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.1000612 | Transcriptional Regulation of Carbohydrate Metabolism in the Human Pathogen Candida albicans | Glycolysis is a metabolic pathway that is central to the assimilation of carbon for either respiration or fermentation and therefore is critical for the growth of all organisms. Consequently, glycolytic transcriptional regulation is important for the metabolic flexibility of pathogens in their attempts to colonize diverse niches. We investigated the transcriptional control of carbohydrate metabolism in the human fungal pathogen Candida albicans and identified two factors, Tye7p and Gal4p, as key regulators of glycolysis. When respiration was inhibited or oxygen was limited, a gal4tye7 C. albicans strain showed a severe growth defect when cultured on glucose, fructose or mannose as carbon sources. The gal4tye7 strain displayed attenuated virulence in both Galleria and mouse models as well, supporting the connection between pathogenicity and metabolism. Chromatin immunoprecipitation coupled with microarray analysis (ChIP-CHIP) and transcription profiling revealed that Tye7p bound the promoter sequences of the glycolytic genes and activated their expression during growth on either fermentable or non-fermentable carbon sources. Gal4p also bound the glycolytic promoter sequences and activated the genes although to a lesser extent than Tye7p. Intriguingly, binding and activation by Gal4p was carbon source-dependent and much stronger during growth on media containing fermentable sugars than on glycerol. Furthermore, Tye7p and Gal4p were responsible for the complete induction of the glycolytic genes under hypoxic growth conditions. Tye7p and Gal4p also regulated unique sets of carbohydrate metabolic genes; Tye7p bound and activated genes involved in trehalose, glycogen, and glycerol metabolism, while Gal4p regulated the pyruvate dehydrogenase complex. This suggests that Tye7p represents the key transcriptional regulator of carbohydrate metabolism in C. albicans and Gal4p provides a carbon source-dependent fine-tuning of gene expression while regulating the metabolic flux between respiration and fermentation pathways.
| Pathogens must be able to assimilate the carbon sources in their environment to generate sufficient energy and metabolites to survive. Since glycolysis is a central metabolic pathway, it is important for this metabolic flexibility. The most commonly isolated agent in human fungal infections, Candida albicans, depends upon glycolysis for the progression of systemic disease. We investigated glycolytic transcriptional regulation in C. albicans and defined two key regulators of the pathway, Tye7p and Gal4p. We demonstrated that these factors are important for the fermentative growth of C. albicans both in vitro and in vivo and also regulate the input and output fluxes of glycolysis. The gal4tye7 strain showed attenuated virulence in a Galleria and two mouse models, potentially due to the severe growth defect in oxygen-limiting environments. Gal4p and Tye7p represent fungal specific regulators involved in the pathogenicity of the organism that may be exploited in the development of antifungal treatments. Our study describes a fungal glycolytic transcriptional circuit that is fundamentally different from that of the model yeast Saccharomyces cerevisiae, providing further evidence that the transcriptional networks in S. cerevisiae need not be generally representative of the fungal kingdom.
| In order to grow and thrive in a wide range of hosts, pathogens not only depend on certain virulence factors but also metabolic flexibility; therefore, they must be able to assimilate various carbon sources. Carbohydrates are the primary and preferred source of metabolic carbon for most organisms, and are used for generating energy and producing biomolecules. Most sugars are converted to glucose 6-phosphate or fructose 6-phosphate before entering the glycolytic pathway. Glycolysis is then responsible for converting these hexose phosphates into the key metabolite pyruvate while producing ATP and NADH (Figure 1). From there, cells carry out two major strategies of energy production: fermentation and respiration. Although both processes regenerate NAD+, respiration is significantly more energetically efficient than fermentation as it produces additional ATP through the tricarboxylic acid (TCA) cycle and oxidative phosphorylation. However, regardless of the mode of energy production, glycolysis is the central, common pathway for both processes. As glycolysis is critical for carbon assimilation, the pathway has been shown to be up-regulated during infections and important to the virulence in pathogenic bacteria, parasites, and fungi [1]–[6].
Since glycolysis is a central metabolic pathway, it is strictly regulated. While there are different levels of regulation of the process, transcriptional control is common to bacteria, fungi, plants, and animals. The glycolytic enzymes are transcriptionally regulated in response to environmental conditions such as oxygen levels, carbon source and availability, and to cellular demands such as metabolite concentrations and energy needs. However, in most species the regulators of glycolytic gene expression have not been identified, so our understanding of transcriptional control of glycolysis in eukaryotes is mainly based on the non-pathogenic yeast Saccharomyces cerevisiae (for review see [7]). In S. cerevisiae, the transcription regulators Gcr1p and Gcr2p are primarily responsible for activating the expression of the glycolytic genes [8],[9]. Gcr1p binds to CT boxes (5′-CTTCC-3′) upstream of the glycolytic genes and Gcr2p acts as a co-activator by forming a complex with Gcr1p [10],[11]. Deleting either gene decreases the expression levels of the glycolytic genes resulting in growth defects during culture on glucose [9],[12]. However, the mutant strains display wild type growth rates on non-fermentable carbon sources [9],[12]. The factor Tye7p (also referred to as Sgc1p) is another glycolysis-specific regulator in S. cerevisiae. Tye7p has been shown to be involved in the activation of several glycolytic genes, although not to the same extent as Gcr1p and Gcr2p [13]. This activation is independent of GCR1 and the tye7 strain displays no growth defects under any carbon source regime [13],[14]. The transcription factors Rap1p, Abf1p, and Reb1p also have roles in activating the glycolytic genes, but these are global factors involved in many cellular processes [15]–[18].
Although the glycolytic circuit is well characterized in S. cerevisiae, most organisms to do not have GCR1 or GCR2 homologs [19]. Furthermore, it is well established that the Saccharomyces-lineage exhibits a unique dependence on the fermentation pathway: these yeasts mainly ferment sugars to ethanol instead of using respiration, even under aerobic conditions [20]. S. cerevisiae up-regulates glycolysis and represses the TCA cycle in the presence of glucose allowing this aerobic fermentation behavior to occur [21]. Only when no fermentable carbon sources are present, after the post-diauxic shift, will S. cerevisiae switch to the respiratory mode. This phenomenon is known as the Crabtree effect and is due to a glucose repression circuit that is largely regulated by the transcriptional repressors Mig1p and Rgt1p, the protein kinase Snf1p, and the protein complex SCFGrr1 [20]. This regulatory circuit is proposed to have developed from the adaptive potential derived from the whole-genome duplication that occurred after the divergence of Saccharomyces from Kluyveromyces [22],[23]; the repression circuit is common to the Saccharomyces-lineage and many of the genes retained from the whole-genome duplication are involved in the lifestyle adaptation to aerobic ethanol production [24]–[26].
The facultative anaerobic lifestyle of Crabtree-positive Saccharomyces yeasts is in contrast to that of most other eukaryotes, which are either facultative or obligate aerobes and lack the glucose repression circuit. Under aerobic conditions, Crabtree-negative cells predominately oxidize pyruvate to carbon dioxide through the TCA cycle. In the absence of oxygen, most aerobic organisms are able to utilize the fermentation pathway to some extent to continue regenerating NAD+. This difference in metabolic flux is highlighted by transcription profiles of the aerobic fungi Trichoderma reesei, Neurospora crassa, and Aspergillus oryzae, which show little or no repression of the TCA cycle in glucose rich compared to glucose poor growth conditions, and therefore do not rely as heavily on the fermentation pathway as does S. cerevisiae [27]–[29].
The opportunistic human fungal pathogen Candida albicans is a facultative aerobe and thus metabolizes carbon sources in response to oxygen availability similar to that of a typical eukaryotic cell. C. albicans is responsible for various non life-threatening infections such as oral thrush and vaginitis but in extreme cases, especially in immunosuppressed individuals, it can cause potentially lethal systemic infections. In fact, Candida species are the most common isolated agent in fungal infections and the fourth leading cause of nosocomial bloodstream infections in the United States, with an attributable mortality rate of approximately 38–49% and treatment costs estimated to be $1.7 billion annually [30]–[33]. C. albicans accounts for more than half of all Candida infections [30],[33], highlighting the importance of understanding the metabolism of this pathogen for the development of effective antifungal treatments.
Crabtree-negative organisms that lack GCR1/2 homologs, such as C. albicans, must control transcription of glycolytic genes differently than does S. cerevisiae. In this study, we characterized two fungal-specific activators of the glycolytic pathway in C. albicans, Tye7p and Gal4p. Deleting both genes resulted in severe growth defects when the mutant cells were cultured on fermentable carbon sources when respiration was inhibited or oxygen was limited, and chromatin immunoprecipitation coupled with microarray analysis (ChIP-CHIP) and transcription profiling showed these factors bind to and regulate expression of the glycolytic pathway genes. Tye7p and Gal4p are also required for complete pathogenicity as the mutant strains showed attenuated virulence. This work therefore defines the key regulatory elements controlling glycolytic gene expression in a facultative aerobic pathogen.
We investigated possible transcriptional regulators of glycolytic gene expression in C. albicans. In the well-studied yeast S. cerevisiae, Gcr1p, Gcr2p, and Tye7p are key glycolysis-specific activators. Unlike Gcr1p and Gcr2p, which are limited to Saccharomyces and closely related yeasts, Tye7-like transcription factors can be found throughout the Saccharomycotina subphylum. Therefore, while there are no homologs of Gcr1p or Gcr2p in C. albicans, there is a CaTye7p. ScTye7p and CaTye7p share 87% amino acid similarity in the DNA binding basic-helix-loop-helix (bHLH) domain but only 33% in the activation domain. A recent investigation also implicated the CaGal4p transcription regulator in the expression of genes involved in glycolysis [34]; while Gal4p in S. cerevisiae is a well-characterized zinc cluster transcription factor that regulates galactose catabolism, it does not fulfill this role in C. albicans. ScGal4p and CaGal4p also share homology strictly in the DNA-binding domain. Due to the potential or observed involvement of these factors in aspects of carbohydrate metabolism, we tested the role of CaTye7p and CaGal4p in the control of glycolytic gene expression in C. albicans.
We first constructed tye7 and gal4tye7 deletion strains to use in conjunction with our previously generated gal4 strain [34]. We tested the ability of all strains to grow on the fermentable carbon sources glucose, fructose, mannose, and galactose, and the non-fermentable carbon source glycerol. On solid media, no growth defect was evident for any deletion strain regardless of the carbon source or concentration tested (Figure 2A). However, when the more sensitive liquid assay was used with glucose, galactose, and glycerol carbon sources, it was able to identify a minor growth defect for the gal4tye7 strain with glucose media, as the doubling time was 147 min compared to that of the wild type of 123 min (Table 1 and Figure S1A). As C. albicans lacks the glucose repression mechanism that exists in S. cerevisiae, its respiration pathway is active under glucose growth conditions so it is not critically dependent on the glycolytic pathway. Although glycolysis is central to both the respiration and fermentation pathways, it is more important for fermentative metabolism since under these conditions the cell must rely exclusively on the ATP generated by glycolysis. As a result, glycolysis proceeds at a higher rate in fermenting cells [20].
To mimic the fermentative metabolism of S. cerevisiae, the mitochondrial inhibitor antimycin A was added to the solid media, and cells in liquid culture were grown without aeration. These changes disrupt the proton gradient and ultimately prevent the production of ATP by oxidative phosphorylation via the respiration pathway. When C. albicans was forced to use fermentation, a severe growth defect during culture on glucose, fructose or mannose media was evident for the double mutant strain (Figures 2B and S1B and Table 1). Therefore, it appears that both Gal4p and Tye7p are involved in fermentative growth with most fermentable carbon sources. The tye7 strain showed a minor growth defect under these fermentative conditions while the gal4 strain still grew at wild type levels, suggesting that Tye7p is a more important regulator of fermentative growth. This prediction was supported by the complemented strains, as reintroducing one copy of TYE7 was sufficient to restore wild type growth rates to the double mutant strain, while one copy of GAL4 resulted in only partial restoration (Figure S1B). Therefore, although both factors appear to be involved in regulating the fermentative growth pathway, Tye7p plays a more central role.
Galactose was unique among the fermentable carbon sources tested as no distinct phenotype was observed for the gal4tye7 strain compared to the wild type under fermentative growth conditions (Figures 2B and S1B). This is likely due to the Kluyver effect, which is thought to be a result of insufficient sugar uptake, and prevents the growth on certain sugars in the absence of respiration [35]. The fermentative growth conditions used (growth without aeration and antimycin A at 2 µg/ml) do not completely inhibit respiration, which allowed the strains to grow, although very slowly, in galactose media. Most yeast hexose transporters are able to take up glucose, fructose, and mannose, while galactose uptake requires separate transporters. If galactose uptake is the limiting step, then any effect of GAL4 or TYE7 on the fermentation pathway will be minimized. Therefore, the lack of observed difference between the mutant strain and the wild type in Figures 2B and S1B is not because the gal4tye7 strain grows well on galactose media when respiration is disrupted, but instead is a result of the comparably poor growth of the wild type strain (Table S1).
To gain insight into why Gal4p and Tye7p are required for fermentative growth, we performed ChIP-CHIP to determine their binding targets. ChIP-CHIP of chromosomally TAP-tagged Gal4p and Tye7p was first performed during growth in glucose media since glucose is the primary carbon source that stimulates the glycolytic pathway. Two different microarray formats, single spot full-genome arrays and whole-genome tiling arrays, were used to provide different strategies for data analysis and for validation purposes. Gal4p bound 98 targets and Tye7p bound 271 targets with the single probe full-genome array (Tables S2 and S3), so Tye7p appears to function as a more global regulator. However, both gene sets were significantly enriched for glucose/carbohydrate metabolic processes and both factors bound essentially all the glycolytic genes. The Gene Ontology (GO) biological processes that were enriched in the bound-gene sets are displayed in Figure 3A. The results are similar between the two factors except, as is discussed below, Gal4p bound more targets involved in pyruvate metabolism. Although Tye7p bound a large number of targets, no GO process is enriched other than carbon metabolism.
The tiling array showed highly similar results but was able to identify a few additional targets including three glycolytic genes, PFK26-2, GLK1, and GLK4. Smoothed peak intensity curves of the tiling array binding events were created to estimate the largest fold enrichments and thus the most significant targets. For Gal4p, the 13 bona fide glycolytic pathway promoters are in the top 68 smoothed peak intensities while for Tye7p they are in the top 52 peaks (Tables S4 and S5). Therefore, the glycolytic promoters are among the most significant targets for both factors. As well, several genes involved in ethanol fermentation (PDC11, ADH1, and NDE1) are included in this group of targets. Therefore, the ChIP-CHIP data suggests that Gal4p and Tye7p are involved in the entire fermentation pathway from glucose to ethanol.
Although Gal4p and Tye7p bound many common targets, there were a significant number of individual binding events, some of which are related to carbohydrate metabolism (Figure 3B). The clearest example is that Gal4p bound the promoter sequences of the five genes encoding the pyruvate dehydrogenase complex (PDH) while Tye7p bound the promoter sequences of the three genes for the trehalose synthase complex. Tye7p also bound several genes involved in glycogen and glycerol metabolism. A summary of selected metabolic binding targets under glucose growth conditions is shown in Figure 4A and Table S6. Many of the Gal4p and Tye7p targets are linked to the glycolytic pathway suggesting that these factors also regulate the input and output fluxes of the pathway.
ChIP-CHIP is a whole-genome approach for determining binding locations of a transcription factor; however, it is insufficient to give a complete picture of a factor's biological function. As was observed with GAL4, the binding and transcription profiles can provide different insights. Therefore, to complement the ChIP-CHIP analysis, transcription profiling comparing wild type and mutant strains was performed under glucose growth conditions. The expression levels of selected carbohydrate metabolic genes is illustrated in Figure 4B and Table S7. As expected, not all targets bound by ChIP-CHIP showed differential expression and not all differentially regulated genes showed direct binding of the transcription factors; however, in general the most significantly bound targets were down-regulated in the absence of the factor. The glycolytic and fermentation genes were down-regulated in the tye7 strain confirming Tye7p's role as an activator of fermentative metabolism. Gal4p is also involved in the activation of the glycolytic/fermentation genes because the expression levels of these genes were lower in the gal4tye7 strain compared to the tye7 strain. The involvement of Gal4p in the activation of glycolytic gene expression was masked in the gal4 expression profiles, most likely because the absence of GAL4 caused an up-regulation of TYE7 [34]. Therefore, Tye7p appears able to significantly compensate for the loss of Gal4p, further supporting the idea that it plays a more central role in glycolytic gene regulation than does Gal4p.
The ChIP-CHIP data showed that only Tye7p bound the genes involved in trehalose metabolism. Expression of the trehalose metabolic genes was down-regulated in the tye7 strain but not further reduced in the gal4tye7 strain indicating that Gal4p is not involved in their activation. Trehalose is a glucose disaccharide that has a role as a storage carbohydrate in yeast. Another important storage molecule in yeast is glycogen. Tye7p also bound several glycogen metabolic targets (GPH1, GDB1, GLG1, GSY1, and GLC3) that were subsequently down-regulated in the tye7 strain and showed no influence of Gal4p in the gal4tye7 strain. As well, Tye7p was the sole regulator of several glycerol metabolic targets (DAK2, GCY1, GPP1, GPD1, and GPD2). On the other hand, Gal4p activated the PDH genes with no influence from Tye7p. These genes were not down-regulated in the tye7 strain (some were slightly up-regulated along with GAL4) but were reduced in the gal4tye7 strain.
The down-regulated genes were analyzed for GO enrichment. As expected, all the categories were related to carbon metabolism. Generally, the two deletion strains had similar results with glycolysis (tye7 P-value: 1.34E-18; gal4tye7: 1.44E-15) and cellular alcohol metabolic process (tye7: 4.48E-17; gal4tye7: 8.33E-16) being the most significantly enriched categories.
Tye7p directly activated many genes involved in trehalose and glycogen metabolism independently of Gal4p. As well, the genes encoding the glycolytic-committing enzyme phosphofructokinase (PFK1 and PFK2) were among the top six most down-regulated genes in the tye7 strain while the gluconeogenesis-specific gene FBP1 was moderately down-regulated. Therefore, it appears that Tye7p regulates the flux between energy storage and energy production at the glucose-6-phosphate branch point (Figure 1). To support this claim, we investigated whether the levels of trehalose and glycogen were different in the tye7 strain compared to the wild type. Since exponentially growing cells have low trehalose levels that rapidly accumulate during stationary phase [36], trehalose amounts were determined from cells at both phases. We observed significantly increased levels of trehalose in the tye7 strain for both stationary and logarithmic phase cells (Figure 5A). Additionally, iodine staining showed that the glycogen content in the tye7 cells was higher than that of the wild type (Figure 5B). The higher storage carbohydrate levels correlate with the expression profile as FBP1 and the majority of genes involved in trehalose and glycogen metabolism were down-regulated 2–3 fold while PFK1 and PFK2 were down-regulated approximately 6 and 9 fold, respectively. Therefore, the glucose-6-phosphate flux in the tye7 strain would favor trehalose and glycogen synthesis. In contrast, the gal4 strain showed wild type levels of both storage carbohydrates, suggesting that Tye7p alone regulates the cell's decision to commit to glycolysis or energy storage.
It is clear that Gal4p and Tye7p are important for fermentative growth when glucose, fructose or mannose is the carbon source. Although there was no phenotype during growth on galactose or glycerol, we investigated the effect of these carbon sources on binding to identify any differences (Dataset S1). Figure 6A illustrates the ChIP-CHIP results of selected carbohydrate metabolic targets during growth on galactose and glycerol media with the behavior during growth on glucose included as a comparison.
A striking trend was the difference in binding between Gal4p and Tye7p under the various carbon sources. Whereas the peak intensity of Gal4p binding changed based on the carbon source, the peak sizes of Tye7p binding were largely unaffected (Figure 7A). This pattern was consistent for the majority of targets resulting in a decrease in the overall number of Gal4p binding targets from glucose to galactose to glycerol and a similar overall number of binding sites for Tye7p with the different carbon sources (Figure S2). Therefore, Gal4p binds its few targets in a carbon source-dependent manner while Tye7p appears to be a more global regulator that binds its targets independently of the carbon source the cells are growing on. This difference in target binding dependent on the carbon source could be a direct result of the protein levels of the transcription factors. We compared protein levels during growth in YPD, YPGal, and YPGly media relative to YP media and observed that Gal4p is significantly induced by glucose while Tye7p has a more constitutive expression (Figure S3). These results further support the inference that Tye7p is the more central regulator of carbohydrate metabolism.
Another trend was that Gal4p showed stronger binding to the promoters of genes encoding the glycolytic pathway enzymes that acted in the later part of the pathway (from TPI1 on) compared to the early part of the pathway, regardless of the carbon source of the growth medium (Figure 6A). In Figure 7A, HXK2 binding is representative of early pathway genes and PGK1 binding is representative of later pathway genes. Interestingly, this difference in binding within the pathway corresponds at the point where the six carbon glucose molecule has been converted to two three carbon products and also represents the separation between the initial ATP consuming steps and the later energy producing steps (Figure 1). Therefore, while Tye7p appears to be involved in committing the cell to glycolysis, Gal4p appears to focus on the later part of the pathway to promote energy production once the commitment is made.
The binding distribution curves created with the tiling array were also used to predict the motif that CaGal4p and CaTye7p recognize by looking for sequences enriched around the binding sites of the top peak intensity targets. ScGal4p has a well established 5′-CGG(N11)CCG-3′ motif [37]. Analysis of the top CaGal4p binding targets revealed enrichment for this motif (Figure 7B). Since ScGal4p and CaGal4p have 86% sequence similarity in the DNA binding domain, it is reasonable to expect they would recognize a similar sequence. The binding distribution curves of HXK2, PGK1, and PDA1 showed Gal4p motifs near the binding sites of CaGal4p (Figure 7A).
As previously mentioned, CaTye7p is a bHLH transcription factor. These type of factors are known to recognize the E-box sequence 5′-CANNTG-3′ [38]. The motif enriched among the top CaTye7p binding targets contains this bHLH signature (Figure 7B). The binding distribution curves of HXK2, PGK1, and TPS3 showed Tye7p motifs near the binding sites of CaTye7p (Figure 7A).
To gain further insight into how Gal4p and Tye7p regulate their targets in response to different carbon sources, transcription profiles comparing wild type and deletion strains with galactose and glycerol as the sole carbon source were performed. Figure 6B illustrates the expression profiles of selected carbohydrate metabolic targets during growth on galactose and glycerol media with the behavior during growth on glucose included as a comparison (complete lists of down-regulated genes in Tables S8, S9, S10, S11, S12 and S13). The glycolytic genes were down-regulated in the tye7 strain under galactose and glycerol growth conditions but not as significantly as with glucose-containing media. Gal4p strongly activated the glycolytic genes on glucose and galactose media but had only minimal effect when glycerol was the carbon source. This result correlates with the location profiling data as Gal4p displayed reduced binding under glycerol growth conditions. Tye7p and Gal4p's carbon-source dependent roles in glycolytic gene expression were validated by quantitative real-time PCR (qPCR) (Figure S4).
Pathogens must not only be adept at utilizing different carbon sources but must also be able to handle changes in oxygen levels. For C. albicans, this flexibility involves growth in oxygen rich environments such as the skin and oral mucosal layers and oxygen poor niches such as inner organs. Since serious systemic infections are associated with these oxygen poor conditions and the glycolytic genes are known to be up-regulated in response to hypoxia in C. albicans [39], we tested the ability of our deletion strains to grow in oxygen-limiting environments. The gal4 strain was unaffected but the tye7 and gal4tye7 strains displayed a severe growth defect (Figure 8A). This defect was observed not only at 30°C but also at 37°C, the human physiological temperature (Figure S5A). This is further validation for Gal4p and Tye7p's role in fermentative metabolism.
To confirm that Tye7p is responsible for the induction of the glycolytic genes under hypoxic conditions we repeated the expression profile under glucose growth conditions in the presence of nitrogen instead of oxygen (for complete lists see Tables S14 and S15). As observed under oxygen rich (normoxic) growth conditions, the glycolytic genes were down-regulated (Figure 8B and Table S7). However, there were some differences compared to the normoxia profile. First, some metabolic genes altered their expression to adjust to the low oxygen environment in the absence of the key glycolytic activator. These changes include the up-regulation of gluconeogenesis-specific genes, glycerol synthesis genes, pentose phosphate pathway genes, and the PDH genes, all of which were either down-regulated or not significantly regulated in the normoxia profile. The lack of a fully functional glycolytic pathway would result in gluconeogenesis stimulation, glycerol synthesis is an alternative to ethanol fermentation to regenerate NAD+, the pentose phosphate pathway is an alternative to glycolysis to generate reducing equivalents, and increasing PDH expression would promote respiration for any of the available pyruvate. Second, GAL4 expression was significantly up-regulated (5.5 fold compared to 1.4 fold in normoxic conditions). This up-regulation also explains the increase in PDH expression and is likely why some glycolytic genes were not as significantly down-regulated. Third, the expression of many more genes was altered in the hypoxia profile, but many of these could be attributed to the significant growth defect of the tye7 strain (doubling time was approximately 220 min compared to the wild type at around 110 min). We chose to just focus on the effect on metabolic gene expression as we had already established Tye7p as a metabolic regulator under normoxic conditions.
Based on the reduced growth rate of the gal4tye7 strain compared to the tye7 strain in low oxygen growth conditions, we assumed that Gal4p was also involved in the induction of the glycolytic genes. The expression profile with the gal4tye7 strain was not done due to the severe growth defect. However, to confirm our hypothesis we transferred normoxia grown cultures to low oxygen conditions for 30 min and analyzed glycolytic gene expression by qPCR. The gal4tye7 strain showed a significant further down-regulation compared to the tye7 strain (Figure 8C).
To ensure our hypoxia experimental set-up was accurate and that the glycolytic genes were induced, we compared the wild type strain under hypoxic and normoxic conditions (Figure 8B and Table S7). Our results agreed well with previously published data [39] as glycolytic, fermentation, stress response, cell wall, fatty acid, iron metabolism, and hyphae-specific genes were up-regulated while TCA cycle, respiration, and ATP-synthesis genes were down-regulated (for complete lists see Tables S16 and S17).
Surprisingly, TYE7 and GAL4 expression appeared down-regulated in the wild type under hypoxia. Recently, it has been shown that the glycolytic genes can be rapidly induced under hypoxia before subsequently declining [40]. This dynamic expression is especially true for transcription factors that must be quickly up-regulated to activate their target genes but may become down-regulated once their target genes have reached their needed expression levels. Therefore, we measured the levels of TYE7, GAL4, and the glycolytic gene CDC19 at four different time points following a shift to hypoxic growth conditions (Figure S6). We observed a rapid increase in TYE7 and GAL4 expression in the first 15 minutes followed by a sharp decline. In contrast, CDC19 induction was longer and the decline less drastic. Thus, TYE7 and GAL4 are initially induced by hypoxia to activate the glycolytic genes before they are subsequently down-regulated.
Since metabolic flexibility and growth under low oxygen conditions are important for pathogens, we investigated the effect of deleting GAL4 and TYE7 on the virulence of C. albicans. We chose to first test all of our strains using the greater wax moth Galleria mellonella as a host model. Screening the virulence of C. albicans strains in Galleria has been shown to produce similar results to those measured through systemic infections with mice [41]. As controls, injections of PBS or UV/heat-killed BWP17 did not kill any Galleria over seven days, demonstrating that any death was attributable to viable C. albicans cells. The gal4 strain showed a minor, but significant (P = 0.008, log-rank test) difference compared to the wild type, while both the tye7 and gal4tye7 strains showed very significant (P<0.0001, log-rank test) attenuated virulence (Figure 9). These results correlate with the observed growth defects under hypoxic conditions. If the mutant strains grow slower than the wild type due to a lower oxygen environment inside the insect, the insect is able to survive for a longer period of time.
The double mutant strain was also tested in two mouse models, A/J and C57BL/6J, to support the result from the Galleria model and allow for further analysis. The A/J strain is C5 deficient and is highly sensitive to systemic infection with C. albicans while the C57BL/6J strain is C5 sufficient and therefore is less sensitive [42]. As with the Galleria model, the gal4tye7 strain displayed significant attenuated virulence in both A/J (P = 0.0009, log-rank test) and C57BL/6J (P = 0.0007) mouse models (Figure 10A).
Fungal loads from different tissues were examined. For the A/J mice, two sets of six mice were injected with the gal4tye7 strain. Fungal loads for the first set were determined 24 hours after injection, when the mice challenged with the wild type and revertant strains were euthanized due to moribundity. Fungal loads from the kidney, liver, and heart were significantly lower (P = 0.002, 0.002, and 0.009, respectively, Mann-Whitney test) in the gal4tye7 infected mice compared to mice challenged with the wild type strain (Figure 10B and Table 2). The fungal load of the second set was determined when the gal4tye7 infected mice became moribund. The fungal burden of this set showed comparable levels to the mice injected with the wild type strain (Figure 10B and Table 2). For the C57BL/6J mice, kidney fungal loads were determined following euthanization due to moribundity or survival until day 21. Five of the six C57BL/6J mice challenged with the gal4tye7 strain survived until day 21 and four of them completely cleared the infection (Figure 10B).
Histological examination showed in vivo hyphae formation of the mutant strain in the kidney of A/J mice (Figure 10C) and confirmed the observations that the gal4tye7 strain is able form hyphae in the presence of serum, N-acetylglucosamine, and spider media in vitro (data not shown). These results further support the hypothesis that the reduced virulence of the double mutant strain is attributed to a growth defect due to low oxygen environments in the host, which allows the organism to survive with the infection for a longer time and increases the chance the host's immune system is able to clear the infection.
Although fungi generally have similarly designed metabolic pathways, their transcriptional regulation of these pathways can be quite different, as illustrated by the human fungal pathogen C. albicans and the non-pathogenic yeast S. cerevisiae. Crabtree-positive yeasts, such as S. cerevisiae, have developed a circuit that represses the respiration pathway and up-regulates the glycolytic/fermentation pathway in the presence of excess glucose even under aerobic conditions. In contrast, C. albicans is Crabtree-negative and prefers to completely oxidize carbohydrates through the respiration pathway in aerobic conditions, only relying on the fermentation pathway in the absence of oxygen. Furthermore, C. albicans up-regulates the glycolytic pathway in low oxygen conditions while S. cerevisiae does not [39]. These metabolic responses of C. albicans are similar to the majority of fungi and other eukaryotes. Therefore, it is important to extend our understanding of transcriptional control of carbohydrate metabolism beyond S. cerevisiae to other organisms.
Although the glycolytic transcriptional circuit has been studied in the yeast Kluyveromyces lactis [19],[43], which lacks the glucose repression circuit, this organism is closely related to S. cerevisiae and its glycolytic genes are mainly regulated by orthologs of Gcr1p and Gcr2p. It appears that K. lactis is an intermediate between S. cerevisiae and the majority of aerobic fungi that do not contain GCR1/2 homologs and therefore it is not a representative model of Crabtree-negative fungi. We characterized Tye7p and Gal4p, two transcriptional activators of the glycolytic pathway in C. albicans, which lacks GCR1/2 homologs. Deleting both factors resulted in a severe growth defect during culture on several fermentable carbon sources (glucose, fructose, and mannose) when respiration was inhibited or oxygen was limited; the single mutant gal4 strain showed no growth defects while the tye7 strain displayed a slight growth defect under these conditions. All deletion strains grew at near wild type levels on a non-fermentable carbon source or with fermentable sources when the respiration pathway was not disrupted.
ChIP-CHIP and transcription profiling revealed that both Tye7p and Gal4p are directly involved in the activation of the entire glycolytic pathway. Tye7p bound all the glycolytic promoters in a carbon source-independent manner and the tye7 strain showed a down-regulation of these genes that was most significant in glucose growth conditions. Gal4p binding to the promoter sequences of the glycolytic genes was affected by the carbon source with a decrease in binding from glucose to galactose to glycerol. Gal4p's role in the activation of these genes was also carbon source-dependent with the strongest effect during growth on glucose and galactose. Furthermore, transcription profiling and qPCR confirmed that both Tye7p and Gal4p are involved in the induction of the glycolytic genes under hypoxic growth conditions.
Gal4p and Tye7p also regulated metabolic processes linked to the glycolytic pathway. Some of these roles are common while some are independent of one another. Tye7p bound to and activated many genes involved in trehalose and glycogen metabolism without Gal4p's involvement. As well, Tye7p strongly activated the genes encoding phosphofructokinase, which catalyzes an irreversible glycolytic-committing reaction. We showed that deleting TYE7 increases the levels of trehalose and glycogen, likely a result of the severely reduced glycolytic flux due to the extremely low expression of PFK1 and PFK2. Deleting GAL4 had no effect on the storage carbohydrate levels. Therefore, Tye7p has a role in determining whether glucose is stored or utilized for energy derivation, similar to Gcr1p in S. cerevisiae [44],[45]. Coordinately regulating this flux ensures that the cell is committed either to storing energy or producing energy as to avoid futile cycling; therefore, it is logical that the key glycolytic activator also regulates trehalose and glycogen metabolism.
Another equilibrium that requires regulation is the flux between fermentation and respiration. This flux depends on the competition for pyruvate between pyruvate decarboxylase (PDC) and the pyruvate dehydrogenase complex (PDH) [46]. In S. cerevisiae, glucose induces PDC expression to promote fermentation while the PDH genes are unaffected and the TCA cycle is repressed [21]. In A. oryzae and N. crassa, both the PDC and PDH are induced by glucose and the TCA cycle is not repressed [28],[29]. Increasing the transcription level of the PDH may be a mechanism employed by Crabtree-negative cells to allow the PDH to out-compete PDC, thereby increasing the respiratory capacity of the cell. Both Gal4p and Tye7p bound to and activated the genes involved in fermenting pyruvate to ethanol indicating that these factors regulate the entire fermentation process from glucose to ethanol. However, only Gal4p bound and activated the PDH genes suggesting that it plays an important role in the metabolic flux of the cell in directing respiration vs. fermentation modes. Consequently, Gal4p has an indirect effect on the TCA cycle. Although Gal4p did not bind the promoter sequences of the TCA cycle genes in the ChIP-CHIP profile (except LSC1/2), deleting GAL4 subsequently results in the down-regulation of the PDH genes and ultimately several TCA cycle genes [34].
Therefore, although both Gal4p and Tye7p are key regulators of the glycolytic pathway, each has its own distinct role. Tye7p is the central transcriptional regulator of carbohydrate metabolism that provides a strong basal level of glycolytic expression while controlling the flux into the pathway and committing the cell to glycolysis. Gal4p is a carbon source-dependent regulator that fine-tunes gene expression based on the cells' need for the fermentation pathway. It assists Tye7p by increasing glycolytic gene expression during growth on fermentable carbon sources. In the presence of fermentable carbon sources Gal4p is able to significantly enhance the energy producing part of the glycolytic pathway and promote respiration by activating the PDH to meet the increased energy needs of the cell and minimize the dependence on fermentation. This assistance is not required with non-fermentable carbon sources as the glycolytic flux is lower since gluconeogenesis is stimulated and the cells grow at a slower rate with reduced energy needs.
The most common types of antifungal drugs for C. albicans infections, azoles, polyenes, and echinocandins, target cell membrane and cell wall integrity; however, targeting fungal metabolism could provide for future drug development. The importance of glycolysis to C. albicans' pathogenicity has been shown as deleting CDC19 causes avirulence while a conditional mutant of FBA1 results in attenuated virulence [3],[4]. Since the components of metabolic pathways are generally highly conserved, orthologs exist in humans reducing interest in these functions as potential drug targets (i.e. C. albicans CDC19 has approximately 50% amino acid identity with human pyruvate kinases). However, Gal4p and Tye7p are fungal-specific regulators and therefore represent potential antifungal targets. GAL4 and TYE7 were shown to be important for the virulence in a Galleria and two mouse models. This virulence effect is likely a result of a growth defect due to the low oxygen conditions that are present in invasive infections. The ability of two-thirds of the gal4tye7 injected C57BL/6J mice to clear the infection highlights the real potential of Gal4p and Tye7p as drug targets. Although the double mutant strain did show significant attenuated virulence in all three host models, it was not avirulent. This is likely attributed to the metabolic flexibility of C. albicans and its ability to use alternative carbon sources. The importance of alternative carbon metabolism to C. albicans' pathogenicity has been previously demonstrated as deleting key enzymes of the glyoxylate cycle, gluconeogenesis, and ß-oxidation pathways reduces virulence [3],[47],[48].
It appears that Gal4p and Tye7p have altered their function throughout the evolution of fungi. We analyzed other genomes in the Saccharomycotina subphylum and observed a pattern relating the presence of GCR1/2 homologs and the clustering of the Gal4p motif. Species that possess GCR1/2 homologs have an enrichment of the Gal4p motif upstream of the GAL regulon genes while species lacking GCR1/2 homologs have an enrichment of the Gal4p motif in the promoter regions of the glycolytic genes. Therefore, it appears that the rewiring of Gal4p coincides with the loss/gain of GCR1/2, and that Gal4p and Tye7p likely regulate glycolysis in the Saccharomycotina species lacking GCR1/2 homologs. Intriguingly, this rewiring event also appears to coincide with changes in the galactose sensory network [49].
Along with the ribosomal transcriptional network [50],[51], the transcriptional regulatory control of glycolysis represents an example of the plasticity of circuits throughout the evolution of fungi. The Rap1p ribosomal circuit and the Gcr1p/Gcr2p glycolytic circuit are unique to S. cerevisiae and closely related species, suggesting that the transcriptional networks of S. cerevisiae are not representative models of the fungal kingdom. However, there is a key distinction between these two rewiring events. While the regulatory components of the ribosomal network are different, the ultimate outcome is identical. On the other hand, carbon metabolism regulation is fundamentally different between Saccharomyces and other eukaryotes suggesting that the entire carbohydrate transcriptional network has undergone rewiring, not just the regulators. Thus, investigating the regulation of carbon metabolism in Crabtree-negative organisms is important as little can be extrapolated based on S. cerevisiae. This study thus defines the key regulatory elements of glycolytic gene expression and provides insights into the mode of transcriptional regulation of carbohydrate metabolism in a typical eukaryotic cell and a human pathogen.
The C. albicans strains used in this study are listed in Table S18. Cells were generally grown at 30°C in media containing 1% yeast extract, 2% peptone, with either 2% dextrose (YPD), 2% galactose (YPGal), or 2% glycerol (YPGly). All media was supplemented with uridine (50 µg/ml).
Plasmids and oligonucleotides used in this study are listed in Tables S19 and S20, respectively. Gal4p and Tye7p were tagged chromosomally with a TAP-URA3 PCR product [52]. Transformations were carried out using standard procedures [53]. Correct integration of the TAP-tag was confirmed by PCR, and western blots were used to verify protein expression.
BWP17 was used for generating the tye7 strain. CMM3 (gal4) [34] was used to generate the gal4tye7 strain. The deletion and complementation strains were created using the SAT1-flipper cassette as described [54] with some modifications. The tye7 disruption construct was created by cloning 500 bp flanking sequences of TYE7 into the plasmid pSFS2A. The plasmid was linearized prior to transformation. Transformants were selected on YPD plates containing 200 µg/ml of nourseothricin and confirmed by PCR. Excision was performed by incubation at 30°C for 5 hours in YP media with 2% maltose before plating on YPD plates. Confirmation of excision events was done by PCR. The process was repeated for disruption of the second allele. Revertants of tye7 and gal4 were created using the gal4tye7 strain. Complementation of tye7 was carried out by reintroducing the ORF at its native locus by replacing the upstream flanking sequence in the tye7 disruption cassette with the complete TYE7 ORF. A complementation cassette that consisted of the complete GAL4 ORF and a 500 bp downstream flanking sequence was constructed for reintroduction of GAL4 at its native locus. We selected clones that replaced the HIS1 deletion cassette and then restored histidine prototrophy by transformation with NruI-digested pGEM-HIS1 [55].
CMM3 and CMM1 (wild type prototrophic equivalent of CMM3) [34] were also used in the phenotypic assays. In all assays BWP17 and CMM1 were used as control strains but they usually gave identical results so only one was generally shown. The exception was for the liquid growth curves as BWP17 plateaued at a lower OD600 than CMM1. In this case, the tye7 strain was compared to BWP17 and normalized to CMM1 for graphical purposes.
As uridine auxotrophy affects virulence but can be restored by integration of URA3 at the RPS10 locus [56], for the mouse studies the CMM1, gal4/gal4/tye7/tye7, and gal4/GAL4/tye7/TYE7 strains were made prototrophic by targeting the URA3 marker to the RPS10 locus through StuI digestion of CIp10. Correct integration was confirmed by PCR and qPCR verified that only one copy of URA3 was integrated.
Serial spotting plate assays were carried out as described [57] except cells were washed twice with sterile water before plating. Cells were plated on synthetic complete media containing the carbon source at 0.2% or 2% and agarose at 2% to minimize carbon source impurities. Antimycin A (2 µg/ml) was added to inhibit respiration. For liquid assays, cells were grown to log phase in synthetic medium with 5% glycerol, washed twice with sterile water, and resuspended at an OD600 = 0.1 in synthetic media containing the carbon source at 2%. Cells were either grown at 30°C in flasks with shaking (aerobic conditions) or in microtiter plates without shaking (static conditions). Samples were done in triplicate and the average was used for analysis. For growth under hypoxic conditions, cells were spotted on YPD plates and incubated in an anaerobic chamber. The chamber was flushed daily with nitrogen to remove oxygen and any by-products.
ChIP experiments were performed as previously described [52] with some modifications. Briefly, cells were grown to OD600 = 2 in 40 ml of YPD, YPGal, or YPGly. Tagged ChIPs were labeled with Cy5 dye and untagged (mock) ChIPs were labeled with Cy3 dye. Microarray hybridization and washing were performed as described [58]. Scanning was done with a ScanArray Lite microarray scanner (Perkin Elmer). QuantArray was used to quantify fluorescence intensities. Data handling and analysis were carried out using Genespring v.7.3 (Agilent Technologies). The significance cut-off was determined using the distribution of log-ratios for each factor. A minimum of three biological replicates were analyzed for each carbon source condition with hybridization to single spot full-genome (ORF and intergenic) microarrays containing 11,817 70-mer oligonucleotide probes [52]. For determining the number of targets and GO analysis with the single probe microarrays, the cut-off was a fold enrichment >1.5 and a t-test P-value<0.1.
One replicate of the ChIP-CHIP experiments for each carbon source condition was hybridized to a custom designed whole-genome tiling array for further analysis. Using the C. albicans Genome Assembly 21 [59] and the MTL alpha locus [60], we extracted a continuous series of 242,860 60 bp oligonucleotides each overlapping by 1 bp. We eliminated 2062 probes containing stretches of at least 13 A/T nucleotides. The remaining 240,798 probes were used to produce a whole-genome tiling array using the Agilent Technologies eArray service (https://earray.chem.agilent.com/erray/).
Lowess normalization of the intensity ratio of each of the 240,798 probes was done using an in-house software implementation. The signal along each chromosome was smoothed using a median filter (n = 3) followed by a Gaussian low-pass filter (σ = 150 bp). For peak localization, the smoothed signal was interpolated at 10 bp intervals using cubic-spline resampling. Peaks were reported in decreasing order of the smoothed intensities.
RNA was extracted with the RNeasy Kit (Qiagen) as per manufacturer's instructions. Briefly, cells were grown to OD600 = 0.8 in YPD, YPGal, or YPGly in aerated flasks (normoxia) and disrupted using acid washed glass beads. Transcriptional profiling was carried out as described [58] with 20 µg of RNA used for cDNA synthesis. A minimum of three biological replicates on double spotted ORF microarrays (6,394 intragenic 70-mer oligonucleotide probes) were used for analysis [58]. Scanning and analysis were carried out as described for ChIP-CHIP except scanning was performed at two different laser PMTs to avoid saturated signals for a few highly expressed transcripts including several glycolytic genes. For GO analysis and the supplementary complete lists the cut-off was a difference in expression >1.5 or <0.67 and a t-test P-value<0.05.
Expression profiles under hypoxic conditions were performed as described above except bottles containing YPD media were flushed with nitrogen to remove oxygen. Two biological replicates were performed and statistical analysis was done as with normoxic conditions.
For qPCR analysis of glycolytic gene expression under hypoxic conditions, BWP17, tye7, and gal4tye7 cultures were grown to OD600 = 0.7 in YPD in an aerated flask. The cultures were then transferred to bottles flushed with nitrogen and grown for an additional 30 min before the RNA was extracted as above and analyzed by qPCR.
For hypoxic induction kinetic analysis, BWP17 was grown to OD600 = 0.8 in YPD in an aerated flask. Half the culture was transferred to bottles flushed with nitrogen while the other half was left to grow in the aerated flask. At different time points the RNA was extracted and compared by qPCR.
For qPCR, cDNA was synthesized from 5 µg of total RNA using the reverse-transcription system (50 mM Tris-HCl, 75 mM KCl, 5 mM DTT, 3 mM MgCl2, 400 nM oligo(dT)15, 1 µM random octamers, 0.5 mM dNTPs, 200 units Superscript III reverse transcriptase; Invitrogen). The mixture was incubated for 60 min at 50°C. Aliquots were used for qPCR, which was performed using the Corbett Rotor-Gene RG3000A (Corbett Research) with SYBR Green fluorescence (Qiagen). Cycling was 10 min at 95°C followed by 40 cycles (95°C, 10 s; 58°C, 15 s; 72°C, 20 s). Samples were done in triplicate and means were used for calculations. Fold changes were estimated using the comparative ΔΔCt method as described [61] with the coding sequence of the C. albicans ACT1 ORF as a reference.
Trehalose levels were measured as based on previous studies [62]. Briefly, cells were either grown to log phase (OD600 = 2) or grown for 40 hours to reach stationary phase, washed twice with cold water, and resuspended in water. Cells were lysed by incubating at 95°C for 30 min and the supernatant was used for enzymatic analysis. Reactions (50 µl of sample, 100 µl 270 mM citric acid buffer pH 5.7, and 0.15 U trehalase (Sigma)) were incubated at 37°C for 5 hours. Glucose amounts were assayed with the hexokinase glucose kit (Sigma) with endogenous glucose levels determined based on reactions without trehalase. A BCA protein assay (Pierce) was performed as per manufacturer's instructions. Relative trehalose levels were based on nmol trehalose per mg of cell protein. Three biological replicates were performed for each strain and condition. Glycogen levels were estimated using the iodine vapor method [63]. Cells were serially diluted, spotted on YPD plates, and incubated for 24 hours at 30°C. The cells were then exposed to iodine vapor for 5 minutes.
To determine the protein expression of Gal4p and Tye7p under different carbon sources, overnight cultures in YPD were diluted to OD600 = 0.4 and grown for 2 hours in fresh YPD. Cells were washed twice with sterile water and resuspended in YP, YPD, YPGal, or YPGly media and grown for an additional 3 hours. The cells were then washed with TBS buffer and lysed using acid washed glass beads (same lysis buffer as ChIP-CHIP except for addition of a phosphatase inhibitor cocktail (Roche)). The protein extract was clarified by centrifugation and a BCA protein assay (Pierce) was performed as per manufacturer's instructions. Gal4p and Tye7p were separated on a 10% SDS gel and transferred to a PVDF membrane. A rabbit polyclonal antibody directed against the TAP-tag (Open Biosystems) was used (1∶2000 dilution). A mouse anti-actin monoclonal antibody (1∶500 dilution, Chemicon) was used to probe actin as a loading control. HRP-conjugated anti-rabbit and anti-mouse secondary antibodies (Santa Cruz) were used (1∶10,000 dilution). The HRP signal was revealed using the Lumi-Light Western Blotting Substrate (Roche).
For Galleria mellonella studies, overnight cultures were washed twice with PBS and resuspended in PBS at OD600 = 8. Larvae, in the final instar, weighing 180±10 mg were injected between the third pair prothoracic legs with 10 µl of suspension (8×105 cells). Infected larvae were incubated at 37°C in the dark with excess of a multigrain diet supplemented with glycerol and vitamins [64]. Four replicates, each consisting of 20 insects, were carried out with survival rates measured daily for a period of 7 days. Death was determined based on the lack of response to touch and the inability to right themselves. A BWP17 culture was irradiated with UV for 2 hours and incubated at 95°C for 1 hour before injection to confirm that viable C. albicans cells were responsible for death. Kaplan-Meier survival curves were created and compared with the log-rank test (GraphPad Prism 5).
Mouse studies were carried out as previously described [65]. Briefly, 8–12 week-old A/J and C57BL/6J mice (Jackson Laboratories, Bar Harbor, ME) were inoculated via the tail vein with 200 µl of a suspension containing 3×105 C. albicans in PBS. Six mice, three female and three male, were used for each experimental group except for the gal4tye7 injected A/J set where six females and six males were used. Mice were closely monitored and those showing extreme lethargy were considered moribund and were euthanized. Target organs were removed aseptically and homogenized in PBS before plating on YPD plates containing chloramphenicol (34 µg/ml). The number of yeast colonies per organ was determined, log-transformed, and compared using the Mann-Whitney test (GraphPad Prism 5). Comparative genomic hybridization (CGH) analysis was performed on all strains prior to injection to verify that no aneuploidy arose as a result of any of the genetic manipulations. All experimental procedures involving mice were approved by the Biotechnology Research Institute Animal Care Committee, which operated under the guidelines of the Canadian Council of Animal Care.
|
10.1371/journal.pgen.1005005 | Autoselection of Cytoplasmic Yeast Virus Like Elements Encoding Toxin/Antitoxin Systems Involves a Nuclear Barrier for Immunity Gene Expression | Cytoplasmic virus like elements (VLEs) from Kluyveromyces lactis (Kl), Pichia acaciae (Pa) and Debaryomyces robertsiae (Dr) are extremely A/T-rich (>75%) and encode toxic anticodon nucleases (ACNases) along with specific immunity proteins. Here we show that nuclear, not cytoplasmic expression of either immunity gene (PaORF4, KlORF3 or DrORF5) results in transcript fragmentation and is insufficient to establish immunity to the cognate ACNase. Since rapid amplification of 3' ends (RACE) as well as linker ligation of immunity transcripts expressed in the nucleus revealed polyadenylation to occur along with fragmentation, ORF-internal poly(A) site cleavage due to the high A/T content is likely to prevent functional expression of the immunity genes. Consistently, lowering the A/T content of PaORF4 to 55% and KlORF3 to 46% by gene synthesis entirely prevented transcript cleavage and permitted functional nuclear expression leading to full immunity against the respective ACNase toxin. Consistent with a specific adaptation of the immunity proteins to the cognate ACNases, cross-immunity to non-cognate ACNases is neither conferred by PaOrf4 nor KlOrf3. Thus, the high A/T content of cytoplasmic VLEs minimizes the potential of functional nuclear recruitment of VLE encoded genes, in particular those involved in autoselection of the VLEs via a toxin/antitoxin principle.
| The rather wide-spread and extremely A/T rich yeast virus like elements (VLEs, also termed linear plasmids) which encode toxic anticodon nucleases (ACNases) ensure autoselection in the cytoplasm by preventing functional nuclear capture of the cognate immunity genes, but how? When expressed in the nucleus, the mRNA of the VLE immunity genes is split into fragments to which poly(A) tails are added. Consistently, lowering the A/T content by gene synthesis prevented transcript cleavage and permitted functional nuclear expression providing full immunity against the respective ACNase toxin. Thus, internal poly(A) cleavage is likely to prevent functional nuclear immunity gene expression.
| Pichia acaciae and Kluyveromyces lactis each contain two cytoplasmic virus-like elements (VLEs, also known as linear plasmids); i.e. pPac1-1 (12.6 kb), pPac1-2 (6.8 kb) and pGKL2 (13.5 kb), pGKL1 (8.9 kb) respectively [1,2]. The respective larger elements display substantial similarities to each other in terms of organization and gene content. They can exist without the smaller ones as they encode all proteins required for nucleus-independent cytoplasmic replication and maintenance [3]. The smaller VLEs pPac1-2 and pGKL1, respectively, which depend on the larger ones in terms of cytoplasmic transcription and/or replication, encode for the production of killer toxin complexes, zymocin (pGKL1) and PaT (pPac1-2) [reviewed in 4]. One subunit in either zymocin or PaT is highly conserved; it carries chitin binding and chitinase domains that recognize cell wall associated chitin of target cells as primary toxin receptor for subsequent import and/or activation [5,6,7]. In both zymocin and PaT, a rather hydrophobic stretch or subunit appears to manage membrane transfer of the cytotoxic subunits, PaOrf2 (encoded by pPac1-2 ORF2) and γ-toxin (encoded by pGKL1 ORF4). Although they hardly show any sequence similarity, they both act as anticodon nucleases (ACNases). The recently solved crystal structure of PaOrf2 revealed a unique fold, which shows no similarity to any known ribonuclease [8]. PaOrf2 specifically attacks tRNAGln in vivo and additionally cleaves in vitro tRNAGlu and tRNALys or synthetic stem-loop RNA derived from the tRNAGln sequence [8,9]. γ-toxin cleaves the same tRNAs in vitro, but in vivo its preferred target is tRNAGlu [10,11]. While γ-toxin cleaves its target tRNA once at the 3`side of the wobble uridine, PaOrf2 apparently cleaves at the same position and additionally two nucleotides upstream, as judged from the appearance of two alternative cleavage products with full length tRNA from S. cerevisiae [9,10]. Since PaOrf2 but not γ-toxin evades a possible repair of the tRNA halves by cellular tRNA ligases, it was speculated that the presence of two cleavage sites might allow the excision of a di-nucleotide, rendering the target tRNA non-repairable [12,13,14,15].
VLE cured strains of P. acaciae and K. lactis are sensitive to their own respective toxins, proving that not only the killer phenotype but also the cognate immunity are encoded by the elements [1,2]. Indeed, PaT immunity is conferred by the only protein encoded by pPac1-2 (ORF4) that lacks a signal peptide for secretion [16] and immunity against zymocin had been postulated to be encoded by KlORF3 of pGKL1 [17]. There is hardly any homology among PaORF4 and KlORF3, and consistent with it, no cross-protection has been observed against zymocin or PaT [16]. PaT and zymocin are the most thoroughly studied VLE encoded ACNase killer toxins, but there are other systems in yeast [reviewed in 18], such as PiT from Pichia inositovora, a ribonuclease inducing specific fragmentation of 25S and 18S rRNAs [19], or DrT from Debaryomyces robertsiae, an ACNase resembling PaT and cleaving tRNAGln [20]. From an evolutionary point of view, toxin and immunity functions implemented in VLEs have to be considered as players of an autoselection system rather than providing advantages for the respective host [16,21], although the latter, clearly benefits from the conferred killer phenotype.
Intriguingly, PaORF4 encoding PaT immunity could be heterologously functionally expressed solely from VLEs in the cytoplasm [16], i.e. when the gene was integrated into the pGKL-system of Kluyveromyces lactis. All efforts to express the immunity phenotype with PaORF4 on nuclear episomal and centromeric vectors failed to establish self-protection against the ACNase toxin. Here, we show that PaORF4 as well as the immunity genes from other VLEs (KlORF3 and DrORF5) are nevertheless transcribed when the genes are governed by a yeast nuclear promoter in episomal vector backbones, but the mRNA becomes immediately fragmented thereby preventing translation that otherwise would yield a functional immunity protein. As exemplified for PaORF4 and KlORF3, changing the primary structure from a rather high A/T bias to a much lower degree allowed for functional nuclear immunity expression, proving that the gene’s primary sequence information is sufficient to provide ACNase self-protection and that the native ORF context ensures autoselection of the VLE.
For the three known VLE encoded ACNase toxin complexes PaT, zymocin and DrT, immunity functions were proposed to be encoded by PaORF4, KlORF3 and DrORF5, respectively [16,17,20]. Subsequently, PaORF4 and DrORF5 were functionally expressed from their native promoters in the cytoplasm after integration of the genes into a VLE system (the pGKL1/2 system transferred to S. cerevisiae) [16,20,22]. Attempts to express both immunity genes in the nucleus after fusion of the ORFs to the constitutive ADH1 promoter (ADH1pr), however, did not establish toxin immunity. This is in contrast to the putative zymocin immunity gene (KlORF3) which was previously identified on the basis of functional expression from the nucleus after fusion of the KlORF3 to the PGK promoter [17]. Upon expression of the PGKpr-KlORF3 construct, partial zymocin protection was observed in S. cerevisiae cells. However, the phenotype required the presence of the autonomous VLE pGKL2 while, nuclear expression in a standard S. cerevisiae strain devoid of any VLE did not confer detectable zymocin immunity [17]. To reconfirm this latter notion, we fused KlORF3 to the alternative strong constitutive promoter ADH1pr and analyzed the zymocin response of the sensitive S. cerevisiae strain BY4741 containing the ADH1pr-KlORF3 fusion in comparison to the wild type by the microdilution method. As shown in Fig 1, zymocin sensitivity indeed remains unaltered in the presence of the ADH1pr-KlORF3 construct, supporting the conclusion that the VLE encoded immunity factors cannot be functionally expressed in the nucleus in a standard S. cerevisiae strain. In contrast, both DrORF5 and PaORF4 provide resistance to their cognate ACNase toxins (DrT and PaT) when expressed in the cytoplasm of a sensitive S. cerevisiae strain and in all assays conducted, complete, rather than partial immunity was observed [16,20].
To analyze whether the observed failure of immunity expression from the nuclear vectors was due to a barrier in transcription or due to transcript instability, we analyzed the levels of mRNAs encoding immunity (immRNA) and their stability. ImmRNA from S. cerevisiae strains carrying nuclear fusions of the immunity factor encoding ORF (immORF) and the ADH1pr was compared with immRNA from natural expression hosts, where immORFs are expressed from the cytoplasm. In all cases, cytoplasmic expression yielded stable immRNA that exceeded the size of the corresponding full-length immORF (Fig 2). In contrast, nuclear expression of immORFs produced one (KlORF3), two (DrORF5) or four (PaORF4) distinct signal bands in Northern blots, which were significantly smaller in size than their corresponding full-length immORF (Fig 2). Thus, while immRNAs are stable, when expressed from their cognate VLEs in the cytoplasm, they are prone to fragmentation and get particularly instable when expressed in the nucleus. The lack of full length immRNA in the latter case is in line with the observed general lack of functional ACNase immunity when immORFs are expressed in the nucleus.
Since the poly(A) site processing machinery recognizes UA rich elements that appear to be quite diverse in S. cerevisiae [reviewed in 23], we explored the possibility that immRNA fragmentation of the highly A/T biased transcripts could be associated with recognition of random, internal poly(A) sites, leading to immORF fragmentation with the addition of poly(A) tails at the cleavage sites. We isolated total RNA from S. cerevisiae strains expressing ADH1pr-KlORF3, ADH1pr-PaORF4 and ADH1pr-DrORF5 and primed cDNA synthesis with a poly(A)-specific oligonucleotide. Such cDNAs were analyzed by PCR using immORF specific oligonucleotides, binding at the 5’ end together with an oligonucleotide complementary to the poly(A)-specific anchor. As a control, the ERG3 mRNA was amplified from the same cDNA preparations. For all immORFs, several 3’ RACE products were obtained, all of which were smaller than the minimum expected gene size (Fig 3), suggesting the presence of poly(A) stretches at the 3´ ends of the immRNA fragments. Such result agrees with the specific fragmentation of the nuclearly expressed immRNAs followed by the addition of poly(A) tails. Fragments of each PCR reaction were extracted from the gel and cloned; sequencing identified the fragments to perfectly match the 5`terminal regions of the respective immRNAs, which are truncated at their 3´ends and extended by attachment of several (16 to 66) adenines (Fig 4).
To ensure that the data obtained from 3’RACE experiments did not result from unspecific internal priming within A-rich mRNA regions, a linker ligation method was applied to exemplarily identify the immRNA ends of KlORF3. The linked KlORF3 fragments were amplified using a gene specific and a linker specific primer and after cloning analyzed by sequencing. All identified fragments contained a poly(A) tail consisting of 7–49 adenyl nucleotides. The results confirm the previous observations. Both methods identified that cleavage and polyadenylation of each of the immRNAs happens not only at one definite, but at multiple positions. For example, among the 33 sequenced PaORF4 mRNA fragments, 15 different polyadenylation sites in a region between positions 209 nt and 594 nt were mapped (Fig 4 and S1 Fig). The identified positions in the mRNA truncation products and their frequency of occurrence for each immRNA are summarized in Fig 4.
Since the above results suggested the possibility that the high A/T content of immRNAs limits their nuclear expression due to ORF-internal poly(A) site processing, we analysed whether a reduction of the A/T content of PaORF4 and KlORF3 improves their expression from the nucleus. Synthetic variants of both genes were generated, where most of the A/U rich codons were replaced by synonymous more G/C rich codons via gene synthesis (S2 Fig). As a result, the G/C content for PaORF4 increased from 21% to 45% in the synthetic variant PaORF4ms and KlORF3 increased in G/C content from 22% to 54% in the synthetic variant KlORF3ms without altering the amino acid sequence. Both variants were cloned into the same vector backbone previously used to study the native, unchanged genes, resulting in a set of multicopy plasmids carrying immORF fusions to the ADH1 promoter, where immORFs remain either unchanged in codon usage (78–79% A/T) or exhibit a significantly reduced A/T content (46–55%). All constructs were expressed in a PaT or zymocin sensitive VLE-free S. cerevisiae strain and the presence of full length immRNA was comparatively analysed by RT-PCR (Fig 5). As a control, the ERG3 mRNA was detected in all strains in parallel. Full length immRNA was generally absent in the strains expressing the A/T-rich native (non-modified) versions of the immORFs (Fig 5A), which is in agreement with the results obtained by Northern analysis (Fig 2). In striking contrast, however, full length immRNA becomes detectable when low A/T content codon usage variants PaORF4ms and KlORF3ms are expressed from the same nuclear constructs (Fig 5B). Thus, lowering the A/T content clearly improves immORF expression in the nucleus compared to the natural gene variants being expressible only in the cytoplasm.
To check whether such improvement also enables functional immORF expression, ACNase immunity of strains carrying the different immORF expression constructs was scored by the eclipse plate and microdilution assays (Fig 6). Consistent with previous results, expression of A/T-rich versions of PaORF4 or KlORF3 did not confer a detectable immunity phenotype to the PaT or zymocin producers, respectively. When the low A/T-content variants PaORF4ms or KlORF3ms were expressed, however, sensitivity to the cognate ACNase toxin producer was entirely lost. At the same time, the PaORF4ms expressing strain remained sensitive to the zymocin producer (Fig 6A and Fig 6B) and the KlORF3ms expressing strain remained sensitive to the PaT producer (Fig 6A). These results indicate that although lowering the A/T contents in two functionally distinct immORFs suffices to overcome the observed nuclear expression barrier, the immunity factors, once being expressed, do not confer ACNase cross-protection to non-self yeast strains.
Our KlORF3ms expression studies (Fig 6A) confirm the previous proposal by Tokunaga et al., [17] that pGKL1 encoded Orf3 protein confers zymocin immunity. As KlORF3ms yields protection to the zymocin producer K. lactis in the background of S. cerevisiae strain BY4741, there is evidently no general requirement for the presence of the pGKL2 VLE to establish immunity. In previous experiments, the immunity phenotype associated with the PGKpr-KlORF3 construct and the pGKL2 VLE was only partial, since exogenous purified zymocin induced detectable growth inhibition [17]. To check whether the pGKL2 independent immunity conferred by nuclear expression of ADH1pr-KlORF3ms also is partial, we analyzed the zymocin response using the microdilution assay. We compared KlORF3ms-induced zymocin protection in WT cells with elp3 cells not carrying any immORF. The latter condition prevents zymocin induced tRNA cleavage due to absence of the crucial wobble uridine mcm5s2-modification and confers full toxin resistance [10,11]. We observed no difference between the zymocin response of elp3 cells not expressing any immORF and ELP3 cells expressing KlORF3ms; only ELP3 wild type cells without KlORF3ms showed sensitivity to zymocin (Fig 7A). To further check the dominant nature of KlORF3ms induced immunity and to analyze whether the KlOrf3 protein is capable of intracellular inactivation of γ-toxin, as suggested in earlier work [17], we constructed strains co-expressing KlORF3/KlORF3ms and GAL1pr-driven, multi copy KlORF4 devoid of its signal peptide encoding region, leading to intracellular accumulation of the ACNase subunit γ-toxin (KlOrf4). As a control, both immORF constructs were introduced into the elp3 strain, where the need for an immORF is overcome by preventing tRNA cleavage in the first place. Galactose induced expression of the γ-subunit proved inhibitory to the strain expressing the A/T-rich variant KlORF3 only; as expected, the elp3 mutation prevented toxic effects of γ-toxin but also KlORF3ms entirely prevented growth inhibition by intracellular γ-toxin and the additional removal of ELP3 did not improve growth of the strain under inducing conditions (Fig 7B). Thus, zymocin immunity acts, similar to the previously studied PaT and DrT immunity functions at the intracellular stage and provides true immunity rather than partial resistance, independent of any pGKL2 encoded functions. Since we detected no cross resistance of KlORF3ms expressing strains to the non-cognate ACNase toxin PaT (Fig 6A and Fig 6B) and PaORF4ms did not protect detectably from zymocin (Fig 6A), the two immunity proteins PaOrf4 and KlOrf3 are highly specific for each of their cognate ACNase subunits.
PaT, DrT and zymocin are the three known examples of eukaryotic protein toxins with ACNase activity. All three are encoded by non-autonomous VLEs (pPac1-2; pWR1A and pGKL1) persisting in the cytoplasm of different yeast species. Crucial functions for cytoplasmic transcription and DNA replication, processes normally occurring in the nucleus, are supplied in each case by a larger VLE (pPac1-1; pWR1B and pGKL2). Among these are a uniquely structured RNA polymerase [24] as well as a virus like mRNA capping enzyme [25,26] to generate capped, cytoplasmic mRNAs from unique cytoplasmic promoters [27,28,29]. Generally, VLE genes cannot be expressed in the nucleus due to non-recognition of the cytoplasmic promoters by the nuclear (host encoded) RNA polymerases. The presence of genes on pPac1-2 and pWR1A mediating immunity against PaT and DrT was previously shown by integration into the pGKL1/2 system transferred to S. cerevisiae. The parental pGKL1/2 carrying S. cerevisiae strain produces zymocin and its cognate immunity factor but was sensitive to DrT and PaT. This sensitivity was entirely lost upon integration of DrORF5 and PaORF4, respectively [16,20]. Since both, DrOrf5/PaOrf4 and DrOrf3/PaOrf2 display detectable sequence homology and there is significant DrT/PaT cross protection mediated by PaOrf4, a direct recognition of the matching (PaOrf2) or nearly matching (DrOrf3) ACNase by the immunity factor was suggested. In support, PaOrf4 can disable toxic in vivo effects of intracellular PaOrf2 and both proteins were shown to form a complex in vitro that inhibits the ACNase activity of PaOrf2, resembling the mode of action of tRNase colicin immunity factors, which tightly bind and occlude the tRNase active site [8,30]. Similarly, in vivo studies with DrOrf5 showed that it protects against the in vivo tRNase activity of the intracellular DrOrf3 subunit [20], hinting at a similar immunity principle as for PaOrf4. Co-crystal structures of VLE-encoded tRNases with their cognate immunity proteins will be required to determine whether immunity factors against toxic tRNases as evolutionary diverse as prokarytotic colicins and VLE encoded killer toxins indeed share a similar mechanistic strategy to bind and occlude the tRNase active site.
For zymocin, understanding of the immunity factor function had been less advanced; published data [17] indicated an as yet undefined requirement for the VLE pGKL2 to establish KlOrf3 mediated immunity and in contrast to PaT and DrT immunity functions, the zymocin immunity factor appeared to provide partial protection only. Additionally, it was suggested that the zymocin immunity factor protects from intracellular γ-toxin based on the observation that pGKL1/2 carrying cells are resistant to galactose-induced expression of a signalpeptide-less KlORF4 gene [17], but no similar protection has been shown for the isolated KlORF3 gene. Since heterologous expression of KlORF3 precluded the use of the cytoplasmic pGKL1/2 based expression system, we encountered the general problem that even the established immunity genes PaORF4 and DrORF5 could not be expressed in the nucleus after replacement of the cytoplasmic promoter by well characterized nuclear promoters. As the same outcome was observed for the zymocin immunity gene, a general principle inhibitory to nuclear expression of these immunity genes became obvious. Since AT rich immunity genes are efficiently translated by the host’s translational machinery when the corresponding mRNA is generated in the cytoplasm by a VLE encoded transcriptional machinery but not when the same mRNA is generated in the nucleus, a nuclear transcriptional rather than a cytoplasmic translational barrier appeared to exist. Northern, RACE and linker ligation analysis now shows that nuclear expression generally ends up in fragmentation of the immRNAs which goes along with the addition of poly(A) tails.
Poly(A) site recognition is thought to basically involve the presence of the AAUAAA poly(A) signal (PAS), together with a GU-rich sequence as a downstream element [reviewed in 23]. However, unlike higher eukaryotes, yeast apparently tolerates a high degree of variation in individual poly(A) site recognition elements, as was derived from the analysis of expressed sequence tags generated by oligo(dT) primed cDNA synthesis [31]. For example, the PAS element in yeast is simply characterized by being A-rich. Thus, extremely AU-rich transcripts, such as VLE derived genes exhibit a high probability of ORF internal poly(A) site recognition and processing when moved to the nucleus. In support of this, we show that lowering the A/T content by defined gene synthesis is sufficient to prevent immRNA fragmentation in the nucleus allowing for functional expression of PaT and zymocin immunity phenotypes.
Our analysis with the synthetic KlORF3ms construct shows that the zymocin immunity protein indeed acts intracellularly and provides true self-protection rather than partial resistance. Thus, all eukaryotic ACNase immunity proteins may, like PaOrf4, recognize and inhibit their cognate ACNase. In support of a specific recognition, cross immunity against DrT but not zymocin can be provided by PaOrf4, whereas KlOrf3 provides immunity solely and specifically to zymocin. Since ACNase subunits of DrT and PaT are detectably similar but no similarity exists between DrT/PaT and zymocin, such similarity/non-similarity is apparently recognized by the immunity proteins. The initial detection of partial zymocin resistance in a strain carrying PGKpr-KlORF3 in the nucleus as well as pGKL2 in the cytoplasm [17] may be correlated to the fact that the fusion of KlORF3 to the PGKpr did not eliminate the upstream conserved sequence (UCS) element of KlORF3. Since UCS sequences have been shown to be sufficient for mediating cytoplasmic transcription [28] and toxin resistance was only seen when pGKL2, i.e. the UCS-recognizing RNA polymerase was present, it appears possible, that partial zymocin immunity was due to cytoplasmic transcription of the PGKpr-KlORF3 construct, which may have been located transiently in the cytoplasm after transformation. Such transient cytoplasmic availability may constitute the basis for the observed partial zymocin resistance as opposed to full immunity in this study.
Only rather recently nuclear sequences of plasmid and viral origin (NUPAVs) were detected which—eponymously—result from evolutionary capture of plasmid and VLE-genes by the yeast nucleus [32]. Indeed, cytoplasmic VLE based genes can frequently and repeatedly be trapped by the nucleus, as explicitly shown for pDH1A from Debaryomyces hansenii, which represents the most recent ancestor of NUPAVs so far known [33]. Taken also into consideration that some ORFs of VLEs, such as the toxin genes have been cloned and successfully expressed from nuclear vectors [19,34,35] the risk is immediately imposed on a VLE system that upon nuclear immunity gene capture autoselection is disabled, which is yet mandatory for VLE long term propagation. While chromosomally encoded yeast killer toxins are traditionally considered as factors beneficial to the producer cell due to the ability to eliminate competitors, VLE encoded toxins additionally or even predominantly serve to counterselect for spontaneous plasmid free segregants, clearly resembling the autoselective properties of bacterial toxin/antitoxin systems [16,21,36]. In Pichia acaciae, Kluyveromyces lactis and Debaryomyces robersiae toxin encoding cytoplasmic VLEs can be easily eliminated under laboratory conditions, which in all cases generates toxin sensitive segregants. Such situation differs from the vast majority of chromosomally encoded toxins, which are routinely not active against the producing species which supports that VLE encoded killer toxins function to kill spontaneous VLE-free segregants. However, such function does not exclude additional benefits to producer cells that are provided by the ability to kill other yeasts in a given environment. Since VLE-derived NUPAVs in different stages of degeneration can be detected in various yeast genomes [32,33], the high A/T content of VLEs in general may serve to minimize the potential for domestication of VLE based genes by the host, which might be particularly relevant for the immunity function that, if domesticated and separated from a toxin encoding VLE, would eliminate the positive selective pressure incurred by the toxin on the maintenance of the VLE system. In other words, spontaneous VLE-free segregants would no longer be eliminated in an environment of VLE containing, toxin secreting sister cells. Interestingly, among the VLE derived NUPAVs described by Frank and Wolfe, 2009, immunity genes constitute the largest group. We identified an additional immunity derived NUPAV in the yeast Pichia sorbitophila [37], which appears to be almost intact and closely related to the DrORF5/PaORF4 genes (S3 Fig). The gene (Piso0_001880) located close to the end of chromosome F spans 729 bp and contains at its 5’end an extended region of similarity to the immORFs where, however, the ATG and the VLE promoter appear to be lost, leading to the annotation of an internal ATG as the gene’s startcodon. Importantly, Piso0_001880 has an A/T content of 75.5% which resembles the typical VLE characteristics and differs significantly from P. sorbitophila genome average (58.6%). We assume that Pios0_001880 represents a rather recent VLE derived domestication in an early stage of degeneration that may be related to the nuclear expression barrier caused by extreme A/T content. In support, no entirely intact immunity-NUPAV has been identified so far, suggesting a general incompatibility of A/T rich VLE genes with the nuclear transcript processing machinery.
The cloning host Escherichia coli DH5αF’ was grown in Luria-Bertani (LB) medium (0.5% yeast extract, 1% peptone, 0.5% NaCl) supplemented with ampicillin (100 μg ml-1) at 37°C. Yeast strains used in this study are listed in S1 Table. They were grown either in YEPD medium (1% yeast extract, 2% peptone, 2% glucose) or in yeast nitrogen base (Difco, Detroit, MI, USA) at 30°C. Transformation of S. cerevisiae was performed according to the PEG/lithium acetate method [38].
Plasmids used in this study are listed in S2 Table. The immORFs were amplified by PCR using total DNA of the killer yeasts P. acaciae, D. robertsiae, K. lactis as template and the primers listed in S3 Table (PaORF4: PO4-NdeI-rv and PO4-fw, KlORF3: KlO3rev_NdeI and KlO3for, DrORF5: DrO5rev_NdeI and DrO5for). The PCR products were blunt-end cloned into EcoRV restricted pSK- plasmid to yield pSKPaO4, pSKKlO3 and pSKDrO5. The pSKDrO5 plasmid was modified by site-directed mutagenesis to remove a DrORF5-internal NdeI restriction site using the primers mut_NdeI_for and mut_NdeI_rev. The A/T decreased gene versions PaORF4ms and KlORF3ms were synthesized by GeneArt (Regensburg, Germany) and delivered in a vector containing NdeI and HindIII restriction sites upstream and downstream of the respective ORF. All ORFs were released from their vectors via NdeI and HindIII and cloned into a likewise restricted pSKpADH1. The ADH1pr-immORF fusions were then ligated into the 2μ vector YEplac195 using the restriction sites KpnI and SacI (YEPaO4, YCPaO4 YEKlO3, YEDrO5, YEPaO4ms) or SmaI and HindIII (YEKlO3ms). For coexpression of KlORF3ms and γ-toxin, the EcoRI-BglII insert of pABY1643 (GAL1pr-γ-toxin-GST; [10]) was subcloned into EcoRI-BamHI digested YEplac181.
For the microtiter plate assay the partially purified toxins PaT and zymocin were obtained from culture supernatants of P. acaciae NRRL Y-18665 and K. lactis AWJ137 by ultrafiltration as described previously [20]. Different toxin concentrations were applied in microtiter plates as described in Klassen et al., [39]. Cell growth was monitored photometrically in a Multiscan FC Microplate Photometer (Thermo Fisher Scientific, Waltham, MA, USA) at 620 nm.
To check the sensitivity of a strain in the eclipse assay, a drop of 7 μl of the cell suspension was spotted on a YEPD agar plate. The killer strain was then placed at the rim of the drop and the plate was incubated at 30°C overnight.
Effects of intracellularly expressed, galactose-inducible toxin subunits on the growth of certain strains were checked with the drop dilution assay. Cultures were serially diluted and 5 μl aliquots were spotted on YNB medium containing glucose (repressing condition) or galactose (inducing condition) and incubated for several days at 30°C. Zymocin containing YPD plates were prepared by spreading 300 μl of filter sterilized, concentrated supernatant (RCF 5) of K. lactis AWJ137.
Total RNA of different immunity ORF expressing yeast strains was isolated after over night cultivation in YEPD medium. 1.5 μg of each RNA sample were separated by a denaturating 1.5% agarose gel electrophoresis (20 mM MOPS, 8 mM sodium acetate, 1 mM EDTA, 0.74% formaldehyde, pH 7.0) and blotted onto a positively charged nylon membrane (Roche Diagnostic GmbH, Mannheim, Germany). The blotting success and RNA integrity were controlled by methylene blue staining (0.02% methylene blue, 0.3 M sodium acetate, pH 5.2). Hybridization was performed at 57–61°C overnight in hybridization buffer containing 50% formamid and a DIG-labelled RNA probe specific for the mRNAs of PaORF4, KlORF3 and DrORF5, respectively. Probes were prepared by amplifying the gene sequences using the primers (PaORF4_probe_for/PaORF4_probe_revT7, KlO3proberevT7/KlO3res1, DrO5_probe_rev_T7/DrO5_rs1, S3 Table) and labelled with the DIG RNA Labeling Kit (SP6/T7) (Roche Diagnostic GmbH, Mannheim, Germany) according to the manufacturer’s instructions. For detection a phosphatase-conjuncted anti-DIG antibody and a chemilumiscent alkaline phosphatase substrate CDPstar (Roche Diagnostic GmbH, Mannheim, Germany) were applied, and signals were visualized by exposure to X-Ray films.
For cDNA synthesis the RevertAid H minus first strand cDNA synthesis kit (Fermentas, St. Leon-Rot, Germany) was applied according to the manufacturer’s instructions. All primers used for cDNA synthesis are listed in S3 Table.
Total RNA was isolated as previously described (Klassen et al., 2008) or, alternatively, by making use of the RNeasy Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions.
cDNA was synthesized using random hexamer primers and total RNA as template. 1 μl of cDNA was used as the template for PCR reactions applying the Phusion High fidelity DNA polymerase (Thermo Fisher Scientific) and primer combinations PaO4intr-fw/PaO4intr-rv (PaORF4), PaO4ms_for/PaO4ms_rev (PaORF4ms), KlO3_rs3/KlO3pf (KlORF3) and KlO3ms_rev/KlO3ms_for (KlORF3ms) (for primer binding positions see Fig 5C).
Primer design for 3´ RACE experiments was based on the protocol of Scotto-Lavino et al., [40]. Polyadenylated RNA was transcribed to cDNA using the poly(A) complimentary primer QT22. The cDNA was used as template for PCR with primer Q0 (binds to a QT22 anchor) and an immORF specific primer at an annealing temperature of 58°C. PCR products were separated on an 1% agarose gel and extracted fragments were cloned into vector pSKpADH1 via NdeI and HindIII restriction sites. Isolated plasmids were analyzed by sequencing.
For linker ligation 2 μg of total RNA were mixed with 2 μg of a phosphorylated DNA oligo nucleotide (Oligo_5P_3ddC) and ligated using T4 RNA ligase (NEB, Frankfurt, Germany) for 1.5 h at 37°C. Column purified reactions were used for cDNA synthesis with Oligo_rev as primer. cDNA of the immunity gene was amplified using the primers Oligo_rev and KlO3rev_NdeI as and then cloned into pSKpADH1 via NdeI and EcoRV restriction sites. Cloned fragments were analyzed by sequencing.
|
10.1371/journal.ppat.1001089 | The Canine Papillomavirus and Gamma HPV E7 Proteins Use an Alternative Domain to Bind and Destabilize the Retinoblastoma Protein | The high-risk HPV E6 and E7 proteins cooperate to immortalize primary human cervical cells and the E7 protein can independently transform fibroblasts in vitro, primarily due to its ability to associate with and degrade the retinoblastoma tumor suppressor protein, pRb. The binding of E7 to pRb is mediated by a conserved Leu-X-Cys-X-Glu (LXCXE) motif in the conserved region 2 (CR2) of E7 and this domain is both necessary and sufficient for E7/pRb association. In the current study, we report that the E7 protein of the malignancy-associated canine papillomavirus type 2 encodes an E7 protein that has serine substituted for cysteine in the LXCXE motif. In HPV, this substitution in E7 abrogates pRb binding and degradation. However, despite variation at this critical site, the canine papillomavirus E7 protein still bound and degraded pRb. Even complete deletion of the LXSXE domain of canine E7 failed to interfere with binding to pRb in vitro and in vivo. Rather, the dominant binding site for pRb mapped to the C-terminal domain of canine E7. Finally, while the CR1 and CR2 domains of HPV E7 are sufficient for degradation of pRb, the C-terminal region of canine E7 was also required for pRb degradation. Screening of HPV genome sequences revealed that the LXSXE motif of the canine E7 protein was also present in the gamma HPVs and we demonstrate that the gamma HPV-4 E7 protein also binds pRb in a similar way. It appears, therefore, that the type 2 canine PV and gamma-type HPVs not only share similar properties with respect to tissue specificity and association with immunosuppression, but also the mechanism by which their E7 proteins interact with pRb.
| Human papillomaviruses (HPVs) are estimated to cause the most common sexually transmitted infection in the world, and these infections are recognized as the major cause of cervical cancer. One of the papillomavirus oncoproteins, E7, plays a major role in both the viral life cycle and progression to cancer. In cells E7 associates and inactivates pRb, a tumor suppressor protein. For the vast majority of papillomaviruses, E7 binds to pRb using a small amino acid sequence, LXCXE. However, we have now identified a papillomavirus E7 protein that lacks the LXCXE domain yet still binds and degrades pRb. This E7 protein, derived from a carcinogenic canine virus, uses its C-terminal domain to bind pRb. In addition, we discovered that a family of papillomaviruses, the gamma type HPVs, also lacks the LXCXE domain and binds pRb using a similar mechanism.
| Human papillomaviruses (HPVs) mediate the initiation and maintenance of cervical cancer [1], [2]. Based upon DNA sequence homology, there are more than 150 different HPV genotypes, 40 of which infect anogenital and oral mucosa [3]. In addition to genotyping, HPVs can also be classified as low-risk and high-risk based on the clinical prognosis of their associated lesions. Low-risk HPVs cause benign epithelial hyperplasias while high-risk HPVs cause lesions that can progress to malignancy. Integration of the HPV genome into a host cell chromosome is a frequent event during malignant progression and it may play a significant role in dysregulated expression of the HPV E6 and E7 proteins [4]. The high-risk HPV E6 binds to several cell targets, including p53, Myc, E6AP, PDZ proteins and other cellular proteins to alter apoptotic/growth regulatory pathways and induce cellular telomerase activity [5]. The E7 protein binds and sequesters pRb and directs its ubiquitin-mediated proteolysis [6], thereby altering E2F-regulated genes and allowing cells to enter the S phase of the cell cycle.
The E7 oncoprotein is approximately 100 amino acids in length and contains two highly conserved regions (CRs), the amino-terminal CR1 and CR2 domains [7]. The E7 CR1 and CR2 domains share strikingly high homology with the CR1 and CR2 regions of adenovirus (Ad) E1A and related sequences in simian vacuolating virus 40 (SV40) large tumor antigen (T Ag) [4], [8]. For each of these viruses, the CRs contribute significantly to cell transformation [9], [10], [11], [12]. A conserved Leu-X-Cys-X-Glu (LXCXE) motif in the CR2 domain of HPV E7, as well the ones in Adenovirus E1A and SV40 LT, are necessary and sufficient for association with pRb [13]. The crystal structure of pRb bound to an E7 peptide was resolved, and revealed that LXCXE of HPV E7 binds entirely through the B domain of pRb [14]. For high risk HPV, the LXCXE motif is also required for pRb degradation[15], [16]. The carboxyl-terminal domain of E7 consists of a metal binding domain composed of two CXXC motifs separated from each other by 29 amino acids [14]. This zinc-binding region is important for E7 dimerization and intracellular stabilization [10], [17]. The carboxyl-terminal domain also contributes to E7 association with chromatin-modifying enzymes, particularly histone deacetylases and histone acetyl transferases [18]. Although the carboxyl-terminus of high-risk HPV E7 does not appear to have a direct role in the binding and degradation of pRb [15], [19], it has been proposed to be important for releasing E2F from pRb [20], [21].
Papillomavirus can be isolated from a wide range of vertebrates, ranging from birds to manatees [22], [23] and infection by these viruses is, in general, species-specific. The canine papillomavirus model has been used successfully for vaccine [24], [25] and therapeutics studies [26]. Recently, our lab isolated and sequenced canine papillomavirus type 2 (CPV-2, previously named CfPV2), which we showed to be an epidermotropic virus that occurred frequently in immunosuppressed animals and induced tumors that progressed to aggressive cancers [27], [28]. The E7 gene of CPV-2 appeared unique in that it lacked the conserved LXCXE motif. In this study we show that this variant E7 protein is still able to bind and degrade pRb and that the primary domain for binding pRb is in its carboxyl-terminus. Interestingly, upon searching the HPV genome database, we observed that the gamma HPVs also contained the variant LXSXE E7 domain and, similar to the canine E7 protein, could still bind and degrade pRb. In addition to this similarity, we also noted that both the type 2 canine PV and gamma HPVs both exhibited a tropism for skin and for immunocompromised hosts.
Wild-type CPV-2 E6 and E7 were generated by PCR using CPV-2 genome as template [27] and subcloned into retrovirus vector pLXSN (Clontech) at the sites EcoR I and BamH I. The CPV-2 E7 mutants CPV-2 E7 ΔLXSXE, S26C, and S26G were generated by using the QuikChange Site-Directed Mutagenesis kit (Stratagene), and CR1, CR2, CR1CR2 and CT were generated by PCR using pLXSN.CPV-2E7 as template. All the wild type E7 and mutants were cloned into the sites EcoR I and Not I of the pGEX4T-2 (GE Healthcare) for GST fusion protein expression. CPV-2 E7 and mutants with a hemagglutinin (HA) epitope tag at their amino terminal or carboxyl terminal were generated by PCR. PCR products were then subcloned into the mammalian expression vector pJS55 [29] at the sites EcoR I and BamH I. All plasmids were sequenced to confirm the presence of corresponding mutations. All primer sequences used in subcloning and site-directed mutagenesis please see Supporting Information S1. All the wild type and mutants of HPV-4 E7 DNA (GenBank NC_001457.1) were synthesized (Celtek Bioscience), and cloned into pGEX4T-2 (GE Healthcare) for GST fusion protein expression.
U2OS cells, Hela cells and SD3443 cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM) (Invitrogen) supplemented with 10% Fetal Bovine Serum (FBS). Primary human keratinocytes were derived from neonatal foreskins as described previously [30] and were grown in Keratinocyte-SFM medium (Invitrogen).
U2OS cells were co-transfected with RcCMV-Rb and pJs55, pJS55-HA.HPV16 E7, pJS55-HA.CPV-2 E7ΔLXSXE, pJS55-HA.CPV-2 E7CR1CR2, pJS55-HA.CPV-2 E7C-RT or pJS55-HA.CPV-2 E7 using Lipofectamine 2000 (Invitrogen) as specified by the manufacturer. Hela cells were transfected with pJs55, pJS55-HA.HPV16 E7 or pJS55-HA.CPV-2 E7 using Lipofectamine 2000 (Invitrogen) as specified by the manufacturer. Cell were harvested and lysed by RIPA buffer (25 mM Tris•HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS) 24 hours post transfection.
To prepare retrovirus stocks, SD3443 cells were transfected with E7 retrovirus constructs using Fugen (Roche applied science, US.) as specified by the manufacturer. Culture supernatants containing retrovirus were collected 48 h post-transfection. Viral titers of the supernatants were determined using 3T3 cells. The primary HFK cells (passage 0) were infected at a multiplicity of 10 PFU/cell with retrovirus expressing wild type E6, E7 or E7 mutants. Retrovirus-infected cells were selected in G418 (50 ng/ml) for 2 days.
GST and GSTE7 fusion proteins were expressed in BL21pLysS cells (Invitrogen). The cells were induced with 100 µM isopropyl-β-D-thiogalactopyranoside (IPTG) 6 hours at 25°C once the optical density at 600 nm reached 0.8–1.0. Recombinant CPV-2 E7 and mutants were purified from the supernatant of disrupted cells by glutathione-Sepharose chromatography as previously described [24].
Proteins were extracted from cells and measured concentration as previously described[31]. Proteins were separated on a 4 to 20% Tris-glycine gradient gel (Novex) and then were electrophoretically transferred to an Immobilon-P polyvinylidane difluorid (PVDF) membrane (Millipore). The primary antibody was used at a dilution of 1∶1,000 or 1∶3,000. The secondary antibodies, alkaline phosphatase-conjugated goat anti-mouse IgG and anti-rabbit IgG (Tropix) antibodies, were used at a dilution of 1∶2,000. Western blots were visualized by using SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific). The following commercial antibodies were used: for pRb (1∶1000 dilution), Rb (4H1) Mouse mAb (Cell Signaling technology); for glutathione S-transferase (1∶3000 dilution), catalog no. 3818-1 (Clontech); for HA (1∶1000 dilution), HA.11 clone 16B12 (Covance).
For GST pull-down assays, Jurkat cell or CPEK cell nuclear extract (50 µg) was incubated with 5 µg of GST or GST fusion protein in binding buffer [20 mM Hepes/150 mM KCl/4 mM MgCl2/1 mM EDTA/0.02% Nonidet P-40/10% glycerol/0.035% 2-mercaptoethanol/1% (vol/vol) Sigma protease inhibitor mixture] and rocked for 1 h at 4°C. Glutathione-Sepharose beads (Amersham Pharmacia Biosciences) were added to each reaction and rocked for another 1 h at 4°C. The beads were then washed with 1 ml of washing buffer (125 mM Tris, 150 mM NaCl, pH 8.0) four times and boiled with 2× SDS sample buffer, and the proteins were separated by SDS/PAGE. Western blots were used to measure the level of pRb and GSTE7 proteins. The bands of pRb and GSTE7 proteins were quantified by densitometry using Quantity One (BioRad). The relative binding activities were calculated using pRb bound by wild type CPV-2 E7 GST fusion protein as 100%, and normalized with GSTE7 bands.
For co-immunoprecipitation assays, N-terminal HA-tagged CPV-2 E7 proteins were immunoprecipitated with a polyclonal anti-pRb antibody (Santa cruz). C-terminal HA-tagged CPV-2 or HPV16 E7 proteins were immunoprecipitated with a polyclonal anti-HA antibody (Santa Cruz). Bead washing buffer was 25 mM Tris•HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS, 1% Sigma protease inhibitor mixture. The pulled down complexes were resolved on a 4 to 20% gradient gel and then analyzed by Western blotting using either anti-pRb antibody, anti-HA antibody or anti-cullin2 antibody (Invitrogen).
Cells in a 3.5 cm diameter dish were lysed with 1 ml TRIZOL (Invitrogen). Total RNA was isolated according to manufacturer's protocol. Reverse transcription PCR was performed using ONE STEP RT-PCR KIT (QIAGEN) as specified by the manufacturer. All primer sequences and condition please see Supporting Information S1.
Multiple sequence alignments of E7 were prepared using Clustal W. The phyllogenetic analysis was conducted using the Mega version 4.0 [32].
Several studies have demonstrated that the conserved LXCXE motif in the HPV E7 CR2 domain is necessary and sufficient for binding pRb [4]). Studies have also revealed that the substitution of C or E or complete deletion mutation of the LXCXE motif destroys pRb binding. Without binding to the pRb, E7 is unable to degrade pRb [4]. Our laboratory recently isolated CPV-2 from footpad and interdigital papillomas of immunosuppressed dogs. Sequencing revealed that the CPV-2 E7 protein contained the typical two C-X-X-C motifs within its carboxyl terminal half but lacked the conserved pRb binding site (LXCXE) which is present in COPV (CPV-1) E7 and most HPVs (Figure 1A). CPV-2 E7 has a serine (amino acid 26) in the position of cysteine in the LXCXE motif. In order to test whether CPV-2 E7 could degrade pRb, canine E7 was transduced into human keratinocytes (HFKs), canine kidney cells (MDCKs) or canine keratinocytes (CPEKs) by using retrovirus infection. Cell lysates were collected, and pRb levels were measured by western blots. Surprisingly, despite lacking the conserved LXCXE motif, CPV-2 E7 was still able to degrade pRb in HFKs, MDCKs and CPEKs (Figure 1B, C and D). To test whether the lower level of pRb was due to a change at the transcriptional level, RT-PCR was performed to measure the level of pRb mRNA. There was no significant difference between the amount of pRb mRNA in control cells and cells with CPV-2 E7 (Figure 1E). In addition, treatment of the E7 expressing cells with the proteasome inhibitor, MG132, restored the level of pRb (Figure 1F). These data suggest that the reduction of pRb by CPV-2 E7 occurs at the protein level rather than mRNA level, and that degradation is most likely responsible.
The degradation of pRb by HPV16 E7 requires high affinity binding [19]. Since CPV-2 E7 lacks the conserved pRb binding motif, LXCXE, there could be two possibilities for the high affinity binding of pRb by CPV-2 E7. It could be either that the LXSXE motif has the same binding properties as LXCXE, or that CPV-2 E7 has an alternative dominant binding site. We generated E7 mutants (Figure 2A) with mutations within the LXSXE domain to investigate whether the LXSXE motif exhibits similar binding to pRb as LXCXE. The GST E7 wild type and mutant fusion proteins were purified from bacteria (Figure 2B), and tested for their binding to pRb. As demonstrated in Figure 2C, wild type CPV-2 E7 binds well to pRb. Substitution of serine26 to cysteine in CPV-2 E7 significantly increased the interaction between E7 and pRb. This is in agreement with mutagenesis studies showing that mutation of HPV16 E7 cysteine24 to serine substantially decreased the binding activity [13], [33]. In the studies of HPV16 E7, the substitution of cysteine in LXCXE for glycine abolished the binding between E7 and pRb [14], [33], [34], [35]. However, for CPV-2 E7, the pRb binding activity of the glycine mutant protein was still retained (Figure 2C). More importantly, the deletion of the entire LXSXE motif did not significantly affect the ability of CPV-2 E7 to bind pRb. This is very surprising since the deletion of the LXCXE motif in HPV 16 E7 totally abolished pRb binding [14], [33], [34], [35]. Thus, while the CPV-2 LXSXE motif exhibits lower pRb binding than the conserved LXCXE motif, it is clear that there is an alternative pRb binding site in the CPV-2 E7 protein.
In order to locate potential alternative pRb binding sites in CPV-2 E7, several CPV-2 E7 truncation mutants were generated (Figure 3A). CPV-2 E7 protein was divided into CR1 (amino acids 1 to 15), CR2 (amino acids 16 to 38) and the C terminal domains (amino acids 39 to 98). The GST E7 wild type and mutant fusion proteins were purified from bacteria (Figure 3B). A binding assay with GSTE7 mutants and pRb was performed. In agreement with earlier studies [4], [14], [36], HPV16 E7CR1CR2 bound pRb much more efficiently than the HPV16 E7 carboxyl-terminus domain (Figure 3C), with 6 times more pRb being bound by the combined CR domains. In contrast, the carboxyl-terminus of CPV-2 E7 has much higher binding than the CR1 and CR2 domains, on the average 7 times higher. Importantly, similar binding affinities were also observed between the canine pRb and CPV-2 E7 (Figure 3D), indicating that these unique binding interactions were not due to differences in the species of Rb used for analysis. Thus, CPV-2 E7 uses its carboxyl-terminus instead of the CR1 and CR2 domains to associate with pRb.
To verify that the above in vitro binding studies were relevant to in vivo conditions, U2OS cells transduced by HA-tagged CPV-2 E7 constructs were used to study the in vivo association of CPV-2 E7 protein and pRb. Co-precipitation experiments were performed with the anti-pRb antibody (Figure 4). In contrast to results with HPV 16 E7, the CPV-2 E7 mutant deleted of the LXCXE-like motif bound pRb as well as the wild-type E7 did. Furthermore, the carboxy-terminal domain of CPV-2 alone bound to pRb very well. Thus, both in vitro and in vivo studies indicate that the E7 carboxyl-terminus mediates pRb binding. The level of canine E7 CR1CR2 construct was not detectable in the cell since it appears to be unstable. The doublet band noted for both wild-type E7 and the LXCXE deletion E7 proteins is most likely due to alkylation by protease inhibitors during the IP procedure. In previous studies, we also observed two distinct forms of HPV-16 E7 and showed that they were generated in vitro by the alkylating reagents, TPCK and TLCK [37]. These reagents are used as a component of a protease-inhibitor cocktail during IP studies to prevent protein degradation. We identified cysteine 27 near the amino terminus of HPV-16 E7 as the alkylation target. In our current study, we did not observe this modification of HPV-16 E7, apparently due to interference by the amino-terminal epitope tag. In the published study where we observed alkylated HPV-16 E7, we had used an E7 protein tagged at its C-terminus. We presume that the CPV-2 E7 protein is being modified in a similar fashion, although different sites might be altered than in HPV-16 E7.
Since the mechanism used by CPV-2 E7 to bind pRb is different from that used by HPV E7, it was also possible that the two E7 proteins used alternative methods to degrade pRb. To test this possibility, a series of E7 deletion and single amino acid substitution mutants were generated. Keratinocytes were transduced with retrovirus encoding either wild type E7 or E7 mutants. Cell lysates were collected, and the level of pRb was measured with immunoblots. Surprisingly, the LXSXE-deletion mutant (which retains pRb binding) lost the ability to degrade pRb (Figure 5A). In addition, when the serine at 26 was changed to cysteine (which increases pRb binding), the degradation of pRb was not enhanced (Figure 5A). Thus, although the primary pRb binding of CPV-2 resides in the carboxyl-terminus, we observed that the amino-terminal LXSXE sequence is necessary for the degradation of pRb. Neither the E7 amino- nor carboxyl-terminal domains could independently degrade pRb (Figure 5B). This is in contrast to studies performed on high risk HPV E7 (reviewed by Munger [38]) showing that the sequences important for binding and degradation of pRb localized to CR1 and CR2. Another difference between HPV16 E7 and CPV-2 E7 is highlighted by previous studies showing that HPV16 E7 associates with the cullin 2 ubiquitin ligase complex and that this association contributes to degradation of pRb [4], [14], [36]. This does not appear to be true for CPV-2 E7 protein. Co-immunoprecipitation assays failed to detect any association of Cullin2 with CPV-2 E7 (Figure 5C).
Our current results demonstrate that at least one animal papillomavirus uses a different mechanism to bind and degrade pRb. To investigate whether some HPVs might use a similar alternative binding mechanism, we screened a papillomavirus phylogentic tree based upon the E7 protein sequence (Figure 6A). E7 proteins from 33 HPVs and 15 animal papillomaviruses were selected according to their genus [39] and aligned using Clustal W [40] with MEGA version 4.0 [32]. Based on the alignment, a phyllogenetic tree was assembled by using the minimum evolution method with MEGA version 4.0 [32]. CPV-2 E7 was most closely related to the genus gamma-papillomaviruses (HPV-4, 48, 50, 60, 65, 88, 95 and 116) and only distantly related to the genus lambda-papillomaviruses (CPV-1 and Felis domesticus papillomavirus). Interestingly, the alignment of E7 proteins revealed that all the gamma HPVs lack the LXCXE motif (Figure 6B) and, indeed, nearly all these gamma-HPVs contain the same LXSXE sequence found in CPV-2. One of the gamma-HPVs, HPV60, contains alanine rather than serine in the LXSXE sequence. To test the binding of a representative gamma-HPV E7 to pRb, HPV-4 E7 was synthesized and cloned into an expression vector. HPV-4 E7 protein was divided into CR1 (amino acids 1 to 15), CR2 (amino acids 16 to 38) and carboxyl-terminal domains (amino acids 39 to 100) (Figure 6C). Wild type HPV-4 E7 and truncation mutants were expressed as GST fusion proteins (Figure 6D) and tested for their abilities to bind pRb. Similar to CPV-2 E7, the HPV-4 E7 protein contained an LXSXE motif and bound pRb (Figure 6E). More interestingly, as shown in Figure 6E, the carboxyl- terminus of HPV-4 E7 bound pRb more efficiently than the CR1CR2 domains. It appears, therefore, that the gamma-type HPVs and CPV-2 share a mechanism by which their E7 proteins interact with pRb via the carboxyl-terminal domain.
Due to the similar ability of the CPV-2 and HPV-4 E7 proteins to bind pRb, we also evaluated whether HPV-4 E7 could degrade pRb. In HFKs, HPV-4 E7 reduced the level of pRb in transduced cells, although somewhat less than observed with CPV-2 or HPV16 E7 (Figure 7A). To test whether the lower level of pRb protein might be due to altered gene transcription, we measured pRb mRNA levels by RT-PCR. There was no significant difference in the amount of pRb mRNA in the control cells compared to cells expressing CPV-2 E7 (Figure 7B), indicating that the pRb protein changes were post-translational. More importantly, treatment of the E7 expressing cells with proteasome inhibitor, MG132, restored the level of pRb protein (Figure 7C). These data, similar to that for CPV-2, suggest that the reduction of pRb by HPV-4 E7 is most likely the result of protein degradation.
Small DNA tumor viruses, such as HPV, Adenovirus, and Polyomavirus, produce viral oncoproteins that can interact with pRb and alter its function. Targeting pRb appears important for the ability of these viruses to regulate E2F and cell DNA replication and to complete the virus life cycle. All these oncoproteins, E7, E1A, and LT, use a conserved CR2 domain and LXCXE motif to bind pRb [41]. However, for CPV-2 E7, the LXCXE-like motif is not necessary for association with or degradation of pRb. It appears that CfPV has evolved an alternative mechanism to bind pRb by using the E7 carboxyl-terminal domain.
Although the HPV E7 carboxyl-terminus has been proposed to have an independent, low affinity pRb binding site [14], [36], HPV16 E7 mutants with a deletion of the LXCXE motif in CR2 fail to associate with pRb family members as determined by Western blotting and extensive proteomic analyses of associated cellular protein complexes [4]. In contrast, the carboxyl-terminus of CPV-2 E7 exhibits greater pRb binding than the CR1 and CR2 domains. Even more interesting is the finding that the CPV-2 E7 mutant deleted of the LXSXE domain still can bind to pRb with high efficiency. Furthermore, the carboxyl-terminus of CPV-2 E7 alone can bind pRb in vitro and in vivo. The carboxyl-terminus of HPV E7 has been proposed to be important for releasing E2F from pRb [20], [21]. It will be interesting to map the association site on pRb, and to determine whether the binding induces the release of E2F from pRb.
During preparation of this manuscript, three new canine papillomaviruses were identified [42]. One of the viruses, CPV7, shares high sequence homology with CPV-2 and its predicted E7 ORF encodes a protein with the LXSXE motif. Overall, however, 5 of the 7 identified canine papillomaviruses contain the LXCXE motif in their E7 protein. CPV-1 E7, which has LXCXE motif, degrades pRb in cells as anticipated (data not shown).
As shown in this study, the LXSXE motif is not limited to canine papillomaviruses; the gamma genus of HPVs also have the same motif. This genus consists of eight HPVs, 7 of which contain the E7 LXSXE motif. Interestingly, the gamma HPVs have several other similarities to CPV-2. First, they induce cutaneous rather than mucosal lesions. Second, they most likely persist in the population as subclinical infections and induce visually detectable tumors only under conditions of immunosuppression. Third, their tumor cells are characterized by histologically-distinguishable intracytoplasmic inclusion bodies [39]. The canine model may provide a new approach for studying the biology of this unique category of papillomaviruses and their stringent regulation by the host immune response.
|
10.1371/journal.pgen.1000293 | The Genomic Analysis of Lactic Acidosis and Acidosis Response in Human Cancers | The tumor microenvironment has a significant impact on tumor development. Two important determinants in this environment are hypoxia and lactic acidosis. Although lactic acidosis has long been recognized as an important factor in cancer, relatively little is known about how cells respond to lactic acidosis and how that response relates to cancer phenotypes. We develop genome-scale gene expression studies to dissect transcriptional responses of primary human mammary epithelial cells to lactic acidosis and hypoxia in vitro and to explore how they are linked to clinical tumor phenotypes in vivo. The resulting experimental signatures of responses to lactic acidosis and hypoxia are evaluated in a heterogeneous set of breast cancer datasets. A strong lactic acidosis response signature identifies a subgroup of low-risk breast cancer patients having distinct metabolic profiles suggestive of a preference for aerobic respiration. The association of lactic acidosis response with good survival outcomes may relate to the role of lactic acidosis in directing energy generation toward aerobic respiration and utilization of other energy sources via inhibition of glycolysis. This “inhibition of glycolysis” phenotype in tumors is likely caused by the repression of glycolysis gene expression and Akt inhibition. Our study presents a genomic evaluation of the prognostic information of a lactic acidosis response independent of the hypoxic response. Our results identify causal roles of lactic acidosis in metabolic reprogramming, and the direct functional consequence of lactic acidosis pathway activity on cellular responses and tumor development. The study also demonstrates the utility of genomic analysis that maps expression-based findings from in vitro experiments to human samples to assess links to in vivo clinical phenotypes.
| It is well recognized that tumor microenvironments play an important role in modulating tumor progression in human cancers. Although previous studies have highlighted the importance of hypoxia, there is limited knowledge on the effects of other components in tumor microenvironments. Therefore, we use gene expression to compare and analyze how cells respond to lactate, acidity, and hypoxia, as well as how these responses can be utilized to predict the clinical outcomes of patients with breast cancers. We uncover an unexpected association with better clinical outcome of the strong lactic acidosis and acidosis response in breast cancers as a result of their abilities to inhibit glycolysis and favor oxidative phosphorylation for energy generation. This effect is caused by not only the repression of the gene expression of glycolysis genes but also the inhibition of Akt activation of cells exposed to lactic acidosis and acidosis. In conclusion, we propose that lactic acidosis and acidosis to be considered as independent prognostic factors for human cancers.
| The tumor microenvironment is characterized by oxygen depletion (hypoxia), high lactate and extracellular acidosis (lactic acidosis) as well as glucose and energy deprivation [1]. These changes are largely caused by a combination of poor tissue perfusion, abnormal tumor vasculature, uncontrolled proliferation and dysregulated energy metabolism. These microenvironmental features vary widely in different tumors, reflecting heterogeneity in the metabolic status of individual tumors. They can also trigger phenotypic changes in cancer cells and directly modulate biological properties and clinical phenotypes. Although our understanding of hypoxia has advanced tremendously in recent years, relatively little is known about the role of other microenvironmental stresses, especially lactic acidosis and glucose starvation.
The accumulation of lactic acid in solid tumors is often thought to be caused by tumor hypoxia – a by-product of glycolysis as the tumor cells shift to an anaerobic mode of energy production under hypoxia or due to the altered metabolic profiles of cancer cells. In spite of this apparent mechanistic link, these two factors exhibit significant disparities in their spatial and temporal distribution in tumors [2],[3]. This may be due to the fact that some tumors exhibit a predisposition toward glycolysis even in the presence of oxygen, a phenomenon that is referred to as aerobic glycolysis or the Warburg effect [4]. Some tumors may also possess greater capacity to pump protons out to the extracellular space to create a reversed pH gradient – acidic extracellular pH (pHe) and alkaline intracellular pH (pHi) through higher expression of proton transporters [5]. Additionally, lactic acid can accumulate in poorly perfused tissue due to inefficient removal. Thus, it is important to consider these two factors separately in understanding their distinct contributions to tumor phenotypes.
Research on tumor microenvironments relies on our ability to manipulate in vitro conditions of mammalian cell growth to re-create the relevant stresses experienced in vivo. This offers a means to explore the roles of individual environmental factors on cellular behavior and phenotype. Genetic and pharmacological manipulation can then be applied to explore the molecular mechanisms and genetic circuitry underlying cellular responses. Such studies typically observe in vitro cellular behavior and make inferences about contributions to tumor progression in vivo. For example, hypoxia, when applied in vitro, has been shown to promote angiogenesis, cellular migration and energy consumption, thus providing the potential mechanisms for its association with poor clinical outcome [6],[7]. Similarly, lactic acidosis, when applied to cultured cells, has been shown to trigger calcium signaling [8], gene expression of angiogensis (e.g. VEGF, IL8) [9],[10],[11], HIF1α stabilization [12], cell death [13] and affect gene expression [14],[15],[16]. Recent studies also use genomic analysis to identify the cellular response to acidosis and high lactate [14],[16]. These results are summarized in several nice reviews [17],[18],[19]. However, inference using these observations from in vitro perturbations to in vivo cancer phenotypes is frequently challenging and indirect. Gene expression microarray signatures have provided a solution to this gap since they afford an opportunity to develop a surrogate phenotype of the in vitro state which can then be assessed in the in vivo state. This approach generates gene expression signatures from perturbations in cultured cells in vitro to represent a defined biological process [20],[21],[22], which in turn can serve as a common phenotype to recognize similar molecular features in human cancer samples in vivo. Using this approach, we have previously shown that the wound healing, hypoxia responses and various oncogenic mutations can play important roles in tumor progression [20],[21],[23]. Our current study applies this strategy to lactic acidosis, aiming to elucidate the casual roles of lactic acidosis at a molecular level and evaluate the prognostic implications. The analysis reported in this manuscript investigates these molecular mechanisms through integrative genomic analysis of lactic acidosis responses from both in vitro culture cells and in vivo human cancers.
To characterize the gene expression program generated in response to lactic acidosis, hypoxia, and combined lactic acidosis and hypoxia, we made use of human mammary epithelial cells (HMEC) brought to replicative arrest by growth factor/serum withdrawal for 24 hours. Since the HMEC represent normal epithelial cells with intact signaling components, the response elicited in HMECs is likely to reflect the cellular response not biased by genetic mutations present in cancer cell lines. We exposed HMECs to four different culture environments for 24 hours in triplicate samples: 1) control – ambient oxygen level (∼21%O2) with neutral pH; 2) lactic acidosis – 25 mM lactic acidosis with pH 6.7; 3) hypoxia (2% O2) with neutral pH; 4) combined lactic acidosis and hypoxia. We did not find a significant change in media pH at the end of the 24 hour culture. The gene expression of these HMEC samples were interrogated with Affymetrix GeneChip U133 plus 2.0 arrays to measure the expression of more than 54,000 probe sets and at least 47,000 transcripts and variants. Gene expression profiles of cellular responses to hypoxia, lactic acidosis and combined stresses were first normalized by RMA, mean centered and filtered with the criteria of at least 2 (out of 3 samples in each experimental condition) observations with at least 1.75 fold changes to select 4722 probes sets. A clustering analysis on these genes revealed that hypoxia and lactic acidosis induced distinct sets of genes (Figure 1A). The hypoxia induced gene clusters included CA9, stanniocalcin1, EGLN3, BNIP3 and many of the genes seen in our prior studies with spotted cDNA arrays [21] (Figure 1B). Interestingly, induction of some hypoxia-induced genes (e.g. CA9, Stanniocalcin1) was abolished by simultaneous lactic acidosis (Figure 1B), consistent with previous studies [24],[25].
Compared with hypoxia, the lactic acidosis response is more dramatic with alterations of many more genes. Genes induced by lactic acidosis included: PLAUR, ERBB3, CD55, interleukin 15, CXCL16, angiogenin and MHC class I genes (Figure 1B). For a subset of these lactic acidosis-induced genes, the degree of induction was further enhanced by the combined stresses of lactic acidosis and hypoxia. Among genes repressed by lactic acidosis, many are involved in cell cycle, cell proliferation and glucose metabolism (Figure 1B). We further confirmed the induction of ERBB3, CD55, and PLAUR in response to lactic acidosis, and to the combination of hypoxia with lactic acidosis, via real-time PCR (Figure 1C).
A supervised analysis of the full set of data on all 47,000 probe sets was performed using Bayesian multivariate regression analysis (BFRM) that has been utilized in a number of prior studies [26],[27] (Text S1). This analysis includes the ability to use housekeeping gene information on each chip in order to automatically correct for gene-sample specific assay artifacts. The multivariate analysis computes, among other things, gene-specific probabilities of expression changes resulting from lactic acidosis or hypoxia stress (Table S1). At a threshold of probability of expression change of 0.99 (Bayesian significance of 1%) we find 217 genes whose expression is significantly altered by hypoxia and 1585 genes by lactic acidosis; only 54 genes are affected by both individual treatments (Figure 1D). Cellular responses to lactic acidosis are more dramatic and wide-spread than the responses to hypoxia, and involve substantially distinct gene sets (Figure 1B, D).
To survey the molecular pathways triggered by lactic acidosis and hypoxia, we analyzed the Gene Ontology (GO) enrichment in the genes induced and repressed by hypoxia and lactic acidosis using GATHER [28]. Among the genes induced by lactic acidosis, we found enrichment for G-protein coupled receptor signaling, antigen processing and presentation, and cellular catabolism (Table S2). The top GO terms repressed by lactic acidosis were genes involved in cell cycle, RNA metabolism and RNA processing (Table S2). On the other hand, the top GO terms enriched in the hypoxia-induced genes included hexose metabolism, glycolysis, glucose metabolism and glucose catabolism (Table S3); while the GO terms enriched in the hypoxia-repressed genes included cell cycles and RNA metabolism (Table S3). Furthermore, we compared the GO terms enriched when cells are exposed to hypoxia and lactic acidosis together (Table S4).
Previous studies have shown that lowering the extracellular pH from 7.4 to ∼6.7 will lead to a slight lowering of intracellular pH (pHi) from 7.4 to 6.9–7.0, which is likely to be mediated by monocarboxylate transporter (MCT) proteins [29],[30]. Given the importance of MCT family proteins in the regulation of cellular response to lactic acidosis, we analyzed this family of proteins under both hypoxia and lactic acidosis (Table S5). We found that MCT-4 was induced significantly under hypoxia and repressed under lactic acidosis. The induction of MCT-4 by hypoxia has been previously reported [31].
To further assess the extent to which the hypoxia and lactic acidosis response overlaps, we used binary logistic regression to estimate the probability of activation for the overall hypoxia and lactic acidosis pathways revealed in gene expression under individual conditions. We first used the control vs. hypoxia (for hypoxia probability) and control vs. lactic acidosis (for lactic acidosis probability) groups to provide the training sets to generate respectively hypoxia (Figure 1E) and lactic acidosis (Figure 1F) gene signatures. These signatures were then used to estimate their probability in the remaining two groups of samples in the same experiment. We found that the there is only marginal hypoxia pathway activation evident in lactic acidosis samples compared to controls (0.158 (control) vs. 0.286 (lactic acidosis), p = 0.14, Figure S1A). A similar analysis on the lactic acidosis pathway in hypoxia response samples indicated slightly elevated levels of lactic acidosis relative controls (0.362 (hypoxia) vs. 0.231 (control), p = 0.005, Figure S1B). We also used the BFRM analysis to identify genes which are altered only in the presence of both hypoxia and lactic acidosis. At a threshold of 99% probability, we found 127 induced genes and 320 repressed genes in the presence of both hypoxia and lactic acidosis (Figure S2A, B). Among the GO terms enriched in the induced genes were several transcription factors involved in regulating transcriptional activities (ATF3, YY1, CPEB2/3, SAP18, SREBF2, Figure S2A). Among the GO terms enriched in the repressed genes were genes encoding proteins involved in apoptosis process (FADD, CASP6, PDCD6, ERCC3, Figure S2B). The down-regulation of these pro-apoptotic genes may be important in maintaining the cellular survival during the simultaneous presence of both environmental stresses.
High lactate (lactosis) and low pH (acidosis) often co-exist in tumor lactic acidosis, but these two factors are not necessarily present simultaneously. To determine the respective contributions of lactosis and acidosis in the lactic acidosis response, we created culture conditions to separate the lactic acidosis condition (pH 6.7 created by 25 mM lactic acid) into lactosis (25 mM sodium lactate with neutral pH) and acidosis (pH 6.7 created by HCl) conditions. We analyzed the transcriptional responses of HMECs under lactosis and acidosis via microarrays. Gene expression profiles of the cellular responses to lactosis and acidosis were normalized by RMA, mean centered and filtered with the criteria of at least 4 (out of 6 sample in each experimental condition) observations with at least 1.75 fold change. 213 probes were identified and clustered in Figure 2A. Acidosis induced a much more dramatic change in gene expression than lactosis. Acidosis-induced genes include many of the genes also induced by lactic acidosis in our previous analysis (Figure 1B). We also confirmed with realtime RT-PCR of the induction of ERBB3 and SOD2 in response to acidosis (Figure S3). The acidosis gene signatures were determined by comparing the 6 control vs. 6 acidosis samples (Figure 2B). Most genes in the acidosis gene signature were not altered under lactosis (Figure 2B). To assess the relative contributions of acidosis and lactosis to the lactic acidosis response, we compared the expression level of genes changed in the three groups. We find a high concordance between lactic acidosis and acidosis responses, with similar sets of genes induced and repressed under these two stresses (Figure 2C, Table S6). In contrast, this concordance is not present in other comparisons between lactosis vs. lactic acidosis (Figure 2C), hypoxia vs. lactic acidosis or hypoxia vs. acidosis treatment (Figure 2D). This suggests that lactic acidosis (created by lactic acid) and acidosis (created by HCl) trigger similar genetic responses, distinct from genetic responses to lactosis and hypoxia.
We previously showed that the hypoxia response elicited in cultured epithelial cells provides a molecular gauge of hypoxia response for cancerous human tissues in vivo and predicts poor clinical outcome [21]. This involves projecting the in vitro gene signature of hypoxia “response” or “pathway activity” into numerical scores on each tumor sample in the in vivo expression data to assess the corresponding predicted levels of hypoxia pathway activity in each tumor. Similar approaches have been used in other studies using in vitro generated signatures to infer in vivo responses or pathway activities in tumors [20],[21],[23],[32],[33]. The hypoxia response signature was evaluated in a number of breast data sets via a weighted average of the signature gene set based principal components analysis (full details in Statistics Supplement). Analysis of a Cox survival model indicates patients with tumors showing higher levels of hypoxia pathway activity had poorer clinical outcomes (Figure 3A, Miller), consistent with our previous studies [21]. (Reported p-values are from the variable relevant to the figure, pathway activity score for the indicated pathway, used in a Cox survival model.) Completely concordant results were also observed in three other breast cancer expression studies with different stages of diseases (Figure 3A). These datasets include a study of 286 lymph node negative early breast cancers from NKI (Wang), and two studies of invasive breast carcinomas (Sotiriou, Pawitan) [34],[35]. Similar trends are present but not statistically significant in one smaller study of 82 breast cancers with information on distant metastasis [36] (data not shown).
Given the perceived relation of lactic acidosis with hypoxia in tumors, we evaluated the prognostic value of the lactic acidosis signature using the same statistical approach. We found that tumors with high lactic acidosis response signatures have significantly improved overall survival, in contrast to that for hypoxia response signature. This association with favorable clinical outcomes is consistent across all four breast cancer datasets with their different target populations and stages of cancers (Figure 3B). Given the similarity of lactic acidosis and acidosis signatures, we also assessed the acidosis gene signature alone and confirmed that tumors with high acidosis response signature activity exhibited better clinical outcome and survival (Figure 3C). In the Miller dataset with information on p53 status of individual tumors, we also found a strong association between lactic acidosis and p53 status: the estimated “lactic acidosis pathway activity” based on gene expression is significantly higher in wild type p53 tumors than in those with p53 mutations (p = 3.288×10−11) (Figure 3D).
Among the breast cancer patients included in the Sotiriou dataset, 64 patients were treated with tamoxifen while the remaining 125 patients were untreated. Both hypoxia and lactic acidosis pathway signatures are predictive of poor and favorable outcomes respectively in the patients treated with tamoxifen (Figure S4A, B), but less so among the untreated patients in this cohort (Figure S4C, D).
To understand whether the prognostic value of lactic acidosis is evident in a cell autonomous manner, we tested the prognostic value of the lactic acidosis signature in various breast cancer cell lines grown in a controlled culture milieu. We used binary logistic regression to determine the probability of acidosis response in each of a set of breast cancer cell lines that vary in their potential for distant metastasis [36]. The acidosis signature probability varied greatly among different cell lines; cell lines with high acidosis signature demonstrated low levels of aggression in the xenograft model when compared to those with low acidosis signatures (Figure 3E) [36]. This indicates that the prognostic information contained in the acidosis signature reflects the intrinsic phenotypes of these cancer cell lines.
We also tested the prognostic value of gene signatures reflecting lactic acidosis response and hypoxia response in a multivariate survival analysis using the Sotiriou data set. When both gene signatures are included in the Cox survival model on all samples for which we have survival, ER status, tumor size, and data on node involvement, the p-values for lactic acidosis and hypoxia are 0.0379 and 0.0069 respectively. When we include these two pathway variables along with clinical variables indicating ER status, tumor size >2 cm and node involvement, the p-values are .07, .008, .85, .0032 and .77 respectively. Dropping ER status and node involvement (poor predictors for this dataset) gives p-values of .06, .008, and .002.
We further evaluated parametric Weibull survival models involving different combinations of clinical variables and expression signatures. Figure S5 presents some summary survival curves comparing “high” versus “low” risk groups based on samples “below” versus “above” the median predicted survival time. This analysis shows that survival model involving lactic acidosis, hypoxia, and node size perform almost identically to models involving all variables (Figure S5A, B). Further, models involving just clinical variables perform worse than those that include both clinical and signatures. Thus the clinical and pathway variables provide synergistic value in predicting outcomes in breast cancer patients.
Since lactic acid production and accumulation in solid tumors is likely to relate to the shift to glycolysis under hypoxia, a high degree of correlation is expected between hypoxia and lactic acid in solid tumors. What is the relationship between the degree of hypoxia and lactic acidosis responses in the same tumors? With our quantitative, probabilistic assessment of the pathway activities based on tumor gene expression, we can directly investigate the relationship between these two factors in individual tumors. Unexpectedly, the lactic acidosis response score correlated significantly in a negative fashion with the hypoxia score in all four expression studies, with R ranging from −0.35 to −0.45 and p values from 2.2E-7 to1.4E-15 (Figure 4A). Although this negative correlation between the lactic acidosis and hypoxia responses is somewhat unexpected, it is consistent with the findings that a strong lactic acidosis response predicts favorable prognosis, (Figure 3B) while strong hypoxia response predicts poor prognosis (Figure 3A). To test the potential for synergistic value when both gene signatures are combined in patient stratification, we found the ability to predict survival is significantly improved when we combine the two signatures in breast cancers survival prediction (Figure 4B).
To understand the unexpected association between lactic acidosis response and favorable clinical outcomes, we compared the pathway composition between tumors with high vs. low lactic acidosis responses in all four expression datasets using Gene Set Enrichment Analysis (GSEA) [37]. GSEA uses a Kolmogorov-Smirnov statistic to determine whether specific biological processes (represented by gene sets) are significantly enriched in a subset of breast tumors with strong lactic acidosis response [37]. Among the top gene sets enriched in tumors with strong lactic acidosis responses, two biological processes are prominent: aerobic/mitochondria respiration (e.g. Kreb cycles, electric transportation and oxidative phophorylation) and the metabolism of fatty acids and amino acids (Figure 4C, Table S7). For example, the 46 genes in the TCA cycles compiled by KEGG pathway database are highly enriched in the breast cancers with strong lactic acidosis pathways in the Pawitan study (Figure S6). Several gene sets representing aerobic respiration from other sources also show significant enrichment. When all the samples in the Pawitan study were ranked based on their lactic acidosis score, the tumors with high lactic acidosis response tend to have higher expression level of genes in TCA cycles (Figure 4D). In contrast, tumors with high hypoxia response tend to have lower expression level of genes in the TCA cycle (Figure S7). Thus a high lactic acidosis response identifies a group of breast cancers enriched in the use of aerobic respiration. It is interesting to note that several gene sets representing amino acid and fatty acid metabolism are also enriched in these tumors (Table S7), reflecting the distinct metabolic profiles and mode of energy generation in tumors with high lactic acidosis responses.
There are two major pathways for ATP-generation in mammalian cells – glycolysis or aerobic respiration. One of the fundamental properties of cancer cells is their preferential utilization of glycolysis over aerobic respiration to produce ATP. The glycolytic phenotype of cancer cells is thought to offer selective advantages since the disruption of glycolysis phenotype (e.g. silencing of LDH-A) results in stimulation of mitochondrial respiration and significantly compromises their tumorigenicity and the proliferation under hypoxia [38]. Our GSEA result suggests the lactic acidosis gene signature can identify a subgroup of tumors with a higher level of aerobic respiration and more favorable clinical outcomes. This association between high lactic acidosis activity and strong aerobic respiration is also consistent with its links to wild type p53 (Figure 3D) since p53 has been shown to redirect the metabolic pathways toward aerobic respiration [39],[40].
To investigate the possibility that lactic acidosis directly modulates the balance of energy production, we measured its influence on ATP production in cultured cells when aerobic respiration is inhibited by rotenone. In control conditions without hypoxia or lactic acidosis, we found that approximately 35 %of ATP production in Siha cells is sensitive to rotenone at 48 hours (Figure 5A). Under hypoxia, only 16% of ATP production was sensitive to rotenone at 48hours, reflecting the increased use of glycolysis for energy generation when oxygen is limited (Figure 5A). In contrast, this balance is dramatically changed under lactic acidosis with 72 % ATP production sensitive to rotenone. This effect is even more dramatic at 72 hours – while 35% of ATP production was sensitive to rotenone inhibition under control conditions, this increased to 82% under lactic acidosis and decreased to 18% under hypoxia (Figure 5A). We also tested the contribution of ATP from glycolysis with 2-DG. At 48hours, we found that 71 %of ATP production is sensitive to 2-DG. This is increased to 77% under hypoxia and reduced to 63% under lactic acidosis (Figure 5B). These results indicate that lactic acidosis redirects energy production toward the aerobic respiration in vitro, which may explain why lactic acidosis response can identify tumors with higher level of aerobic respiration.
These data reveal the distinct manner by which lactic acidosis and hypoxia redirect energy utilization – hypoxia favors the glycolytic pathways while lactic acidosis favors the aerobic respiration. This suggests these two stresses may impact cellular metabolism in distinct manner and have synergic effects on ATP inhibition when cells are exposed to simultaneous hypoxia and lactic acidosis. We measured ATP production under control, lactic acidosis, hypoxia and combined hypoxia and lactic acidosis. We found that lactic acidosis and hypoxia reduced the ATP production to ∼50% and 63% respectively in 48 and 96 hours (Figure 5C). This result indicates that while lactic acidosis also caused a reduction in energy production similar to hypoxia. When the cells were exposed to both hypoxia and lactic acidosis, ATP production was dramatically decreased to about 17.3% (on average) of the control cells (Figure 5C).
To further understand how lactic acidosis and hypoxia modulate the balance between the aerobic respiration and glycolysis, we mapped their respective effects on gene expression onto the framework of metabolic pathways in energy metabolism (Figure 5D, E). From this analysis, we found that most genes in the glycolytic pathways, including PFK, ALDO, GADPH, PGK, PGM, ENO and PK, were significantly induced by hypoxia as expected from previous studies [21] (Figure 5D, E). Hypoxia induced the expression of LDHs, the enzymes required for the conversion of pyruvate to lactate. Hypoxia also induced the expression of gene encoding pyruvate dehydrogenase kinase 1 (PDK1), which inactivates the PDH by phosphorylation and prevents the conversion of pyruvate into acetyl-CoA, the essential substrate for TCA cycles during the aerobic respiration [41]. The PDK1 induction under hypoxia directs energy generation towards glycolysis instead of aerobic respiration [41]. Similarly, the average expression of these glycolysis genes was induced under hypoxia (Figure 5F). When these glycolysis genes are analyzed for their Gene Ontology enrichment, the top GO terms, as expected, are processes involved in the glycolysis and glucose metabolism (Table S8).
In contrast, expression levels of these glycolytic genes were consistently repressed by lactic acidosis (Figure 5D, E). For example, the expression levels of phosphofructokinase, fructose-1.6-bisphophatase and lactate dehydrogenase were all induced by hypoxia and repressed by lactic acidosis (Figure 5E). Lactic acidosis reduced the expression levels of all these genes from both the baseline level under normoxia and induced levels under hypoxia (Figure 5F). The changes in expression levels of these glycolysis genes are also shown (Table S9). Overall, hypoxia-induced change in gene expression favors the utilization of glycolytic pathways for energy generation whereas lactic acidosis represses the glycolysis process and favors the use of aerobic respiration as a mode of energy production, consistent with our experimental data of higher reliance on aerobic respiration (Figure 5A).
Under hypoxia, there was also a noticeable reduction in the expression of the genes in the TCA cycles and other mitochondria genes essential for aerobic respiration, consistent with previous studies [42],[43]. Lactic acidosis, by contrast, has no significant effect on their expression levels. The in vivo relevance of these changes in energy production by hypoxia and lactic acidosis was tested by examining survival predictions based on signatures of these two small sets of genes representing glycolysis and the TCA cycle. When the hypoxia-induced changes of glycolysis are projected into the same breast cancer expression studies, we found that high glycolysis pathway activities have prognostic significance similar to the significance of the whole hypoxia gene signatures (Figure 5G). On the other hand, the tumors with strong hypoxia-induced TCA pathways have entirely opposite effects – they have significantly better clinical outcomes (Figure S8A). Further, tumors with the high correlation between the glycolysis genes with lactic acidosis have better survival (Figure 5H), similar to the overall lactic acidosis gene signatures. Among the glycolysis genes, the lactic acidosis response correlated negatively with the hypoxia response in all four expression studies (Figure 5I). The lactic acidosis-induced changes in TCA cycle genes, on the other hand, have no predictive value for clinical outcomes (Figure S8B).
To further explore the signal transduction pathways of lactic acidosis through genomic analysis, we compared the lactic acidosis gene signatures with the database of “connectivity map” [44] composed of the gene expression patterns elicited by 453 different perturbations caused by 164 distinct small molecules. We assessed the similarity between lactic acidosis response and each reference expression profile in the data set with a nonparametric, rank-based pattern-matching strategy based on the Kolmogorov-Smirnov statistic [44]. From this analysis, the top perturbations positively correlated with the lactic acidosis gene signatures were different treatments of Wortmannin and LY-294002, two known inhibitors of phosphoinositide 3-kinases (PI3 kinase) (Figure 6A, detailed in Table S10). PI3 kinase phosphorylates phosphoinositides PtdIns(3,4)P2 (or “PIP2”) to PtdIns(3,4,5)P3 (or “PIP3”) molecule, which in turn recruit Akt to the cell membrane and trigger Akt phosphorylation and activation. Akt activation, in turn, initiates a cascade of cellular events from glucose uptake, energy utilization, cell growth and proliferation to survival and motility, that drive oncogenesis and tumour progression [45]. This association between acidosis and PI3K inhibition is consistent with the observations seen previously in muscle cells [46].
Gene signatures representing different pathways can be evaluated with lactic acidosis response signature in the gene expression data sets of breast cancer to further elucidate the molecular mechanisms of lactic acidosis [20],[21]. We evaluated this using several gene signatures reflecting oncogenic pathway deregulation induced by genetic manipulations [20]. In the three expression studies of breast cancers [35],[36],[47], we found a consistent inverse relationship (p<0.0001) between the acidosis and Akt pathway signatures – tumors with strong acidosis response tend to manifest low Akt pathway activities while tumors with low acidosis response manifest high Akt pathway activities (Figure 6B and Table S11). This inverse relationship was further noted in an independent mouse model of prostate neoplasia with overexpression of constitutively activated Akt [48] – the acidosis response is high in the normal prostate and low in the prostate of the Akt transgenic mouse which exhibit sign of prostate cancers (Figure 6C).
This inverse correlation between the lactic acidosis and Akt pathway activity in the tumor expression data lead us to hypothesize that lactic acidosis can inhibit the Akt pathway in the tumor cells. This possibility is also consistent with the correlation between lactic acidosis response and PI3K inhibition noted in the connectivity map analysis. Since this observed pathway activity seen in the gene expression may be caused by the corresponding change in the Akt enzymatic activity, we formally tested the effect of lactic acidosis on activation of Akt enzymatic activity in prostate cancer cell line DU145 during serum exposure. Growth of DU145 cells in a serum-free condition resulted in the inhibition of the Akt enzymatic activity, as seen by the absence of phosphorylation of Ser473 (Figure 6D). Upon exposure to serum, the Akt enzymatic activity became activated in 30 minutes, as indicated by the phosphorylation of Ser473 [49] (Figure 6D). However, this serum-induced Akt activation was abolished when the cells were simultaneously exposed to lactic acidosis during serum exposure (Figure 6D). Even after lactic acidosis exposure removed once media was changed to neutral pH during serum exposure, this Akt inhibition still persisted (Figure 6D). This suggests that Akt enzymatic and inferred pathway activity can be abolished by prior exposure to lactic acidosis.
Given the important role of the PI3K/Akt pathway in the tumor glycolytic phenotypes [50], we explored whether lactic acidosis, with its ability to inhibit Akt pathways, can inhibit glycolysis phenotypes of cancer cells. This possibility is further supported by the repression of gene expression in the glycolytic pathways and increased reliance on aerobic respiration in lactic acidosis. We examined the glucose consumption and lactate production in two cancer cell lines – WiDr (colon cancer cell) and SiHa (cervical cancer cell). When grown in a control environment (condition 1) with ambient air, colon cancer cell WiDr exhibit features of aerobic glycolysis with modest glucose consumption and lactate production (Figure 6E). While 25 mM sodium lactate (condition 3) alone did not affect the glucose consumption significantly, all the remaining conditions (including lactic acidosis (condition 2), acidosis at pH 6.7, pH 6.5 and pH 6.0 (condition 4, 5, 6, created by HCl) significantly decreased glucose consumption. The acidosis-induced decrease in glucose consumption is also accompanied by a corresponding decrease in lactate production. This shows that lactic acidosis and acidosis lead to a significant reduction in glycolysis. Both glucose consumption and lactate production were increased under hypoxic conditions (0.5% O2). This hypoxia-induced increase was also dramatically reduced when cells were placed under lactic acidosis and acidosis conditions (conditions 2, 4, 5, 6) but not lactosis. Similar effects in decreasing glucose consumption and lactate production were also seen for SiHa (Figure 6F) and mouse MEF cells. This inhibition of glycolytic phenotypes by lactic acidosis can also explain the increased reliance on aerobic respiration as well as the ability of lactic acidosis response in identifying breast cancers with higher level of aerobic respiration.
Taken together, we propose a model (Figure 6G) in which lactic acidosis favors the utilization of aerobic respiration as the mode of energy production by inhibiting the glycolysis pathways. This is likely due to both the repression of expression of glycolysis genes and the inhibition of Akt enzymatic and pathway activities. This switch to aerobic respiration in vitro may also explain the ability of lactic acidosis gene signatures to identify breast cancers enriched in molecular pathways of aerobic respiration/mitochondria in vivo and more favorable clinical outcomes. Thus, the lactic acidosis gene signatures allow us to identify tumors with distinct metabolic profiles and clinical phenotypes. The reciprocal exchange of in vitro and in vivo global gene expression information via the common language of microarrays greatly enhanced our understanding of how individual microenvironmental stresses lead to relevant clinical phenotypes.
Lactic acidosis and hypoxia are two well recognized features in human cancers. Although tumor lactic acidosis is often thought to co-exist with hypoxia, relatively little is known about its cellular response, relationship with hypoxia or its role in tumor progression. Our current study presents, to our best knowledge, the first genomic analysis of lactic acidosis activities in human breast cancers in vivo. These analyses reveal that tumors exhibiting strong lactic acidosis and acidosis responses are associated with favorable clinical outcomes – in direct contrast to the poor clinical outcome associated with strong hypoxia response. Through various genomic analyses, this association with favorable outcomes is likely to be explained by the ability of lactic acidosis and acidosis to direct energy utilization toward aerobic respiration through the inhibition of glycolysis. Lactic acidosis mediates this effect both by inhibiting gene expression of glycolytic pathways and repressing Akt activation. In contrast to hypoxia, lactic acidosis represses the tumor “glycolytic” phenotype with the reduction of both glucose consumption and lactate production in tumor cells. Since tumor glycolysis is a crucial component of the malignant phenotypes and confers a significant proliferative advantage during somatic evolution [4], this “anti-Warburg” effect by lactic acidosis is likely to contribute to favorable clinical outcomes seen in tumors with strong lactic acidosis programs. The inhibition of Akt/glycolysis may also hamper the adaptive shift to anaerobic glycolysis under hypoxia and renders the cell vulnerable to energy depletion and cell death in hypoxic environments [13]. For example, lactic acidosis represses the LDH-A induction by hypoxia and LDH-A inhibition by RNAi has been previously shown to lead to poor tumor survival and diminished tumorgenicity [38].
A number of studies have now described the power in utilizing large scale gene expression data to develop signatures representing important biological states – in this context, the signature becomes a surrogate phenotype that can be used to explore the biological relevance in the diverse space of in vitro and in vivo systems, including human tumors [20],[51]. In this study, we have taken this approach further to bring signatures together to develop a mechanistic understanding of a clinically important biological process – lactic acidosis response. By investigating its relationship with various known molecular processes in the space of gene expression, we uncovered a positive association with a different mode of metabolic status (aerobic respiration and fatty acid/amino acid metabolism) and an inverse relationship with the PI3K/Akt pathways. These findings reveal the previously undefined complexity of lactic acidosis response program. These in vitro observations, in turn, help to explain the favorable clinical outcomes associated with strong lactic acidosis response in vivo. This use of gene expression data as common phenotypes to facilitate the reciprocal exchange of information between experimental perturbations in vitro and clinical phenotypes in vivo has greatly enhanced our understanding of the lactic acidosis response. We expect this approach is generally applicable and will broadly enhance our understanding in other biological processes. This unexpected result demonstrates the importance of analyzing individual microenvironmental factors separately to determine their respective contributions to tumor phenotypes.
One of the principal metabolic properties of cancer cells is their preferential use of glycolysis for energy generation even in the presence of oxygen (aerobic glycolysis). Warburg has speculated that this is caused by defective or functionally impaired mitochondria. Recent studies suggest that mutations affecting mitochondrial DNA or enzymes of the TCA cycle might contribute to tumor formation, tumor progression and the Warburg effect. Two enzymes in the TCA cycle enzymes, SDH and FH, are found to be tumor suppressor genes [52]. Given the importance of the choice of glycolysis vs. aerobic respiration for energy production in determining the tumor phenotypes and clinical outcomes of cancer patients, the regulatory circuits affecting for this choice have become subjects of intense investigation. For example, hypoxia triggers coordinated changes toward glycolysis through both the induced expression of genes encoding glucose transporters and glycolytic enzymes and repression of mitochondrial function [42],[43]. Hypoxia also induces the expression of PDK1. This phosphorylates and inactivates pyruvate dehdrogenase (PDH), which converts pyruvate to acetyl-CoA. In addition, Akt activation also favors glycolysis since it increases glucose transport and makes cancer cells dependent on glucose for their survival [50]. The cellular rate of glycolysis is also regulated on many levels, such as the availability of oxygen/substrate, NAD/NADH, AMP/ATP ratio, enzyme modification and allosteric inhibition. Other regulators, in contrast, can direct the cells toward aerobic respiration and suppress cancer phenotypes. For example, the tumor suppressor gene p53 can redirect toward aerobic respiration by both inducing genes required for aerobic respiration [39] and repressing genes inhibiting glycolysis [40]. Since the accumulation of lactic acidosis comes from the excess of product from the glycolysis, the inhibition of glycolysis by lactic acidosis can be thought of as a negative feedback. Extracellular acidity is determined both by the abundance of different acidic substances (e.g. lactic acid, CO2) [53] and the buffering capacity of the extracellular fluid. This negative feedback mechanism may allow cells to avoid acidity-induced cell death by adjusting the rate of glycolysis based on extracellular pH. Given the association of lactic acidosis pathway activity with favorable clinical outcomes, understanding the mechanism by which lactic acidosis inhibits Akt and glycolysis may lead to novel therapeutic strategies to modify tumor behavior. The lactic acidosis gene signature can also be used to identify cancers preferentially using aerobic respiration and therefore likely to respond better to chemotherapeutics targeting mitochondrial functions [54].
In addition to generation of ATP, glycolysis is responsible for the generation of acetyl-CoA, which feeds into the TCA cycle for aerobic respiration. Thus, the reduction in glycolysis under lactic acidosis may imply a reduction in the amount of acetyl-CoA derived from the glycolytic process. To maintain cellular energy, cells may increase the use of other energy sources, such as the β-oxidation of fatty acids as source of acetyl-CoA. Since Akt pathway activity is known to suppress β-oxidation and thus energy generation from fatty acids [50],[55], Akt inhibition by lactic acidosis may increase β-oxidation of fatty acid and thereby compensate for the increased demand of acetyl-CoA for energy generation. In support of this change in metabolic state, breast cancers with strong lactic acidosis responses are highly enriched in gene sets of fatty acid degradation and amino acid metabolism (GSEA analyses). There is increased expression of PPAR-alpha and several other genes mediating fatty acid oxidation in the lactic acidosis response. Together, these in vivo and in vitro observations indicate that lactic acidosis induces extensive and coordinated changes in the mode of metabolism and energy utilization, which in turn may account for their effects on the clinical phenotypes of breast cancer.
These data suggest that the lactic acidosis-induced gene expression program may have a direct causal role in impacting tumor biology to affect clinical outcomes. It will be important to understand what specific lactic acidosis-driven biological processes underlie the phenotypic differences between groups of tumors separated by lactic acidosis pathway activity. The mechanisms underlying the variation in lactic acidosis responses in breast are still unknown. There are three reasonable possibilities; variations in the lactic acidosis-response program could reflect: 1) actual variations in lactate and/or acidity in the tumors 2) cell-type-specific variations in the magnitude of, or threshold for, the response to bona fide lactate and/or acidity in tumors or 3) inappropriate activation of the lactic acidosis response resulting from genetic and/or epigenetic alterations in cancers. Given these possibilities, it is important to note that several previous studies have suggested that high level of tumor lactate is associated with poor clinical outcomes in several solid tumors [56],[57],[58]. This is in contrast to our results based on the analysis of lactic acidosis gene signatures, but is not necessarily contradictory as the two may be decoupled in tumors. It is relevant to note that acidosis was able to impose strong selection pressure and likely contributes to the observed responses seen in a group of tumors [4],[14],[18]. This difference highlights the distinction between the measured physiological parameters (pO2, lactate and acidity) and observed cellular responses (hypoxia and lactic acidosis response genes) in human cancers. For example, the expression of carbonic anhydrase IX (CA9) expression is induced by hypoxia. Therefore, CA9 is frequently used as an “endogenous” marker for low tumor pO2 – bona fide hypoxia [59]. But tumor pO2 and hypoxia response are two distinct measurements which may not exhibit tight correlation in all instances. Although many studies have concluded a low tumor pO2 is associated with CA9 expression, other studies point out the lack of direct correlation between CA9 expression and pO2 [60],[61],[62]. In other studies, high levels of tumor lactate [56],[63] and acidic pH [64] are usually associated with tumor metastasis or poor treatment response in solid tumors. The discrepancy between high lactate in tumors as a predictor of poor prognosis and strong lactic acidosis response as a predictor of good prognosis strongly suggests a likely disentanglement between measured physiological parameters and observed cellular responses. This discrepancy may be due to continuous fluctuations in the physiological parameters, different turnovers in the mRNA/protein, cell type-specific variations in the responses or it may result from genetic and/or epigenetic alterations in cancers unrelated to the physiological parameters. It is therefore important to determine whether these variations in the lactic acidosis response program are associated with actual variations in lactate and acidity in the tumors or result from other causes without significant connections to tumor lactate or acidity [65].
It is widely believed that the intimate relationship between tumor progression, tumor microenvironment and the response of the tumor to that environment, when combined with other genetic and clinical factors, offers promise for improving our understanding of the heterogeneity of the disease. Progress in this direction will require a substantial advance in our – currently limited – ability to dissect the roles played by multiple characteristics of the tumor microenvironment. We aim to develop this approach to further dissect other various microenvironmental factors in cancers, such as glucose starvation, reoxygenation and ATP depletion [1]. The availability of this information will allow researchers to gradually establish an integrative metabolic profile of human cancers. Additionally, diverse genetic and molecular characteristics from clinical data will help further elucidate heterogeneous properties of tumors and lead to targeting treatments with better prognosis [26].
Human mammalian epithelial cells (HMEC) were cultured in MEGM (Cambrex) and growth factors were withdrawn for 24 hrs before being placed under different environmental stresses. DU145 cells were cultured in RPMI1640 with 10% FBS, 1% sodium pyruvate, 1% L-glutamine, 1% Hepes and 1% antibiotics (penicillin, 10000 UI/ml; streptomycin, 10000 UI/ml). WiDr and SiHa cells were cultured in DMEM with 10%FBS. Lactic acidosis conditions were created with the addition of 25 mM lactic acid (Sigma) to pH 6.7. Hypoxia was created by lowering the oxygen level to 2%. Similarly, the lactosis condition was created with the addition of 25 mM sodium lactate, while acidosis conditions were created by titrating media to pH of 6.7 with HCl.
RNAs were extracted by miRVana kits (Ambion) and hybridized to Affymetrix Hu133 plus 2 genechips with standard protocol. All microarray data are available on GEO (GEO accession number GSE9649). Hierarchical clustering with weighted average linkage clustering was performed after indicated data filtering based on spot quality and variations in signal intensity as described [66]. The analyses of microarrays from the lactic acidosis/hypoxia and acidosis/lactosis experiments were performed using a sparse ANOVA modeling framework outlined and implemented in the software package, Bayesian Factor Regression Models (BFRM)[67] with detailed parameter files and normalized RMA data available in the supplementary section. In the context of a designed experiment (having controls and experimental groups) BFRM provides a probability of differential expression for each gene and each experimental group. A signature corresponding to each experimental group was compiled by listing those genes with high probability of differential expression as well as high levels of fold change in expression. Principal components were then used to compute weights for each gene such that the weighted average of expression levels showed a clear ability to distinguish the relevant group from others in the experiment. Expression levels of tumor samples on each of the signatures were calculated according to the weighted average obtained from this principal components analysis and after subtraction of mean expression and doping control correction factors. In order to distinguish differential survival associated with expression level of a particular signature, the patient population was split into high and low expression level groups and Kaplan-Meyer survival curves were computed for each group.
RNAs were reverse-transcribed to cDNAs with SuperScript II reverse transcription kit following the manufacturer's protocol (Invitrogen). cDNAs were then used as the substrate for gene expression level measurements by qPCR with Power SYBRGreen PCR Mix (Applied Biosystems) and primers specific for ErbB3(Forward:CAGGGGTGTAAAGGACCAGA, Reverse:CGCCAGTAGAGAAAAGTGCC), CD55(Forward:AGGTCCCACCAACAGTTCAG, Reverse:AAAATGCTTGGTTGTCCTGG), PLAU(Forward:TGTGAGATCACTGGCTTTGG, Reverse:ACACAGCATTTTGGTGGTGA), SOD2(Forward:TTTGGGGACTTGTAGGGATG, Reverse:AGAAAGCCGAGTGTTTCCCT), Actin-beta(Forward:CTCTTCCAGCCTTCCTTCCT, Reverse:AGCACTGTGTTGGCGTACAG), B2M(Forward:TGCTGTCTCCATGTTTGATGTATCT, Reverse:TCTCTGCTCCCCACCTCTAAGT) respectively following the manufacturer's protocol (Applied Biosystem).
DU145 cells were serum-starved (0.2%FBS) for 24 hrs, followed by 24 hr of continuous incubation in serum starved (0.2%FBS) media (as the control) and media with 25 mM lactic acid. 20%, 10%, and 5% FBS were applied for 30 mins to induce Akt activation. Proteins were extracted with PARIS kit (Ambion) and equal amount of protein samples were loaded to SDS-PAGE gels and blotted with pSer473 Akt antibody (Cell signaling) and other indicated antibodies.
WiDr and SiHa cells were plated in six-well dishes (800,000 cells per well). The next day fresh media of respective conditions, including control, 25 mM lactic acidosis, 25 mM sodium lactate, acidosis of pH 6.7, pH 6.5 and pH 6.0 were applied to cells with the continuous incubation of 48 hrs under either normoxia or hypoxia (0.5% O2). After 48hr incubation, media were collected for glucose (ACCU-CHECK, Roche) and lactate (ARKRAY) measurements and normalized against cell number to obtain the glucose consumption/lactate production per million cells.
SiHa cells were plated at the density of 2×104 cells/ml. On the next day, respective media of control and 25 mM lactic acidosis, as well as media containing drug inhibitors, 2-DG and rotenone (Sigma), would be applied. They will then be incubated under normoxia and hypoxia (1% oxygen) respectively. ATP was measured by ATPlite 1 step luminescence ATP detection assay system kit with the protocol provided by the manufacturer after 48 and 72 hours (Perkin Elmer). To prevent the interference caused by different colors of control versus lactic acidosis media, we replaced culture media with PBS right before the addition of substrate solution.
|
10.1371/journal.pntd.0003549 | In-vitro Activity of Avermectins against Mycobacterium ulcerans | Mycobacterium ulcerans causes Buruli ulcer (BU), a debilitating infection of subcutaneous tissue. There is a WHO-recommended antibiotic treatment requiring an 8-week course of streptomycin and rifampicin. This regime has revolutionized the treatment of BU but there are problems that include reliance on daily streptomycin injections and side effects such as ototoxicity. Trials of all-oral treatments for BU show promise but additional drug combinations that make BU treatment safer and shorter would be welcome. Following on from reports that avermectins have activity against Mycobacterium tuberculosis, we tested the in-vitro efficacy of ivermectin and moxidectin on M. ulcerans. We observed minimum inhibitory concentrations of 4–8 μg/ml and time-kill assays using wild type and bioluminescent M. ulcerans showed a significant dose-dependent reduction in M. ulcerans viability over 8-weeks. A synergistic killing-effect with rifampicin was also observed. Avermectins are well tolerated, widely available and inexpensive. Based on our in vitro findings we suggest that avermectins should be further evaluated for the treatment of BU.
| Neglected tropical diseases such as Buruli ulcer predominantly afflict the poorest populations in the world and reduce quality of life. Buruli ulcer is a necrotising infection that destroys the skin and soft tissue, frequently presenting as nodules or open ulcers. Buruli ulcer is treated with antibiotics and sometimes surgery. Unfortunately the antibiotic treatment can have toxic side effects, such as hearing loss. Also, patients must either be hospitalized or report daily to a treatment centre to get their medicine as the treatment is delivered by injection.
In laboratory experiments we tested the susceptibility of Mycobacterium ulcerans, which causes Buruli ulcer, to avermectins. Avermectins are drugs that are used to treat common parasite and worm infections, such as river blindness. These drugs are inexpensive, have few side effects and are widely available. Our findings show that two avermectins called ivermectin and moxidectin inhibit the growth and also kill Mycobacterium ulcerans strains from both Africa and Australia. If their efficacy and safety also can be proven in animal and human studies, these drugs will provide an inexpensive addition to the current treatment of Buruli ulcer.
| Buruli ulcer (BU) is a neglected tropical disease that presents as skin nodules, plaques or oedematous lesions that can progress to open ulcers [1]. BU is caused by infection with Mycobacterium ulcerans, a mycobacterium that is related to the causative agents of tuberculosis and leprosy [2]. Most BU patients are children under the age of 15 [3]. The mode of transmission of the disease is not well understood. Superstitious beliefs predominate in rural West Africa, resulting in delayed treatment seeking and stigmatization of patients [4,5]. Mortality associated with BU is low nevertheless morbidity is high. Extensive ulcers frequently lead to lifelong physical disability [6,7].
No vaccine is available against Buruli ulcer and management focuses on early case detection and treatment with surgery and antibiotics [8]. Previously, Buruli ulcer was treated with surgical excision only, but since 2004, an 8-week course of rifampicin and streptomycin is the standard treatment [9–11].
In case patients report early with limited lesions, the 8-week course of rifampicin and streptomycin delivers a good quality of life at long-term follow up [12]. However, the median time to heal is still 18 to 30 weeks, depending on the size of the lesion [10].
Patients presenting late to health care facilities have a much poorer prognosis and many suffer from functional limitations due to the disease [6,7].
There is also the issue that daily injections with streptomycin are impractical and can have serious side effects, such as ototoxicity [13]. It has been shown that all-oral treatment with rifampicin and a macrolide or quinolone results in high cure rates [14]. Shorter duration of antibiotic courses and more safe treatment regimes that reduce the time to healing are desirable for Buruli ulcer.
Avermectins are macrolides that are used to treat helminth-infection (such as strongyloidiasis or onchocerciasis) and parasitic infection (scabies) in humans and in animals. These orally administered drugs are well tolerated and available worldwide. Ivermectin is on the essential drugs list of the World Health Organization. A recent report showed that avermectins, including ivermectin, moxidectin and selamectin, inhibit the growth of different M. tuberculosis strains in-vitro at concentrations of 2 to 8 μg/ml [15]. Motivated by these findings, we tested if avermectins also inhibit and kill M. ulcerans.
Two different M. ulcerans clinical isolates were used, JKD8049 isolated from a patient in Victoria, Australia in 2004 and 1117–13, a 2013 clinical isolate from Benin. For time-kill assays (see below), M. ulcerans JKD8049 containing a bioluminescent reporter plasmid pMV306 hsp16+luxG13 [16] was employed. The Mycobacterium marinum ‘M’ strain was also used. Mycobacteria were grown at 30°C in 7H9 Middlebrook broth supplemented with OADC (Becton Dickinson, Sparks, MD, USA). M. ulcerans JKD8049 harbouring pMV306 hsp16+luxG13 was grown in the presence of 25 μg/ml kanamycin.
Bacteria in mid-exponential growth phase were used for MIC testing. They were prepared to 0.5 Macfarlane standard and diluted 1:5 in PBS. A 500 μl volume of this preparation was used to inoculate duplicate BBL™ Mycobacteria Growth Indicator Tubes (MGITs) supplemented with 0.5 ml OADC (Becton Dickinson, Sparks, MD, USA) that contained doubling dilutions of moxidectin, ivermectin or rifampicin (Sigma-Aldrich, St. Louis, MO, USA.). The tubes were incubated at 30°C and assessed daily for fluorescence, with a long-wave UV-A lamp (Wood’s lamp) for 21 days. The tube with the lowest drug-concentration displaying no growth after this period was considered as containing the inhibitory concentration. Solvent only and rifampicin 0.1 μg/ml were used controls.
Ten millilitre aliquots of mid-exponential growth phase M. ulcerans JKD8049 were transferred in duplicate to sterile 25cm2 tissue culture flasks containing 0, 8 and 20 μg/ml ivermectin. The aim was to obtain a M. ulcerans concentration of at least 10^6 CFU/ml in each flask. Each week, 50 μl of culture was sampled from each flask to assess viable bacteria remaining by CFU counting. Ten-fold dilutions of the sub-samples were prepared in PBS and a 3 μl aliquot of each dilution was spot plated in quintuplicate onto Middlebrook 7H10 agar containing 10% OADC. After 8 weeks of incubation at 30°C plates were examined for growth. The growth/no growth scores of the five technical replicates were used to calculate the most probable number estimate of CFU per ml. Data was analyzed using GraphPad Prism v5.0d.
Three to five replicates of 200 μl aliquots of bioluminescent M. ulcerans JKD8049 in mid-exponential growth were transferred into a white 96-well plate and antibiotics were added. The plate was placed into a FLUOstar Omega plate reader (BMG LABTECH GmBH, Ortenberg Germany). Light emission was read every 300 s via the top optic with the gain set at 3600 and plate temperature at 30°C. Before each reading, plates were shaken at 100 rpm for 10 s in double orbital mode. The results were recorded using Omega v3.00 R2 and analyzed using Mars v3.01 R2 and GraphPad Prism v5.0d.
M. ulcerans JKD8049, M. ulcerans 1117–13 and M. marinum were grown in MGIT tubes in the presence of increasing concentrations of the ivermectin and moxidectin. At three weeks, no fluorescence was observed at 8 μg/ml of ivermectin for M. ulcerans JKD8049 and at 4 μg/ml for M. ulcerans 1117–13 (Table 1). Moxidectin inhibited the growth of M. ulcerans JKD8049 at 4 μg/ml. The MIC for M. marinum was 32 μg/ml for ivermectin and above 64 μg/ml for moxidectin (Table 1). These data show that M. ulcerans but not M. marinum is susceptible to avermectins. Growth of all bacteria was observed in the control tubes containing only the solvent and no growth was observed in the tubes containing 0.1 μl/ml rifampicin.
Time-kill assays were then performed to assess if avermectins not only inhibit M. ulcerans growth but also kill the bacterium. Based on the MIC results above, a low and high dose of ivermectin was tested and M. ulcerans JKD8049 was exposed to either 8 μg/ml or 20 μg/ml ivermectin for eight weeks. A dose-dependent killing effect was observed with no CFU detected from week-4 onwards at 20 μg/ml and week-5 onwards for 8 μg/ml (Fig 1).
Bioluminescence is an ATP-dependent process and it is therefore an excellent dynamic reporter of cellular metabolic activity [17]. A time-kill experiment was thus performed using bioluminescent M. ulcerans. Although the time frame of this experiment was quite short (21 hours), continuous monitoring over that period again showed a dose-dependent impact of ivermectin on bacterial viability (Fig 2). The reduction in bioluminescence at 21 hours was higher in ivermectin at all concentrations tested (8, 16 and 32 μg/ml) compared to the rifampicin positive control (Fig 2). Interestingly, the greatest impact on bacterial viability was observed in the presence of a combined dose of 8 μg/ml ivermectin and 0.1 μg/ml rifampicin (Fig 2).
The current treatment regimen of 8-weeks streptomycin and rifampicin for Buruli ulcer is highly effective but also problematic and promising alternative all-oral treatments are sought [13,14]. Here we have shown that ivermectin and moxidectin were able to inhibit the growth of different M. ulcerans strains at 4–8 μg/ml. These findings are comparable to previous research where MICs of 1–8 μg/ml against M. tuberculosis with ivermectin, selamectin, moxidectin or doramectin were reported [15].
M. marinum, despite its close relationship to M. ulcerans, showed low susceptibility to both ivermectin and moxidectin. We speculate that this difference may be explained at least in part by the abundance of intact transporters and efflux systems in M. marinum and a corresponding scarcity of the equivalent systems in M. ulcerans [18,19].
The 8-week time-kill experiment (Fig 1) showed that ivermectin at concentrations of 8 μg/ml and above has a likely bactericidal effect on M. ulcerans. Future experiments will test lower drug concentrations and use higher starting doses of bacteria. The long doubling time of M. ulcerans (>48h) complicates laboratory-testing methods using traditional culture and colony counting approaches. Here we used bioluminescence as a rapid read-out of cell viability and found it a useful technique for examining these slow-growing mycobacteria (Fig 2). Given the MIC findings (Table 1), we were surprised to observe in the bioluminescent time-kill experiments that ivermectin performed better than rifampicin at their given MIC concentrations. Further investigation of this difference is warranted, probably by conducting the experiments over an extended time period and beyond the 21 hours we were able to achieve here. A combination of 0.1 μg/ml rifampicin and 8 μg/ml ivermectin showed a synergistic killing effect. Clinically, rifampicin may however decrease the serum concentration of ivermectin by the induction of P-glycoprotein/ABCB1 in humans [20]. The extent to which this might be the case is not known and requires further testing.
In the context of avermectins to treat M. tuberculosis infections, it has been argued that MICs of 1–8 μg/ml are not achievable in humans [21]. Avermectins are mostly administered in much lower doses and in single-doses for the treatment of helminth infection. A maximum plasma concentration of 54.4 ng/ml was observed in healthy volunteers receiving 150 μg/kg ivermectin [22].
There are few public data on the safety of higher doses or prolonged courses of avermectins in humans and it unclear at what doses and over what time avermectins would have to be given to Buruli ulcer patients. We would argue that obtaining avermectin plasma concentrations at concentrations that approach MICs derived from in vitro experiments might not be needed for clinical efficacy. In our bioluminescent time-kill assay, 8 μg/ml ivermectin was superior to 0.1 μg/ml rifampicin. It may be that the treatment duration for BU can be reduced and that complete eradication of the microbe is not needed. Interference with the M. ulcerans mycolactone toxin synthesis machinery at concentrations that are sub-inhibitory for growth might be sufficient to permit host immunity to then clear the infection.
The avermectins ivermectin and moxidectin inhibited growth of M. ulcerans at 4–8μg/ml and showed dose-dependent killing in culture-based and bioluminescence assays. The avermectins are inexpensive and already widely distributed in West Africa through their use to treat river blindness. Thus, there may be a chance to repurpose a well-tolerated drug for the treatment of mycobacterial infections, bypassing the long and expensive pipeline for discovery of new antimicrobials. We suggest that avermectins should be further investigated for the treatment of M. ulcerans, possibly in combination with other antibiotics, such fluoroquinolones.
|
10.1371/journal.ppat.1001203 | Fcγ Receptor I Alpha Chain (CD64) Expression in Macrophages Is Critical for the Onset of Meningitis by Escherichia coli K1 | Neonatal meningitis due to Escherichia coli K1 is a serious illness with unchanged morbidity and mortality rates for the last few decades. The lack of a comprehensive understanding of the mechanisms involved in the development of meningitis contributes to this poor outcome. Here, we demonstrate that depletion of macrophages in newborn mice renders the animals resistant to E. coli K1 induced meningitis. The entry of E. coli K1 into macrophages requires the interaction of outer membrane protein A (OmpA) of E. coli K1 with the alpha chain of Fcγ receptor I (FcγRIa, CD64) for which IgG opsonization is not necessary. Overexpression of full-length but not C-terminal truncated FcγRIa in COS-1 cells permits E. coli K1 to enter the cells. Moreover, OmpA binding to FcγRIa prevents the recruitment of the γ-chain and induces a different pattern of tyrosine phosphorylation of macrophage proteins compared to IgG2a induced phosphorylation. Of note, FcγRIa−/− mice are resistant to E. coli infection due to accelerated clearance of bacteria from circulation, which in turn was the result of increased expression of CR3 on macrophages. Reintroduction of human FcγRIa in mouse FcγRIa−/− macrophages in vitro increased bacterial survival by suppressing the expression of CR3. Adoptive transfer of wild type macrophages into FcγRIa−/− mice restored susceptibility to E. coli infection. Together, these results show that the interaction of FcγRI alpha chain with OmpA plays a key role in the development of neonatal meningitis by E. coli K1.
| Escherichia coli K1 is the most common cause of meningitis in premature infants; the mortality rate of this disease ranges from 5% to 30%. A better understanding of the pathogenesis of E. coli K1 meningitis is needed to develop new preventative strategies. We have shown that outer membrane protein A (OmpA) of E. coli K1, independent of antibody opsonization, is critical for bacterial entrance and survival within macrophages. Using a newborn mouse model, we found that depletion of macrophages renders the animals resistant to E. coli K1 induced meningitis. OmpA binds to α-chain of Fcγ-receptor I (FcγRIa) in macrophages, but does not induce expected gamma chain association and signaling. FcγRIa knockout mice are resistant to E. coli K1 infection because their macrophages express more CR3 and are thus able to kill bacteria with greater efficiency, preventing the development of high-grade bacteremia, a pre-requisite for the onset of meningitis. These novel observations demonstrate that inhibiting OmpA binding to FcγRIa is a promising therapeutic target for treatment or prevention of neonatal meningitis.
| Professional phagocytes, including neutrophils and macrophages (MØ) express a specific set of phagocytic receptors that recognize, bind to and mediate internalization of microbial pathogens [1], [2], [3]. Although MØ receptor-mediated phagocytosis generally leads to the destruction of the pathogen, certain receptor-ligand interactions allow for a permissive environment in which the pathogen can thrive and even proliferate. MØ provide a barrier that pathogens must overcome to adhere to and penetrate into tissues. Nonetheless, diverse strategies are used by different bacterial pathogens to subvert phagocytes. Escherichia coli K1 causes meningitis in neonates, which remains a significant problem for the last few decades with case fatality rates ranging from 5 to 40% of infected neonates [4], [5], [6], [7]. Despite treatment with advanced antibiotics, up to 30% of survivors exhibit neurological sequelae such as hearing impairment, mental retardation, and hydrocephalus. Furthermore, due to the emergence of antibiotic resistant strains, mortality rates may significantly increase in future [8]. The crossing of the mucosal epithelium and the invasion of small subepithelial blood vessels by E. coli K1 represent critical early steps in the pathogenesis of meningitis. During initial colonization, E. coli K1 encounters several host defense mechanisms such as complement, neutrophils, and MØ on its path to the blood-brain barrier (BBB). However, very little is known about the mechanisms by which E. coli K1 finds a niche to avoid these host defenses. Our previous studies demonstrated that E. coli K1 evades complement attack by binding to the complement pathway regulator C4bp via outer membrane protein A (OmpA), which subsequently cleaves C3b and C4b complement proteins [9], [10]. In addition, lack of significant quantities of C9, a terminal complement component necessary for the formation of the membrane attack complex, in neonatal population gives an additional opportunity for E. coli K1 to survive in the blood [10]. However, our studies have shown that an inoculum of >103 CFU/ml of E. coli K1 is required to resist serum bactericidal activity [11], indicating that the bacteria must take a refuge in certain cells to survive and multiply during the initial stages of infection, when fewer bacteria are present in the blood.
Despite the importance of MØ in innate and adaptive immunity, the interaction of E. coli K1 with these cells is poorly defined. MØ phagocytose a broad range of pathogens by recognizing pathogen-associated molecular pattern (LPS and peptidoglycans) via pattern recognition receptors (PRR), which include TLRs, the mannose receptor and the scavenger receptor [12], [13]. Opsonin-dependent phagocytosis involves complement receptors and antibody-dependent phagocytosis requires Fcγ receptors. Studies from our lab have shown that E. coli K1 enters and multiplies in both human and murine MØ, either in the presence or absence of opsonization. OmpA expression is critical for these processes [14]. Therefore, it is important to determine whether E. coli OmpA interacts with any cell surface proteins of MØ for entry. Numerous studies have shown that the expression of FcγRI is increased during septicemia and meningitis caused by a variety of pathogens [15], [16], [17], although its importance in any of these infections has not been explored. Fcγ receptors (FcγR) recognize the Fc region of IgG and play a pivotal role in linking the cellular and humoral arms of the immune response. FcγR comprises a multigene family divided into 3 classes (FcγRI, II and III), which are defined by their affinity for IgG [18], [19], [20], [21], [22], [23]. FcγRI is a transmembrane receptor, which binds IgG with high affinity and induces the association of the γ-chain for signal transduction and triggering of effector responses such as MØ phagocytosis [23]. The ligation of FcγRI with IgG also mediates antibody-dependant cellular cytotoxicity induced transcription of cytokine genes and release of inflammatory mediators [24]. The cytoplasmic domain of the γ-chain contains an immunoreceptor tyrosine activation motif (ITAM), which is necessary for the signaling cascade of FcγRs. The cytoplasmic domain of FcγRI has been shown to modulate receptor function, although it does not contain any recognized signaling motif [25], [26]. In this study, we examined the role of MØ and FcγRI α-chain (FcγRIa) in the pathogenesis of neonatal E. coli K1 meningitis by depleting MØ from newborn mice and utilizing a FcγRIa−/− knockout mouse model. Our studies provide evidence of a role for a novel interaction between FcγRIa and OmpA in the onset of meningitis due to E. coli K1.
Our previous studies have shown that OmpA+ E. coli enters and survives in human and mouse MØ [14]. To determine the role of MØ in the pathogenesis of E. coli K1 induced meningitis, MØ were depleted in newborn mice by the administration of carrageenan [27], [28]. MØ readily ingest carrageenan in contrast to lymphocytes, which are not actively phagocytic and lack a well-developed lysosomal complex. Due to its unique secondary and tertiary structure, carrageenan is resistant to biochemical degradation by lysosomal glycosidases. Carrageenan containing phagolysosomes eventually rupture due to osmotic swelling. The consequent release of hydrolytic enzymes into the cytosol causes irreversible damage and eventual lysis of MØ [29]. Following three days of carrageenan administration starting at day 1 after birth, the animals showed >95% depletion of MØ from livers and spleens, as shown by flow cytometry after staining with F4/80 antibody (5.33%±0.4% before and 0.17%±0.1% after carrageenan treatment) (Figure 1A). However, treatment with carrageenan did not affect B cells (39.81%±0.7% before and 40.19%±0.9% after carrageenan treatment), CD4+ T cells (17.56%±0.5% before and 18.02%±0.6% after carrageenan treatment), CD8+ T cells (2.11%±0.4% before and 2.53%±0.3% after carrageenan treatment), DCs (5.67%±1.2% before and 6.09%±0.9% after carrageenan treatment), or PMNs (3.98%±1.2% before and 4.13%±1.4% after carrageenan treatment) in spleens of MØ-depleted mice compared with untreated mice (Figure S1). The MØ-depleted mice were then infected with 103 CFU of E. coli K1 by intranasal instillation and examined for progression of the disease as previously described [28]. Control animals (n = 15 for each group) developed bacteremia at 6 h post-infection, which was increased to 5.5 log10 CFU per ml of blood by 48 h (Figure 1B). In contrast, the MØ-depleted mice, despite having a similar number of bacteria in the blood at 6 h, cleared these bacteria from the circulation by 48 h post-infection. In agreement with the bacteremia levels, >90% of control mice developed meningitis at 72 h after infection, whereas none of the MØ-depleted animals showed signs of meningitis and all survived beyond 7 days (Figure 1C). Determination of serum cytokine levels at various times post-infection revealed that control animals produced an initial burst of IL-10, which peaked at 12 h, and then declined to basal levels by 48 h (Figure 1D). In contrast, the pro-inflammatory cytokines, TNF-α, IFN-γ, IL-1β, IL-6 and IL-12p70 only became detectable in the blood at 12 h post-infection and peaked by 72 h (Figure 1D and Figure S2). Of note, although the MØ-depleted mice had early production of pro-inflammatory cytokines, their levels were significantly lower than those in the control mice. In these mice, IL-10 levels progressively rose during the initial stages of infection and peaked at 72 h at which time the bacteria were completely cleared from the circulation (Figure 1D).
Histopathological examination of control mice infected with E. coli K1 revealed marked infiltration of PMNs in the leptomeningeal and ventricular spaces (Figure 1E). The hippocampus was also inflammed and there was apoptosis of neurons, as indicated by pkynotic nuclei in Ammon's horn. Acute hemorrhage and inflammation was observed, most prominently in the white matter of the brain. The cortex and molecular layer had increased cellularity due to inflammatory exudates. The MØ-depleted mice, however, did not reveal such pathological changes. Blood brain barrier (BBB) leakage is the hallmark of neonatal meningitis. Therefore we used the Evans blue extravasation method to quantify BBB leakage in both the control and MØ-depleted mice [28]. The dye was injected intraperitoneally at 68 h post-infection and after four hours, the brains were removed and Evans blue concentration determined. A marked increase in the permeability of the BBB was observed in infected WT animals, which was significantly reduced in MØ-depleted mice, (p<0.001 by student's t test) (Figure 1F). Furthermore, the number of E. coli K1 entering the brain was approximately 6.0 log10 CFU in control animals, whereas the brains of the MØ-depleted animals contained very few bacteria (Figure 1G). These results demonstrate that MØ may be important for E. coli K1 to reach a required level of bacteremia, which is critical for the establishment of neonatal meningitis.
Our previous studies have shown that OmpA+ E. coli binds and enters MØ in vitro irrespective of opsonization status of the bacteria [14]. OmpA− E. coli, although entered in lower numbers but failed to survive inside MØ. This indicates that OmpA mediated entry into MØ enables OmpA+ E. coli K1 to resist the normal antimicrobial mechanisms of MØ. Therefore, to understand the nature of the macrophage surface structures that interact with E. coli K1, biotin-labeled cell surface proteins of THP-1 cells differentiated into MØ (THP-M) and RAW 264.7 cells were incubated with OmpA+ E. coli, OmpA− E. coli or a laboratory E. coli HB101. Bound proteins were then released and analyzed by western blotting with streptavidin peroxidase. A small number of proteins bound to all the bacteria from both the cells. However, OmpA+ E. coli prominently bound to the 110 and 70 kDa proteins from both THP-M and RAW 264.7 cells, whereas OmpA− E coli bound only to the 110 kDa protein (Figure 2A). Although some proteins bound to HB101 were of similar molecular mass to those bound to OmpA+/OmpA− E. coli, other proteins showed different binding patterns. Based on their molecular masses, we speculated that the proteins binding to E. coli K1 could be Toll like receptor-4 (110 kDa) and FcγRIa (CD64, 72 kDa). Since OmpA+ E. coli specifically bound to the 70 kDa protein in contrast to OmpA− E. coli, the blots were reprobed with an anti-FcγRI antibody, which reacted with the 70 kDa protein, suggesting that OmpA+ E. coli binds to FcγRIa. Of note, treating the bacteria with 40% pooled human serum did not alter the binding, indicating that opsonization with complement and/or with non-specific antibody did not alter bacterial interaction with macrophage surface proteins.
Next, we used blocking antibodies to determine the contribution of OmpA-FcγRIa interaction in E. coli entry into MØ. OmpA+ E. coli was incubated with Fab fragments of anti-OmpA antibody (polyclonal) prior to addition to MØ. In other experiments, the RAW 264.7 cells were pre-treated with antibodies to FcγRI, CR3, TLR2, TLR4 or the mannose receptor prior to addition of OmpA+ E. coli. Isotype matched antibodies or anti-S-fimbria antibodies were used as controls. Both anti-OmpA and anti-FcγRI antibodies reduced the number of bound and intracellular E. coli K1 by ∼80%, whereas other antibodies showed no significant inhibition (Figure 2B). To verify that the anti-FcγRI antibody actually inhibited FcγRI–mediated phagocytosis, the effect of this antibody on the entry of zymosan coated with fluorescent-labeled IgG2a that occurs via FcγRI was also determined. The internalized zymosan particles were counted per 100 cells after quenching the external fluorescence by Trypan Blue [30]. As predicted, anti-FcγRI antibodies significantly inhibited the entry of opsonized zymosan (Figure 2C).
MØ pretreated with the anti-FcγRI antibody were also infected with Group B streptococcus (GBS) pre-treated with C8-deficient serum (for deposition of C3 and to avoid bacterial killing by serum), which is known to enter MØ through the CR3 receptor [31], [32], [33]. The internalization of GBS, however, was not affected by pretreatment with anti-FcγRI antibody, suggesting that it did not interfere with CR3 receptor function in MØ (Figure 2D). However, as expected, anti-CR3 antibodies significantly blocked the binding and entry of GBS into RAW 264.7 cells. To further confirm the role of OmpA interaction with MØ in E. coli entry into MØ, OmpA was purified from OmpA+ E. coli and reconstituted into liposomes as previously described [34], which were used to pre-treat RAW 264.7 cells prior to adding the bacteria (Figure 2E). The liposomes containing OmpA blocked both binding and intracellular survival of E. coli K1 by approximately 50%, whereas liposomes containing outer membrane proteins from OmpA− E. coli did not show such inhibition. Increasing concentrations of OmpA liposomes showed no further increase in the inhibition, indicating that the structure of OmpA in liposomes may not be optimal to that of OmpA on E. coli K1 to bind to FcγRIa.
The fate of OmpA+ E. coli after phagocytosis by RAW 264.7 cells was examined by immunocytochemistry after differential staining. Extracellular bacteria were stained with FITC labeled secondary antibody (green) and the intracellular bacteria were stained with a TRITC labeled secondary antibody (red) after incubation with primary anti-S-fimbria antibody. As shown in Figure 2F, a number of OmpA+ E. coli bound to RAW 264.7 cells, whereas very few OmpA− E. coli bound at 30 min post-infection. Analysis of intracellular bacteria over time revealed that OmpA+ E. coli multiplied, whereas OmpA− E. coli were degraded inside the cells. Collectively, these studies suggest that the OmpA of E. coli K1 interacts with regions of FcγRIa similar to those involved in the binding of Fc and that this interaction enables the organism to enter MØ. In addition, the data suggest that other receptors that recognize pathogen-associated molecules may not play a significant role in MØ binding and entry of E. coli K1. However, entry through other receptors in the absence of OmpA-FcγRIa interaction renders the bacteria susceptible to macrophage killing.
To confirm the role of FcγRIa in OmpA+ E. coli entry of MØ, short hairpin RNA (shRNA) sequences for murine FcγRIa and CR3 in pGeneClip Neomycin vectors were used to transfect RAW 264.7 cells. Suppression of FcγRIa and CR3 gene transcription and expression was verified by RT-PCR and flow cytometry, respectively. The respective shRNA suppressed the transcription of FcγRIa and CR3 considerably, but had no effect on GAPDH, TLR2 or TLR4 mRNA transcript levels (Figure 3A). On par with changes in transcription levels, the surface expression of FcγRIa and CR3 was significantly reduced, while TLR2 and TLR4 expression was unaltered (Figure 3B). There was >90% reduction in the OmpA+ E. coli phagocytosed by FcγRIa-shRNA/RAW cells compared to control or CR3-shRNA/RAW cells (p<0.001 by two-tailed t test) (Figure 3C). This reduction was due to inefficient binding of E. coli K1 to these cells, as less than 30% of bacteria were bound by the FcγRIa-shRNA/RAW cells compared to non-transfected or control-shRNA transfected cells. In contrast, both binding and intracellular survival of GBS were not affected in FcγRIa-shRNA/RAW cells, whereas CR3-shRNA transfection caused significant reduction in both of these processes (Figure 3D). Immunocytochemistry of E. coli K1 infected FcγRIa-shRNA/RAW cells revealed that very few cells ingested bacteria and were killed within 2 h post-infection (Figure 3E, fragmented bacteria). However, E. coli K1 entered and replicated in CR3-shRNA/RAW cells similar to control RAW cells. Comparable results were also obtained with THP-M cells transfected with shRNA specific to human FcγRI (data not shown). To further confirm that lack of FcγRIa expression rendered bacteria susceptible to macrophage killing, FcγRIa-shRNA/RAW cells infected with E. coli K1 were examined by transmission electron microscopy. Although few numbers of FcγRIa-shRNA/RAW cells engulfed E. coli K1, several of them were either degraded or in the process of degradation by 1 h post-infection and were completely killed by 8 h post-infection (Figure 3F). In contrast, CR3-shRNA/RAW cells showed intact bacteria in endosomes undergoing significant multiplication by 8 h post-infection. Taken together these results demonstrate that OmpA-FcγRIa interaction is critical for E. coli K1 to bind to, enter and survive in MØ.
The activation of FcγRI in phagocytic cells by the binding of the Fc region of IgG requires the association of FcγRIa with the IgG γ-chain [19]. To examine whether γ-chain association with FcγRI is also necessary for E. coli K1 invasion, COS-1 cells were transfected with pcDNA3 plasmids containing Myc-tagged human FcγRIa (Myc-hFcγRIa), C-terminal truncated Myc-hFcγRIa which lacks the cytoplasmic tail (CT) or Myc-hFcγRII. Expression of these proteins was verified by Western blotting using the anti-Myc antibodies (Figure 4A) and flow cytometry (Figure 4B). OmpA+ E. coli binding to, and invasion of, hFcγRIa+/COS-1 cells was significantly greater compared to that of mock-transfected cells (Figure 4C). OmpA− E. coli showed very negligible binding to, and invasion into, both FcγRIa transfected and mock-transfected COS-1 cells (data not shown). The invasion of E. coli K1 into Myc-hFcγRIa-CT/COS-1 cells was significantly reduced, although the binding of bacteria to these cells was decreased by only 30% compared to Myc-hFcγRIa+/COS-1 cells. In contrast, overexpression of FcγRII did not increase E. coli binding to, or invasion of, COS-1 cells. These data suggest that FcγRIa acts as receptor for OmpA mediated entry of E. coli K1 into COS-1 cells and that the C-terminal portion is required for this invasion.
Next, to examine whether FcγRIa interacts with OmpA, recombinant hFcγRIa (rhFcγRIa) was purified by Myc-affinity column chromatography from COS-1 cells and incubated with OmpA+ or OmpA− E. coli. The bound proteins were released and subjected to Western blotting with antibodies to Myc or FcγRI. The purified rhFcγRIa bound to OmpA+ E. coli but not to OmpA− E. coli, whereas BSA, used as a control, did not interact with the bacteria (Figure 4D). rhFcγRIa used to pre-treat bacteria prior to adding them to COS-1 monolayers in the invasion assays resulted in much more significant inhibition of E. coli K1 binding to, and entry into the cells in a dose dependent manner when compared to the BSA control (Figure 4E). These results suggest that the alpha chain of FcγRI is sufficient for E. coli K1 to bind to, and invade, COS-1 cells.
One important question to address in these studies is how OmpA of E. coli K1 binds to FcγRIa at the same region as the Fc-portion of IgG in the context of whole blood. Generally, specific or even non-specific IgG in circulation binds invading bacteria and thereby presents the pathogen to FcγR receptors on MØ. Therefore, it is possible that OmpA+ E. coli may be displacing IgG for binding to FcγRI. We tested this hypothesis by performing two different competitive binding experiments. First, OmpA− E. coli were coated with anti-S-fimbria antibody and added to FcγRIa+/COS-1 cells treated with cytochalasin D to prevent internalization. The cells were washed and then various quantities of OmpA+ E. coli were added and incubated for 10 min. After washing the monolayers, the number of OmpA− E. coli that remained bound to COS-1 cells were determined by plating on agar containing tetracycline (OmpA+ E. coli is sensitive to tetracycline). As shown in Figure 5A, IgG2a opsonized OmpA− E. coli bound COS-1 cells in significantly greater numbers compared to unopsonized bacteria and progressively more bacteria were released from the cells as more OmpA+ E. coli were added to the wells. In contrast, OmpA− E. coli could not displace bound OmpA− E. coli. In separate experiments, peritoneal MØ were incubated with FITC-IgG2a (1 µg) for 1 h in the presence of cytochalasin D, washed and then various quantities of OmpA+ E. coli or OmpA− E. coli were added. The cells were incubated for 10 min, washed and the amount of FITC-IgG that remained bound to the MØ was determined by flow cytometry. As shown in Figure 5B, the amount of FITC-IgG2a bound to peritoneal MØ was decreased when OmpA+ E. coli were added, whereas addition of OmpA− E. coli had no effect. These results indicate that the interaction of E. coli K1 with FcγRIa via OmpA can displace bound IgG2a.
Binding to the γ-chain of FcγRIa is crucial for inducing the anti-microbial activity of MØ [20]. Since OmpA binding to FcγRIa prevented the killing of the bacteria, we hypothesize that E. coli K1 interaction with FcγRIa avoids the association of the γ-chain. Consistent with this assumption, OmpA+ E. coli interaction with MØ in the presence or absence of IgG2a opsonization induced far less γ-chain association with FcγRIa in comparison to OmpA− E. coli, as shown by immunoprecipitation studies (Figure 5C). Similarly, OmpA+ E. coli induced a distinct tyrosine phosphorylation pattern of macrophage cytoplasmic proteins compared to OmpA− E. coli opsonized with IgG2a (Figure 5D). Taken together, these studies suggest that the interaction of E. coli K1 with FcγRIa can displace the bound IgG2a, which is mediated by OmpA. They also indicate that OmpA-FcγRI interaction induces novel signaling patterns, which may abrogate the normal antimicrobial response of these cells.
To confirm the role of FcγRIa in the pathogenesis of E. coli K1 meningitis, FcγRIa−/− mice were used for infection studies. MØ isolated from FcγRIa−/− mice did not express FcγRIa but had unchanged expression of other FcγRs, TLRs, mannose receptor and CR3 were unchanged compared to normal littermates (data not shown). The newborn animals were intranasally infected with E. coli K1 and examined for disease progression. Of note, the FcγRIa−/− animals did not develop bacteremia even at a 100 fold higher infectious dose, even though E. coli K1 entered the circulation within two hours of infection (Figure S3A). In contrast, wild type (WT) animals showed 7.0 log10 CFU of E. coli K1 in blood at 72 h post-infection (Figure 6A). The FcγRIa−/− mice did not develop meningitis even when infected with a 100-fold greater inoculum (data not shown). These mice did not show any signs of meningitis even after 7 days of infection, whereas 90% of WT mice showed positive CSF cultures by 72 h post-infection (Figure 6B). Cytokine analysis in the sera of these animals demonstrated that infected WT animals generated significant amounts of TNF-α, IL-1β, IL-6, IFN-γ and IL-12, but FcγRIa−/− mice did not (Figure 6C and Figure S3B–D). On the other hand, IL-10 production peaked at 24 h post-infection and subsequently returned to basal levels in WT mice, whereas FcγRIa−/− showed increased IL-10 production at 72 h post infection (Figure 6D). We next examined blood-brain barrier leakage in FcγRIa −/− mice. Infection with E. coli K1 caused no leakage in FcγRIa−/− mice, whereas WT animals had significant leakage of Evans blue dye (Figure 6E). Furthermore, no bacterial colonies were detected in the brains of FcγRIa−/− mice, while WT animals had a high bacterial load (Figure 6F). Similarly, the pathology of the brains from FcγRIa−/− mice revealed no infiltration of neutrophils, neuronal damage or gliosis, which are the characteristic pathological features of E. coli K1 meningitis observed in WT bacteria infected mice (Figure 6G). In contrast, infection of FcγRIa−/− mice with GBS resulted in significant bacteremia and development of meningitis (Figure S4A-C). Together these results suggest that FcγRIa expression is critical for E. coli K1 to achieve high-grade bacteremia and for subsequent development of meningitis in newborn mice.
Our studies have shown that MØ isolated from E. coli K1 infected mice exhibit increased expression of FcγRI and TLR2, as well as increased production of nitric oxide (NO) due to iNOS activation [28]. We also observed that upregulation of CR3 expression on MØ led to enhanced killing of E. coli K1, whereas this effect was completely abrogated in CR3 siRNA transfected MØ in vitro. Other investigators have also demonstrated that CR3, TLR2 and TLR4 play important roles in the phagocytic ability of MØ [35]-[42]. Therefore, we examined whether the inability of E. coli K1 to survive in FcγRIa−/− mice was due to altered expression of surface receptors using flow cytometry. Peritoneal MØ isolated from infected FcγRIa−/− mice exhibited increased expression of CR3 and TLR4, but lower expression of TLR2 (Figure 7A). These cells also produced lower or negligible quantities of inducible NO upon challenge with E. coli K1, whereas MØ from WT mice generated six-fold higher amounts of NO at 6 h post-infection (Figure 7B). Furthermore, E. coli K1 binding to, and entry into, bone marrow-derived MØ (BMDMs) from FcγRIa−/− mice were significantly lower compared to WT MØ (Figure 7C and D). Some bacteria entered FcγRIa−/−BMDMs, but they were killed within a short period of time as determined by immunocytochemistry (data not shown). To substantiate the role of FcγRI in E. coli K1 entry, FcγRIa−/−BMDMs were transfected with hFcγRIa, FcγRIa-CT or FcγRII and then used for binding and invasion assays. As shown in Figure 7C, E. coli K1 binding to FcγRIa−/−BMDM/FcγRIa and FcγRIa−/−BMDM/FcγRIa-CT increased significantly compared to FcγRIa−/−BMDMs and FcγRIa−/−BMDM/FcγRII. However, entry was limited to binding to FcγRIa−/−BMDM/FcγRIa cells only. These results suggest that FcγRIa expression is critical for E. coli K1 binding to, and entry into, MØ and that the C-terminal domain plays a significant role for the entry. FcγRIa−/−BMDM transfected with a FcγRIa construct exhibited decreased expression of TLR4 and CR3 and increased expression of FcγRI and TLR2 in comparison with FcγRIa−/−BMDM after challenge with E. coli K1 (p<0.01) (Figure 7E). Transfection with FcγRIa-CT, however, resulted in only a partial increase or decrease of these surface molecules. In contrast, FcγRIa−/−BMDM transfected with FcγRII showed basal level expression of these molecules. Confirming the requirement of FcγRIa interaction with E. coli K1 to induce NO production, FcγRIa−/−BMDM/FcγRIa cells generated greater quantities of NO by 6 h post-infection as compared to FcγRI−/−BMDM/FcγRI-CT and FcγRI−/−BMDM/FcγRII cells (Figure 7F). Taken together, these data suggest that FcγRIa interaction with OmpA of E. coli K1 is necessary for suppression of CR3 and TLR4 expression and to enhance the expression of FcγRI and TLR2, and maximal NO production.
To confirm the role of FcγRIa expression on MØ in the pathogenesis of E. coli K1 meningitis, FcγRIa−/− mice were reconstituted with FcγRIa+/+ or FcγRIa−/− MØ and then infected with E. coli K1. FcγRIa−/− mice that received FcγRIa+/+ MØ showed higher blood bacterial numbers compared with animals replenished with FcγRIa−/− MØ (Figure 8A). 94% of CSF cultures were positive for E. coli K1 in FcγRIa+/+ MØ reconstituted mice, whereas all cultures were sterile in animals that received FcγRIa−/− MØ (Figure 8B). BBB disruption was significant in FcγRIa+/+ MØ-replenished animals compared to FcγRIa−/− MØ reconstituted mice (Figure 8C). Higher numbers of bacteria were also recovered from the brains of mice replenished with FcγRIa+/+ MØ compared to animals those received FcγRIa−/− MØ (Figure 8D). These results confirm that FcγRIa expression on MØ is critical for the onset of E. coli K1 meningitis.
The host response to infection starts with the identification of invading microorganisms via innate immune surveillance systems [43]. Nonetheless bacterial pathogens utilize very effective mechanisms to avoid host defenses in order to promote successful replication and dissemination [44]. MØ provide an important innate and adaptive immune coverage in the host, although their importance in E. coli K1 meningitis is unexplored. In the present study, we demonstrate that the expression of FcγRIα-chain in MØ is critical for the survival of E. coli K1 inside these immune cells by using MØ-depleted and FcγRIa−/− mice. It is tempting to speculate that the ability of E. coli K1 to survive inside MØ might enable these bacteria to infect the central nervous system via a “Trojan horse” mechanism. Pathogens that naturally infect the central nervous system, such as Brucella, Listeria, and Mycobacterium, have been demonstrated to use this mode of entry [45], [46]. We observed that the interaction of OmpA with FcγRIa in MØ is critical for bacterial binding to, entry into, and subsequent survival in these cells. Generally various FcγRs recognize microbes coated with either specific or non-specific antibodies. However a select number of microbes have developed methods to avoid this recognition. Protein A of S. aureus is known to bind to the Fc portion of the antibodies so that it avoids interacting with FcγRI, whereas most other microbes either downregulate phagocytic mechanisms or avoid phagocytosis entirely [47], [48]. This study therefore depicts the first evidence that a bacterial protein binds directly to FcγRIa to divert anti-microbicidal mechanisms.
Our competitive inhibition studies demonstrated that OmpA interacts with FcγRIa and can displace the binding of Fc portion of IgG. Therefore, it is possible that the bacteria in circulation, despite being coated with non-specific IgG, interact with MØ via FcγRIa for binding to and entering the cells for subsequent multiplication. OmpA− E. coli could not survive in MØ, suggesting that the interaction of OmpA with FcγRIa induces survival strategies or suppresses anti-microbial pathways in MØ. However, OmpA− E. coli has been shown to express reduced levels of type 1 fimbriae and susceptible to chemical stresses [49], [50]. Therefore, it is possible that OmpA− E. coli could be less capable of dealing with macrophage-induced stresses. Listeria, Shigella, and Rickettsia escape from the phagosome to the cytosol to avoid destruction in phagolysosomes [51]. Other pathogens interfere with the normal biogenesis of phagolysosomes, thus leading to the formation of replicative vacuoles [52], [53]. Since E. coli K1 continue to multiply inside phagosomes, one can speculate that phagosomes containing OmpA+ E. coli avoid lysosomal fusion by blocking phagosome maturation. The receptors expressed on the surface of MØ play a decisive role in the course of infection, whether pathogens are killed or the MØ machinery is taken over by the microbes [54]. Receptors like TLR2, TLR4 and CR3 have been implicated in the phagocytic ability of MØ [55], [56], [57]. Downregulation of CR3 expression on the surface of MØ has been associated with the decrease in the phagocytosis of pathogens and hence survival inside MØ [58]. TLR2 expression has been shown to prolong survival of Staphylococcus aureus inside phagosomes in MØ, which may be a strategy adopted by this pathogen to evade innate immunity. On par with this concept, TLR2 or MyD88 KO mice have been demonstrated to be resistant to sepsis, indicating that TLR2 mediated signaling is playing an important role in the survival of bacterial pathogens [59]. Activation of MØ through TLR4 has been shown to direct the induction of Th1 and Th-17 cells, which mediate protective cellular immunity to Bordetella pertussis by enhancing the bactericidal activity of MØ [60]. It is still to be determined whether TLR2 expression upon E. coli K1 infection has any role in the pathogenesis of meningitis.
We recently demonstrated that iNOS−/− mice and aminoguanidine (iNOS specific inhibitor) treated MØ showed enhanced expression of CR3 and TLR4 and very low levels of TLR2 and FcγRI, indicating that iNOS suppression results in decreased expression of FcγRI [28]. In agreement with these studies, we showed here that lack of FcγRI in MØ prevented the production of inducible NO and increased the expression of CR3 and TLR4, indicating that OmpA-FcγRIa interaction is critical for manipulating the surface expression of CR3 and TLRs in MØ. Our current results indicate that in E. coli K1 pathogenesis, FcγRI interaction with OmpA enhances the expression of TLR2, which in turn can be utilized by the bacteria as a receptor to modulate the efficiency of phagosome formation. Alternatively, E. coli K1 interaction with FcγRIa activates non-microbicidal mechanisms for the bacterial survival in MØ. Our studies have demonstrated that E. coli K1 infected MØ also exhibit increased expression of gp96, a known chaperone for TLR2 and TLR4 [28]. These interactions may also induce effector proteins into MØ by E. coli K1 that eventually are responsible for the control of macrophage environment. Further studies are in progress to examine these possibilities. As cytokines are known to modulate MØ microbicidal activity, it is also possible that the surface expression of TLRs and CR3 could be controlled by the circulating cytokines in E. coli K1 infection. Of note, we have demonstrated that IL-10 administration suppressed the expression of FcγRI and enhanced the expression of TLR4 and CR3, which in turn prevented the survival of E. coli K1 in MØ [61]. In contrast, for several other pathogens, circulating IL-10 supports intracellular replication, indicating that E. coli K1 pathogenesis is distinct from that induced by other bacterial pathogens [62].
Previous studies have shown that the cytoplasmic (CY) domain of FcγRIa plays an important role in phagocytosis and antigen presentation [63]. However, lack of the CY domain neither alters the association of γ-chain with FcγRIa nor influences the tyrosine phosphorylation of γ-chain in response to receptor specific cross-linking [63]. In contrast to these findings, we observed that OmpA binding to FcγRIa did not induce the association of γ-chain despite the presence of the CY domain. This binding also induced a different tyrosine phosphorylation response in MØ. Therefore, the CY domain of FcγRIa induces signaling events independent of γ-chain during the invasion of E. coli K1. Similarly, Qin et al demonstrated that the CY domain induces different gene expression in murine MØ compared to MØ stably transfected with CY-deleted FcγRIa [64]. Alteration of signal transduction pathways to impair FcγR-mediated phagocytosis has also been observed in HIV infected MØ, which have downregulated the expression of the γ-subunit [65]. Moreover, direct interaction of periplakin with the CY domain of human FcγRIa can confer unique properties on this receptor [66]. It should be noted that there are significant differences in the cytoplasmic regions of human and murine FcγRIa. However, our data demonstrate that the interaction of OmpA induced a similar response in both human and murine MØ. In summary, our studies provide the first evidence that a bacterial protein interacts directly with FcγRIa in order to bind to and enter MØ and manipulates the intracellular signaling for bacterial survival and multiplication. The new repertoire of interaction also suggests that MØ function may be manipulated by targeting additional epitopes without activating MØ microbicidal function. This strategy will be useful for devising novel methods of therapy for other diseases involving FcγRIa in addition to neonatal E. coli K1 meningitis.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal care and Use Committee (IACUC) of The Saban Research Institute of Childrens Hospital Los Angeles (Permit number: A3276-01). All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering.
E. coli E44, a rifampin-resistant mutant of E. coli K1 strain RS 218 (serotype O18:K1:H7), has been isolated from the cerebrospinal fluid of a neonate with meningitis and invades human brain microvascular endothelial cells (HBMEC) [34]. E91, a derivative of E44 in which ompA gene is disrupted (designated as OmpA− E. coli) and HB101 (a laboratory E. coli strain that expresses K-12 capsular polysaccharide) are noninvasive in HBMEC [34]. Group B streptococcus type III strain COH-1 used in these studies was provided by Dr. Craig Rubens of Seattle Children's Hospital, Seattle [67]. All bacteria were grown in brain heart infusion broth with appropriate antibiotics as necessary. Bacterial media were purchased from Difco laboratories (Detroit, MI).
Murine MØ cell line RAW 264.7, human macrophage like cells, THP-1 and COS-1 cells were obtained from American Type Culture Collection (Manassas, VA). COS-1 cells were stably transfected with cDNA encoding human FcγRIa, a mutant form of FcγRIa containing a stop codon after first amino acid of the cytoplasmic domain (Lys315→Stop 315) (FcγRI-CT), or with human FcγRII [67]. Anti- FcγRI (blocks the binding of Fc-portion of IgG to FcγRI), anti-CD11b, anti-CD32, anti-MR, anti-TLR2, anti-TLR4 and anti-Myc antibodies were obtained from Cell signaling. Purified IgG2a and FITC-IgG2a were obtained from Sigma (St. Louis, MO). Anti-gp96 antibody was raised in our lab as previously described [34], [68]. Anti-phospho-tyrosine antibody (4G10) was obtained from BD Sciences and all secondary antibodies coupled to various fluorophores were obtained from Bio-Rad Labs (Hercules, CA).
Confluent MØ monolayers in 24-well plates were incubated with 1×106 E. coli K1 (multiplicity of infection of 10) in experimental medium (1:1 mixture of Ham's F-12 and M-199 containing 5% heat-inactivated fetal bovine serum) for 60°min at 37°C, whereas COS-1 cell monolayers were infected with E. coli K1 at an MOI of 100 for 1.5 h. The monolayers were washed three times with RPMI 1640°and further incubated in experimental medium containing gentamicin (100 µg/ml) for 1 h to kill extracellular bacteria. The monolayers were washed again and lysed with 0.5% Triton X-100. The intracellular bacteria were enumerated by plating on sheep blood agar. In duplicate experiments, the total cell associated bacteria were determined as described for invasion except that the gentamicin step was omitted.
SureSilencing shRNA plasmids to mouse FcγRIa and CR3 (CD11b) in the pGeneClip Neomycin Vector were obtained from Super Array Inc., (Frederick, MD). RAW 264.7 cells were transfected with shRNA plasmids using Lipofectamine 2000 and later selected for G418 resistant colonies.
The cell surface proteins of THP-1 cells differentiated into MØ (THP-M) and RAW 264.7 cells were biotinylated by adding to 0.1 M sodium bicarbonate buffer (pH 8.0) containing 0.5 mg/ml NHS-LC-Biotin (Pierce Co, Rockford, IL) at a final protein concentration of 2 mg/ml in tissue culture flasks. The flasks were incubated on ice for 1 h, the cells were extensively washed with ice-cold PBS and solubilized in 5% Triton X-100 in PBS. Total membranes from the cells were isolated following extensive dialysis against PBS and then were concentrated using Centricon tubes (Millipore, Bedford, MA; 10-kDa cut-off). Biotinylated proteins (2–5 µg) were incubated with various bacteria from a 5-ml overnight culture in a volume of 0.5 ml at 37°C on a rotator for 1 h. The bacteria were then centrifuged and the pellets were washed three times with PBS containing 0.1% Triton X-100. After a final wash, the bound proteins were released with Laemmli buffer in the presence of β-mercapto-ethanol and analyzed by SDS-PAGE. The separated proteins were transferred to nitrocellulose and immunoblotted with streptavidin coupled to peroxidase. The protein bands were visualized by ECL reagent (Amersham Biosciences, Piscataway, NJ).
Total RNA was isolated from various transfected RAW 267.4 cells with TRIZOL-LS-reagent (Gibco BRL, Gaithersburg, MD) and quantified using a nanodrop machine. RT-PCR was performed using the following primer sequences: FcγRIa (321 bp) FP 5′-TCCTTCTGGAAAATACTGACC-3′ and RP 5′ GTTTGCTGTGGTTTGAGACC-3′; TLR2 (459 bp) FP 5′-TGAGAGTGGGAAATA TGGAC-3′; RP 5′-CCTGGCTCTATAACTCTGTC-3′; TLR4 (506 bp) FP 5′- TGGAT ACGTTTCCTTATAAG-3′ and RP 5′-GAAATGGAGGCACCCCTTC- 3′; GAPDH (479 bp), FP 5′-CACAGTCCATGCCATCACTG-3′ and RP 5′- TACTCCTTGGAG GCCATGTG -3′. Negative control assays without primers were performed in parallel for every reaction. The amplified products were separated on a 1% agarose gel and were stained with ethidium bromide.
Expression of FcγRI, CR3, TLR2 and TLR4 was detected by staining with appropriate FITC-, phycoerythrin (PE)-, PE-CY5.5-, or allophycocyanin (APC)-coupled mouse monoclonal antibodies (eBiosciences, San Diego, CA). Cells were first pre-incubated for 20 min with IgG blocking buffer to mask non-specific binding sites and then further incubated with the indicated antibodies or an isotype control antibody for 30 min at 4°C. The cells were subsequently washed three times with PBS containing 2% FBS and fixed with BD Cytofix (BD Biosciences). Cells were then analyzed by four-color flow cytometry using FACS calibur Cell Quest Pro software (BD Biosciences, San Jose, CA). Side and forward scatter parameters for which F4/480 was used as a MØ-gating marker, which formed the collection gate and at least 5000 events within this gate were collected for analysis.
Newborn C57BL/6 mice were injected intraperitoneally with (20-mg/Kg body weight) α-carrageenan (Sigma, St. Louis, MO) on days 1, 2 and 3 before infecting with E. coli. In control groups, mice were treated with equal volumes of saline.
Three-day old mice were randomly divided into various groups and infected intranasally with 103 CFU of bacteria. Control mice received pyrogen free saline through the same route. Blood was collected from the tail or facial vein at designated times post-infection and plated on LB agar containing rifampicin to assess bacteremia and level of infection. CSF samples were collected aseptically under anesthesia by cisternal puncture and directly inoculated into broth containing antibiotics. Mice were perfused intracardially with 0.9 % saline to remove blood and contaminating intravascular leukocytes. Brains were aseptically removed and homogenized in sterile PBS. Bacterial counts in all tissues were determined by plating ten-fold serial dilutions on rifampicin LB agar plates. Growth of E. coli in rifampicin containing LB broth from the CSF samples was considered positive for meningitis [28].
Determination of leukocytes in livers and spleens of untreated and carrageenan treated mice was done using flow cytometry [61]. PMNs were identified by staining with anti-Ly6-G (GR-1) followed by goat anti-rat- phycoerythrin (PE). CD4+ and CD8+ T lymphocytes were stained with rat anti-mouse-CD4 followed by goat anti-rat-PE and anti-CD8-FITC. DCs were stained with APC conjugated anti-CD11c antibody. B lymphocytes were detected by staining with anti-CD45R (B220)-FITC. Flow cytometry was performed on a FACScan instrument (BD Biosciences, CA) and the data were analyzed with Cell Quest Software.
Total cell lysates of RAW 264.7 cells infected with bacteria for varying time periods were centrifuged at 16,000 X g for 20 min at 4°C. The supernatants were collected and the protein contents determined. For immunoprecipitation studies, 300–500 µg of protein was incubated with the appropriate antibody overnight at 4°C, washed and further incubated for 1 h with protein A-agarose. The immune complexes were washed four times with cell lysate buffer and the proteins bound to agarose were eluted in SDS sample buffer for further analysis by Western blotting. Portions of the cell lysates were subjected to electrophoresis on a 10% SDS-polyacrylamide electrophoresis gel. The proteins were transferred to a nitrocellulose membrane, which was then blocked with 5% bovine serum albumin (BSA) in Tris-buffered saline containing 0.05% Tween 20 (TBST) for 2 h at room temperature. The blot was then incubated with the primary antibody overnight at 4°C in 5% BSA/TBST. The blot was washed with TBST and further incubated with the horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. Subsequently, the blot was washed four times with TBST for 1 h, developed with SuperSignal chemiluminescence reagent, and exposed to x-ray film to visualize the proteins.
RAW 264.7 cells were incubated with E. coli K1 at an MOI of 10 for varying times, washed and then fixed with 2% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.1. All samples were washed three times in 0.1 M cacodylate buffer for 15 minutes each. The cells were then post-fixed for 20 minutes in 1% osmium tetroxide at 4°C followed by addition of EtOH (60%). Samples were dehydrated through 70, 80, 95, and 100% EtOH (two times, 15 min each), then into propylene oxide (two time, 15 min each), and into a 1:1 propylene oxide/Eponate, left overnight, capped, at room temperature. The propylene oxide/Eponate mixture was decanted off and replaced with 100% Eponate mixture. The samples were polymerized at 70°C for 48 h. Thin sections (∼80 nm) were cut using a diamond knife, mounted on un-coated 300 mesh copper grids and stained with 5% uranyl acetate for 20 min. Photographs were take with a transmission electron microscope (JEOL JEM 2100 LaB6) equipped with a Gatan Ultra Scan 1000 CCD camera.
COS-1 cells were grown in 24-well tissue culture plate to confluence and then treated with 0.5 µg/ml of cytochalasin D for 30 min prior to addition of bacteria. OmpA− E. coli were incubated with anti-S-fimbria antibody for 1 h on ice, washed, and then added to the COS-1 monolayers at an MOI of 100 for 1 h. OmpA− E. coli alone infected monolayers served as controls in these experiments. The monolayers were then washed to remove unbound bacteria and incubated with OmpA+ E. coli at an MOI of 10 and 100 for 10 min, washed the monolayers, and then dissolved in 150 µl of PBS containing 0.3% Triton X-100. Serial dilutions were made and plated on agar containing tetracycline (12.5 µg/ml) in which only OmpA− E. coli grow. The number of CFU was counted and determined the percent displacement by OmpA+ E. coli. In some experiments, FITC-IgG2a (1 µg) was incubated with cytochalasin-D treated peritoneal MØ while rotating the test tube at a low speed for 30 min and washed to remove unbound IgG. Various inocula of OmpA+ E. coli or OmpA− E. coli were added to the cells and incubated for 10 min, washed and the bound FITC-IgG was determined by flow cytometry.
RAW 264.7 cells were grown in eight-well chamber slides and infected with E. coli K1 as described above. The monolayers were then washed with PBS and fixed in 2% paraformaldehyde for 10 min at room temperature. Subsequently, anti-S-fimbria antibody (1:1000 dilution) was added to the cells and incubated for 1 h at room temperature. The cells were then washed with PBS and incubated with secondary antibodies conjugated to FITC for 30 min at room temperature. The monolayers were washed four times with PBS and incubated with excess amounts of secondary antibody coupled to horseradish peroxidase for 1 h at RT to block the external primary antibody sites. After thorough washing of the cells, the monolayers were permeabilized with 5% normal goat serum in phosphate-buffered saline containing 1% Triton X-100 (NGS/PBST) for 30 min. The cells were again incubated with anti-S-fimbria antibody for 1 h in Triton/NGS/PBST buffer, washed and further incubated with secondary antibody coupled to Cy3 for 30 min. The cells were washed again, the chambers removed, and the slides mounted in Vectashield (Vector Laboratories) anti-fade solution containing 4′, 6-diamidino-2-phenylindole. Cells were viewed using a Leica (Wetzlar, Germany) DMRA microscope with Plan-apochromat ×40/1.25 NA and ×63/1.40 NA oil immersion objective lenses. Image acquisition was with a SkyVision-2/VDS digital CCD (12-bit, 1280×1024 pixel) camera in unbinned or 2×2-binned models into EasyFISH software, saved as 16-bit monochrome, and merged as 24-bit RGB TIFF images (Applied Spectral Imaging Inc., Carlsbad, CA). The images were assembled and labeled using Adobe PhotoShop 7.0.
BBB permeability was quantitatively evaluated by detection of extravasated Evans blue dye [28]. Briefly, 2% Evans blue dye in saline was injected intraperitoneally into infected or uninfected mice and after 4 h, mice were deeply anesthetized with Nembutal and transcardially perfused with PBS until colorless perfusion fluid was obtained from the right atrium. Brains from infected animals were harvested, weighed and homogenized. Tissue supernatant was obtained by centrifugation and protein concentration was determined. Evans blue intensity was determined on a microplate reader at 550 nm. Calculations were based on external standards dissolved in the same solvent. The amount of extravasated Evans blue dye was quantified as micrograms per milligram protein.
Peritoneal MØ were isolated from mice according to the method of Mittal et al [28], [69], [70]. Briefly, the mouse peritoneal cavity was exposed carefully without disrupting blood vessels and 2–3 ml of RPMI was slowly injected. The lavage was collected and cultured in tissue culture flasks for 2 h at 37°C under 5% CO2 to allow adherence of MØ. Non-adherent cells were removed and the flasks washed three times with Hanks' solution. The adherent cells were harvested from the flasks using a rubber policeman and were resuspended in 10% FCS-RPMI 1640 medium. MØ were then positively selected using Miltenyi biotech kit and percentage purity examined by FACS analysis using F4/80 antibody, which was >97%. Viability of MØ following interaction with bacteria was assessed using an Annexin V kit (BD Biosciences, San Diego, CA). Mouse bone marrow cells were isolated from the tibias and femurs of 6- to 10-wk-old WT and FcγRI−/− mice [71]. After euthanasia of mice by CO2 asphyxiation, femurs were harvested and bone marrow cells aseptically flushed from the marrow cavities with ice-cold PBS. Cells were collected by centrifugation and erythrocytes were lysed by resuspending in 0.15 M NH4Cl for 3–5 min. Celle were washed with PBS and resuspended in complete DMEM medium supplemented with M-CSF (10 ng ml−1) and IL-3 (10 ng ml−1), plated and allowed to differentiate into MØ. After 5–7 days in culture, adherent MØ were washed with PBS, scraped gently into suspension and counted. The purity of the MØ was determined by flow cytometry using F4/80 antibody and found to be >95%. Fresh bone marrow derived MØ (5×106 cells) were transferred by intraperitoneal injection into mice 6 h before infecting with E. coli K1.
NO production was determined in MØ supernatants by a modified Griess method as described earlier [28], [72], [73]. Briefly nitrate was converted to nitrites with β-nicotinamide adenine dinucleotide phosphate (NADPH, 1.25 mg ml−1) and nitrate reductase followed by addition of Griess reagent. The reaction mixture was incubated at room temperature for 20 min followed by addition of TCA. Samples were centrifuged, clear supernatants were collected and optical density was recorded at 550 nm. The amounts of NO produced were determined by calibrating standard curve using sodium nitrite.
Half of the brain was fixed in 10% buffered formalin, routinely processed and embedded in paraffin. 4–5 µm sections were cut using a Leica microtome and stained with hematoxylin and eosin (H & E). Pictures were taken using a Zeiss Axiovert Microscope connected to a JVC 3-chip color video camera and read by the pathologist in a blinded fashion.
Cytokine (TNF-α, IL-1β, IL-6, IL-12 p70 and IL-10) levels in sera from various animals were determined using Biosource ELISA kits (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions.
For statistical analysis of the data, two tailed Fischer test, Wilcoxon signed rank test and Student's t-test were applied and p value <0.05 was considered statistically significant.
|
10.1371/journal.pntd.0003170 | Nematode-Induced Interference with Vaccination Efficacy Targets Follicular T Helper Cell Induction and Is Preserved after Termination of Infection | One-third of the human population is infected with parasitic worms. To avoid being eliminated, these parasites actively dampen the immune response of their hosts. This immune modulation also suppresses immune responses to third-party antigens such as vaccines. Here, we used Litomosoides sigmodontis-infected BALB/c mice to analyse nematode-induced interference with vaccination. Chronic nematode infection led to complete suppression of the humoral response to thymus-dependent vaccination. Thereby the numbers of antigen-specific B cells as well as the serum immunoglobulin (Ig) G titres were reduced. TH2-associated IgG1 and TH1-associated IgG2 responses were both suppressed. Thus, nematode infection did not bias responses towards a TH2 response, but interfered with Ig responses in general. We provide evidence that this suppression indirectly targeted B cells via accessory T cells as number and frequency of vaccine-induced follicular B helper T cells were reduced. Moreover, vaccination using model antigens that stimulate Ig response independently of T helper cells was functional in nematode-infected mice. Using depletion experiments, we show that CD4+Foxp3+ regulatory T cells did not mediate the suppression of Ig response during chronic nematode infection. Suppression was induced by fourth stage larvae, immature adults and mature adults, and increased with the duration of the infection. By contrast, isolated microfilariae increased IgG2a responses to vaccination. This pro-inflammatory effect of microfilariae was overruled by the simultaneous presence of adults. Strikingly, a reduced humoral response was still observed if vaccination was performed more than 16 weeks after termination of L. sigmodontis infection. In summary, our results suggest that vaccination may not only fail in helminth-infected individuals, but also in individuals with a history of previous helminth infections.
| Parasitic worms, called helminths, infect one-third of the world population. Despite exposure to their host's immune system many helminths establish chronic infections and survive several years within their host. They avoid elimination by dampening the immune response of their hosts. This immune suppression also affects immune responses to third-party antigens such as vaccines. Indeed, accumulating evidence suggests that helminth-infected humans display impaired responses to vaccination. Thus, anthelminthic treatment before vaccination is discussed. Here, we use helminth-infected mice to analyse kinetics and mechanism of helminth-induced interference with vaccination efficacy more precisely. We show that chronic helminth infection completely suppressed antibody responses to a model vaccine. Thereby helminths suppressed the antibody-producing B cells indirectly via suppression of accessory T helper cells. The suppression was more pronounced at later time points of infection and still observed in mice that had terminated the helminth infection for more than 16 weeks. In summary, our results suggest that vaccination may not only fail in helminth-infected individuals, but also in individuals with a history of previous helminth infections. Thus, our report highlights the importance to develop vaccination strategies that are functional despite concurrent helminth infection rather than deworming humans before vaccination.
| More than 1 billion people are infected with helminths worldwide, predominantly in the tropics and subtropics [1]. To avoid their elimination and to limit pathology, helminths have developed sophisticated strategies to dampen the immune response of their hosts [2], [3]. This helminth-induced immune suppression also affects the immune response to third-party antigens and thus may interfere with efficient response to co-infecting pathogens and to vaccination [4]. Reduced cellular and humoral responses have been observed in helminth-infected humans after cholera [5], Bacillus Calmette-Guerin (BCG) [6], [7], tetanus toxoid [8], [9], [10], [11], [12] and anti-Plasmodium falciparum vaccination [13] (reviewed in [14], [15], [16]). Drug-induced termination of helminth infection improved responses to BCG [6], [17] and cholera [5], [18] vaccination. These collective studies emphasize that in addition to the pathology caused by helminth infection itself, helminth-induced interference with vaccination efficacy represent a global health problem.
As the nature of human studies is predominantly descriptive, murine models of helminth infection have been established to further investigate these issues. Responses to HIV or BCG vaccination were compromised in mice infected with the pathogenic trematode Schistosoma mansoni [19], [20]; similarly infection with the gastrointestinal nematode Heligmosomoides polygyrus interfered with humoral and cellular responses to malaria vaccinations [21], [22]. The nematode Litomosoides sigmodontis is used to model human filarial infections [23], [24]. L. sigmodontis third stage larvae (L3) are transmitted during the blood meal by the arthropod intermediate host, the mite Ornithonyssus bacoti. The natural definitive host is the cotton rat Sigmodon hispidus; however, some laboratory mouse strains such as BALB/c mice are fully susceptible to infection. L3 migrate during the first three days via the lymphatic system to the thoracic cavity. They moult to fourth stage larvae (L4) within 10 days and to immature adults within 30 days. Mature adults mate around day 55 post-infection (p.i.) and females release first stage larvae known as microfilariae (MF) into the peripheral circulation by day 60 p.i. Although BALB/c mice eventually control parasite burden by innate and adaptive immunity, they remain infected for several months until the parasites are fully eradicated. Thus, L. sigmodontis-infected BALB/c mice provide a suitable model to analyse the impact of chronic nematode infections on co-infections [25], [26], autoimmune diseases [27], [28] and allergy [29].
We used L. sigmodontis infection to study the effect of concurrent nematode infection on vaccination efficacy. An experimental vaccination against the liver stage of Plasmodium berghei using a single injection of a P. berghei circumsporozoite (CSP) fusion protein induced lower numbers of CSP-specific CD8+ cytotoxic T cells, if vaccinated mice were infected with L. sigmodontis [30]. Using semi-resistant C57BL/6 mice we have shown that acute L. sigmodontis infection suppressed humoral responses to thymus-dependent (TD) model antigen vaccinations performed at day 14 of infection [31]. Since C57BL/6 mice terminate L. sigmodontis infection around day 60 before patency is established [32] the impact of chronic nematode infection can not be modelled using this mouse strain.
In the current study we use the fully susceptible BALB/c mice to analyse the impact of different L. sigmodontis life stages on vaccination efficacy. We report an almost complete suppression of IgG response to TD vaccination in chronically infected mice. Suppression was induced by L4 and by parasitic adults that outcompeted the pro-inflammatory effect stimulated by MF. Suppression was observed even when vaccination was performed several months after termination of L. sigmodontis infection. We provide evidence that suppression of TD response during chronic nematode infection was established by interference with follicular B helper T cell (TFH) induction, independent of Foxp3+ Treg.
Animal experimentation was conducted at the animal facility of the Bernhard Nocht Institute for Tropical Medicine (BNITM) in agreement with the German animal protection law under the supervision of a veterinarian. The experimental protocols have been reviewed and approved by the responsible Federal Health Authorities of the State of Hamburg, Germany, the “Behörde für Gesundheit und Verbraucherschutz” permission number 98/11. BALB/c mice, cotton rats (S. hispidus) and BALB/c “Depletion of Regulatory T cell” (DEREG) mice were bred in the animal facilities of the BNITM and kept in individually ventilated cages under specific pathogen-free conditions. Mice were sacrificed by deep CO2 narcosis.
The life cycle of L. sigmodontis was maintained as described [31]. Six- to eight-week-old female mice were naturally infected by exposure to L. sigmodontis-infected mites (O. bacoti) that transmit infectious L3 during the blood meal. L4, adult worms and granulomata were harvested by flushing the thoracic cavity of infected mice with 8–10 mL of PBS. Parasites were counted subsequently. To detect MF in the circulation, blood of infected mice was collected in EDTA tubes; subsequently 20 µL of blood was added to 100 µL of ddH2O, and centrifuged at 10,000× g for 5 min. The pellet was resolved in 20 µL of Gentian violet and all MF were counted.
L. sigmodontis MF were purified from blood of infected cotton rats by density gradient centrifugation on Percoll. EDTA blood of cotton rats was collected and diluted 1∶2 with PBS. Iso-osmotic Percoll (Sigma-Aldrich, Munich, Germany) was prepared by mixing 9 parts of Percoll (density, 1.130 g/mL) with 1 part of 2.5 M Sucrose (Sigma-Aldrich, Munich, Germany). The following dilutions of 90% Percoll in 0.25 M Sucrose were made: 25% and 30% and layered with the diluted blood on top. After centrifugation at 400× g for 30 min at room temperature (RT) without brakes, MF are located between the 25% and 30% layer. MF were harvested, washed twice with PBS by 30 min centrifugation at 400× g and counted. Mice received 10,000 viable MF i.v.
Non-infected and L. sigmodontis-infected mice were vaccinated at indicated time points post infection by i.p. injection of either 100 µg alum-precipitated dinitrophenol-keyhole limpet hemocyanin (DNP-KLH, Sigma-Aldrich, Munich, Germany), 100 µg 4-hydroxy-3-iodo-5-nitrophenylacetyl (NIP) conjugated to Ficoll (NIP-Ficoll) (Biosearch Technologies, Navato, USA), or by s.c. injection of 30 µg alum-precipitated DNP-KLH into the hind footpad. For analysis of serum antibodies, blood was collected from mice by submandibular bleeding of the facial vein 7, 14 and 21 days after vaccination and allowed to coagulate for 1 h at RT. Serum was collected after centrifugation at 10,000× g for 10 min at RT and stored at −20°C for further analysis. For analysis of spatial separated cellular responses, mice were sacrificed at the indicated time point, and spleen and popliteal lymph nodes (popLN) were dissected. A total of 2.5×105 splenocytes or popLN cells were cultured in 3–5 replicates in 96-well round-bottom plates in RPMI 1640 medium supplemented with 10% FCS, 20 mM HEPES, 2 mM L-glutamine and gentamicin (50 µg/mL) at 37°C and 5% CO2. The supernatant was harvested after 21 days of culture and DNP- and L. sigmodontis-specific IgG1, IgG2a and IgG2b were quantified by ELISA.
DNP-KLH was chosen as TD model antigen as no cross-reaction between DNP-specific Ig and L. sigmodontis antigen was detected and comparison of DNP7-BSA and DNP38-BSA allows detection of high affinity only as well as high and low affinity Ig. For the detection of DNP-, NIP- and L. sigmodontis-specific Ig, ELISA plates were coated overnight with 1 µg/mL DNP7-BSA, (Sigma-Aldrich, Munich, Germany), 1 µg/mL NIP7-BSA (Biosearch Technologies, Navato, USA), or 4 µg/mL L. sigmodontis extract in carbonate buffer pH 9.6. Low affinity DNP-specific IgG was detected by coating ELISA plates with 1 µg/mL DNP38-BSA (Biotrend Chemikalien, Cologne, Germany). Plates were washed, blocked by incubation with PBS 1% BSA for 2 h and incubated for 2 h with serum or cell culture supernatant. Plates were washed and incubated for 1 h with horseradish peroxidase-labelled anti-mouse IgM, IgG1, IgG2a, IgG2b (Life Technologies, Carlsbad, USA), or IgG3 (Southern Biotechnology Associates, Birmingham, USA). Plates were washed and developed by incubation with 100 µL tetramethylbenzidine (0.1 mg/mL), 0.003% H2O2 in 100 mM NaH2PO4 pH 5.5 for 2.5 min. Reaction was stopped by addition of 25 µL 2 M H2SO4, and OD450 was measured. For the more abundantly produced isotypes IgG1 and IgM, titres were calculated by defining the highest serum dilution in a serial dilution (1∶1000 to 1∶128,000) resulting in an OD450 above the doubled background. For the less abundant isotypes IgG2a, IgG2b, and IgG3, arbitrary units were calculated by subtraction of OD450 of the background from OD450 of one fixed serum concentration (1∶100 for IgG2a, 1∶1000 for IgG2b, and 1∶100 for IgG3). Background was generally below OD450 = 0.1. Calculation of titres by serial dilution for random control samples revealed similar differences in DNP-specific IgG2a and IgG2b production as the arbitrary units and thus did not yield additional information (data not shown).
For analysis of B and TFH cells, mice were sacrificed at the indicated time points and popLN were dissected. Cells (1×106) were stained with Live/Dead Fixable Blue Dead Cell Stain Kit (Life Technologies, Carlsbad, USA) according to the manufacturer's instructions. For surface staining, cells were stained with anti-CXCR5-Biotin (clone: 2G8) for 30 min at 37°C or with anti-IgG-phycoerythrin-Cy7 (PE-Cy7) (clone: Poly4053) and Biotin-labelled peanut agglutinin (PNA) (Galab Technologies, Geesthacht, Germany) for 30 min at 4°C. After washing the cells were stained with anti-CD3e-allophycocyanin (APC) (145-2C11), anti-CD4-APC/Brilliant Violet 510 (BV510)/PE (clone: RM4-5), anti-CD44-BV421 (clone: IM7), anti-PD1-fluorescein isothiocyanate (FITC) (clone: 29F.1A12), anti-ICOS-PE (clone: 7E.17G9), anti-CD19-PE (clone: 6D5), DNP-BSA-AF647, anti-IgM-FITC (clone: DS1), Strepavidin-APC and Strepavidin-BV421 for 30 min at 4°C. Foxp3 expression was determined using PE-anti-mouse Foxp3 Staining Set (clone: FJK-16S, Affymetrix eBioscience, Frankfurt, Germany) according to the manufacturer's instructions. Foxp3+ Treg depletion was controlled by analysis of eGFP, Foxp3 and CD4 expression. Samples were analysed on a LSR II Flow Cytometer (Becton Dickinson, Mountain View, USA) using FlowJo software (TreeStar, Ashland, USA). DNP-BSA was fluorescence labelled using Alexa Fluor 647 Protein Labeling Kit (Life Technologies, Carlsbad, USA) according to the manufacturer's instructions. Unless otherwise stated all staining antibodies were purchased from BioLegend (Fell, Germany), BD Biosciences (Heidelberg, Germany) or Affymetrix eBioscience (Frankfurt, Germany).
Heterozygous BALB/c DEREG mice and non-transgenic littermate control BALB/c mice received 0.5 µg diphtheria toxin (DT) (Merck, Darmstadt, Germany) dissolved in PBS i.p. on three consecutive days, starting either two days prior to L. sigmodontis infection or one day before vaccination. Successful depletion of Foxp3+ Treg was routinely controlled by staining for CD4+Foxp3+ cells in the peripheral blood after the third DT treatment.
Statistical analysis was performed by ANOVA with Bonferroni post-test or student's t-test using Prism software (GraphPad Software, San Diego, USA). Results are presented as mean ± SEM; p≤0.05 was considered statistically significant.
We used BALB/c mice that are fully susceptible for L. sigmodontis infection to analyse the impact of different parasitic life stages on vaccination efficacy. By choosing different durations between exposure to infected mites and subsequent vaccination we defined the life stage of L. sigmodontis present at the moment of vaccination (Figure 1A). We analysed the impact of L3 migrating via the lymphatic vessels to the thoracic cavity by vaccinating mice at the day of infection. To analyse the impact of L4 and young immature adults on vaccination efficacy, we vaccinated 14 and 30 days after infection. Vaccination at day 60 p.i. resulted in the presence of mature adults that had mated and released MF into the peripheral circulation.
We employed alum-precipitated DNP-KLH as a model vaccine for TD humoral response. High affinity DNP-specific IgG responses were analysed in nematode-infected and in age-matched non-infected control mice on three consecutive weeks following vaccination. Simultaneous nematode infection increased early responses to DNP-KLH vaccination (Figure 1B). In contrast, presence of L. sigmodontis L4 during vaccination reduced DNP-specific IgG1, whereas DNP-specific IgG2a was reduced by trend and IgG2b was unchanged (Figure 1C). Vaccination at later time points of infection, i.e. day 30 or day 60 p.i., resulted in statistically significant suppression of IgG1, IgG2a, and IgG2b responses to DNP-KLH, in comparison to age-matched non-infected mice (Figure 1D,E). DNP-specific IgG2 response was almost absent in day 60 infected mice.
The capture agent used in these experiments, i.e. BSA coupled to 7 DNP molecules, was suited to detect specifically high affinity IgG. Re-analysis of these sera with BSA coupled to 38 DNP molecules, a setting that will capture IgG displaying a lower affinity to DNP in addition to high affinity DNP-specific IgG, revealed similar results (Figure S1 in Text S1). Thus, the quantity but not the quality of IgG response to vaccination was reduced by established nematode infection.
In line with our previous results [31], [33], nematode infection did not interfere with the humoral response to the polyvalent TI model antigen NIP-Ficoll that activates B cells in the absence of T helper cells by strong crosslinking of the B cell receptor. Due to absent T cell co-stimulation NIP-Ficoll induces predominantly IgM responses and limited IgG1 and IgG3 responses [34]. Consequently we did not detect NIP-specific IgG2a or IgG2b in NIP-Ficoll vaccinated mice (data not shown) but NIP-specific IgM, IgG1 and IgG3 were produced (Figure 2). Thereby, NIP-specific humoral responses were similar, or in the case of IgG3, even increased in NIP-Ficoll-vaccinated non-infected in comparison to day 60 L. sigmodontis-infected mice.
Taken together, these results show that the presence of L4 and adult L. sigmodontis, but not of recently transmitted L3, suppressed humoral response to vaccination specifically in T cell-dependent settings. Intensity of suppression was positively correlated to duration of nematode infection. Chronic nematode infection suppressed both, TH2-associated IgG1 and TH1-associated IgG2 responses to vaccination, thus, inflicting generalized suppression and not polarization towards a type 2 immune response.
Despite the apparent suppression of TD humoral response in infected mice, L. sigmodontis-specific IgG1, IgG2a and IgG2b responses were detectable during infection (Figure S3 in Text S1). Additional DNP-KLH vaccination did not modulate the L. sigmodontis-specific Ig response, as expected (Figure S3 in Text S1: black circles and grey squares).
Between 40 and 60% of infected BALB/c mice developed detectable microfilaraemia by day 60 p.i., leading to simultaneous presence of two different life stages in mice vaccinated at this late time point. Stratification of DNP-specific IgG response of day 60 infected mice that were positive (n = 7) or negative (n = 12) for MF in the peripheral circulation revealed no differences in suppression (Figure S2 in Text S1). To differentiate between the impact of adults and MF on response to DNP-KLH vaccination, we injected 10,000 purified MF at the day of vaccination to model the recent release of MF by females. Interestingly, presence of isolated MF increased the IgG2a response to vaccination while the IgG1 and IgG2b responses remained unchanged (Figure 3A). Thus, MF displayed a pro-inflammatory effect, increasing TH1-associated Ig responses to third-party antigens. This finding suggests that the anti-inflammatory effect observed in day 60 L. sigmodontis-infected microfilaraemic mice was induced by adults and outcompeted the pronounced pro-inflammatory effect of isolated MF.
However, injection of 10,000 MF in a bolus may trigger stronger pro-inflammatory signals than the putative effects mediated by MF released gradually by female adults in vivo. To investigate if adults would also suppress the possibly stronger pro-inflammatory signals delivered by isolated MF we injected purified MF into day 60 L. sigmodontis-infected mice (Figure 3B). While injection of MF into non-infected mice increased IgG2a responses compared to naïve mice as observed before, MF-treated L. sigmodontis-infected mice displayed reduced IgG2a responses in comparison to MF-treated non-infected mice. IgG response in MF-treated L. sigmodontis-infected mice was also significantly lower than IgG response in non-infected mice. Taken together, these results show that adults suppress the IgG response to vaccination in the presence of circulating MF, despite pro-inflammatory stimuli transduced by MF.
Foxp3+ Treg are central regulators of adaptive immune responses and have been shown to mediate helminth-induced immune suppression [35]. Although our previous study did not indicate a function for Foxp3+ Treg in the suppression of CD4+ T cell proliferation during acute L. sigmodontis infection in C57BL/6 mice [31], accumulating evidence suggests that the dominance of Treg-mediated regulation differs in different mouse strains [36], [37], [38], [39]. Therefore we evaluated the contribution of Foxp3+ Treg to the suppression of vaccination efficacy in day 60 L. sigmodontis-infected BALB/c mice. To this end, we employed BALB/c DEREG mice that express a fusion protein consisting out of the human diphtheria toxin (DT) receptor and enhanced green fluorescent protein (eGFP) under the control of the Foxp3 promoter [40]. Injection of DT results into transient depletion of Foxp3+CD4+ T cells in DEREG mice while Treg frequencies in non-transgenic littermates remain unchanged [41]. As Foxp3+ Treg depletion is not permanent in DEREG mice, we investigated the effect of transient Foxp3+ Treg depletion either during initial infection (Figure 4A) or during vaccination of chronically infected mice (Figure 4D).
Absence of Foxp3+ Treg during the first days of L. sigmodontis infection did not abrogate suppression of IgG response to vaccination in nematode-infected BALB/c mice (Figure 4C). DNP-specific IgG1, IgG2a and IgG2b responses were equally reduced in nematode-infected mice containing Foxp3+ Treg (black squares) or not containing Foxp3+ Treg (blue squares). Successful depletion of Foxp3+ Treg was verified by flow cytometry at the day of L. sigmodontis infection (Figure 4B).
Transient Treg depletion at the moment of vaccination (Figure 4D) that was confirmed by flow cytometry one day after vaccination (Figure 4E) increased the DNP-specific IgG2a response in non-infected mice (Figure 4F). Similar increases in pro-inflammatory IgG2a responses upon Treg depletion were observed in a recent study using DEREG mice in a model of atopic dermatitis [42]. Increased IgG2a responses reflected most likely inefficient regulation due to the absence of Treg and, thus function as internal control for depletion efficacy. Nevertheless, increased DNP-specific IgG2a in Treg-depleted mice was still suppressed upon L. sigmodontis infection. Treg depletion during vaccination did not modulate the more abundant IgG1 or IgG2b responses and did not abrogate nematode-induced suppression of IgG response to vaccination. Taken together, these results rule out a contribution of Foxp3+ Treg to the suppression of IgG response to TD vaccination in nematode-infected BALB/c mice.
Natural infection with L. sigmodontis that predominantly dwell in the thoracic cavity induced a systemic immune response. Nematode-specific T and B cell responses are detectable in the draining lymph nodes and in the spleen (data not shown). In the experiments performed so far, mice were vaccinated i.p., thereby inducing systemic responses to DNP-KLH that are initiated mostly in the spleen. To rule out that suppression of IgG response to vaccination was caused by a simple competition of nematode- and vaccine-specific B and T cells in the same lymphatic organ, we separated the sites of L. sigmodontis-specific and DNP-KLH-specific immune responses (Figure 5).
To this end we vaccinated day 60 L. sigmodontis-infected and age-matched non-infected mice with DNP-KLH subcutaneously (s.c.) into the hind footpad. This regimen is suited to induce DNP-KLH-specific T and B cell responses predominantly in the draining popliteal lymph node (Figure 5AB). Fully differentiated plasma cells will predominantly migrate into the bone marrow and secrete Ig into the peripheral circulation. The induction of humoral responses in lymph nodes or in the spleen can be visualized by presence of B cells that secrete Ig spontaneously at low concentrations in cell cultures of these lymphatic organs. Cultured splenocytes derived from vaccinated and L. sigmodontis-infected mice secreted nematode-specific IgG1, IgG2a and IgG2b but did not secrete any DNP-specific IgG (Figure 5C). Cultured popliteal lymph node cells derived from the same mice produced DNP-specific but not nematode-specific IgG1, thus visualizing the spatial separation of B cell responses to parasite and vaccine. DNP-specific IgG2a and IgG2b concentrations in the culture supernatant were below the detection limit. Systemic serum titres of DNP-specific IgG1 and IgG2b was still suppressed in day 60 L. sigmodontis-infected mice after s.c. vaccination (Figure 5D), demonstrating that nematode-induced suppression acted on B cell responses that were primed in a local lymph node as well. As nematode-specific lymphocytes were not present in the local lymph nodes that drained the site of DNP-KLH vaccination (Figure 5C), the observed suppression of systemic DNP-KLH-specific IgG response was not mediated by competition within the same site. Mice that were L. sigmodontis-infected but not vaccinated produced no DNP-specific IgG at all (Figure 1B–E and Figure 5D: black circles) and mice that were DNP-KLH vaccinated but not L. sigmodontis-infected did not produce any L. sigmodontis-specific IgG at all (Figure S3 in Text S1: white squares).
To quantify vaccine-induced B cells, we directly stained DNP-specific CD19+ B cells in the lymph nodes of vaccinated mice. DNP-KLH vaccination into the hind footpad induced DNP-binding CD19+ B cells in the draining lymph nodes (Figure 6) while no DNP-binding B cells were detectable in the non-draining contralateral lymph nodes (data not shown). DNP-specific B cells also bound peanut agglutinin (PNA) (Figure 6A), indicating localization in the germinal centre [43]. Chronic nematode infection reduced the numbers of DNP-specific PNA-binding B cells (Figure 6AB). Reduced DNP-specific B cell numbers were observed in the IgM/IgG double positive and in the terminally switched, IgG single positive B cell population. Thus, the reduced quantity of DNP-specific Ig detected in the serum of nematode-infected vaccinated mice was reflected by reduced numbers of DNP-specific B cells in the draining lymph nodes.
Accumulating evidence suggests that a specialized T cell subset, the TFH, is responsible for provision of co-stimulation to B cells in the germinal centre [44], [45]. TFH can be identified by expression of the chemokine receptor CXCR5 and the regulatory receptor programmed death 1 (PD1), in addition to activation markers such as CD44 [46]. DNP-KLH vaccination into the hind footpad did not change the frequency of total CD4+ T cells in draining compared to non-draining lymph nodes (Figure 7B). However, the frequency of TFH, defined as PD1+CXCR5+ cells within the CD4+CD44+ population (Figure 7A), increased selectively in the lymph node draining the site of vaccination (Figure 7C). The strict correlation of TFH expansion within the stable CD4+ compartment to the site of vaccination in both, non-infected and nematode-infected mice, strongly suggests that these TFH were generated in response to vaccination.
Draining lymph nodes of mice that were infected with L. sigmodontis for 60 days at the moment of vaccination displayed a significant reduction in both, TFH frequency within activated CD4+ T cells and absolute TFH numbers in comparison to non-infected vaccinated mice (Figure 7C). Further characterization of TFH induced in non-infected and nematode-infected mice revealed no significant differences in the expression of inducible co-stimulator (ICOS) or Foxp3 (Figure 7DE). Thus reduced numbers of DNP-specific B cells in DNP-KLH-vaccinated nematode-infected mice were correlated with reduced numbers of vaccination-induced TFH whereas the phenotype remained comparable.
The collective data presented above suggest that vaccinations may fail in individuals carrying chronic nematode infections due to impaired induction of TFH. To evaluate the kinetics of this nematode-induced suppression, immune competent BALB/c mice were first allowed to terminate L. sigmodontis infection and then vaccinated with DNP-KLH (Figure 8). Immune-mediated control of infection was indicated by clearance of MF from the circulation that was observed between day 180 and day 280 p.i. (Figure 8B and data not shown). Responses to vaccination were still reduced in mice that were previously nematode-infected when vaccination was carried out immediately after termination of microfilaraemia and four or eight weeks after clearance of MF from the circulation (data not shown). Therefore we finally introduced an additional recovery period of 16 weeks after clearance of microfilaraemia before vaccination was performed (Figure 8A). Strikingly both, IgG1 and IgG2b responses to DNP-KLH vaccination were still significantly suppressed in mice that had terminated L. sigmodontis infection at least 16 weeks before vaccination (Figure 8C). Thereby DNP-specific IgG2b responses were almost absent in mice with a history of previous L. sigmodontis infection.
It should be noted that although clearance of MF from the circulation at these late time points of infection strongly suggests impaired fitness of the adult parasites due to immune-mediated control and extermination, it does not confer direct information about the presence and the status of adult parasites in the thoracic cavity at the moment of vaccination. No living parasites were detectable in 100% of infected and vaccinated mice three weeks after vaccination (data not shown). We observed limited amounts of remaining dead material in the thoracic cavity of some mice and we cannot formally exclude contribution of this helminth-derived material to suppression of vaccination.
In this study we demonstrate that chronic L. sigmodontis infection prevents humoral responses to bystander antigen vaccination. DNP-KLH-vaccinated and nematode-infected mice displayed reduced titres of DNP-specific IgG in the serum and reduced numbers of DNP-specific B cells in the lymph nodes compared to vaccinated, non-infected mice.
L. sigmodontis-induced suppression was restricted to TD vaccination and not observed during T cell-independent B cell vaccination using the TI-2 antigen NIP-Ficoll. In congruence with our previous study performed in semi-permissive, day 14 L. sigmodontis-infected C57BL/6 mice [31], this study strongly suggests that nematodes suppress antibody-producing B cells indirectly via suppression of accessory T cells. Within the CD4+ T cell population a specialized T cell subset, the TFH, is central for the initiation of classical B cell responses [44], [45]. Next to PD1 expression, TFH are further characterized by the continued expression of the chemokine receptor CXCR5 that regulates their localisation within the B cell follicle [46]. Separating the sites of nematode- and vaccine-induced immune responses we distinguished between nematode- and vaccine-induced TFH. L. sigmodontis-infected mice displayed significantly reduced numbers and frequencies of vaccine-induced TFH in the lymph nodes draining the site of vaccination. Interestingly, we did not detect differences in the phenotype of TFH regarding the expression of ICOS, a central co-stimulatory receptor for B and T cell interaction that is essential for antibody responses to TD antigens [47]. Follicular regulatory T cells (TFR) that arise from thymus-derived Foxp3+ Treg and display a TFH-like, CXCR5+ PD1+ phenotype have been implied in regulation of TD B cells responses via limitation of TFH and B cell numbers in the germinal centre [48], [49]. As Foxp3+ expression in TFH was unchanged and depletion of Foxp3+ T cells did not abrogate nematode-induced suppression, we ruled out a significant contribution of TFR to nematode-induced immune suppression.
We have shown before that transient gastrointestinal nematode infection predominantly suppressed TH1-associated IgG2 responses to vaccination [33], whereas L. sigmodontis infection of semi-permissive C57BL/6 mice [31] as well as chronic infection of fully susceptible BALB/c mice in the current study induced a generalized suppression of both TH1- and TH2-associated isotypes. We also did not observe differences in the affinity of vaccine-induced IgG in nematode-infected and non-infected mice. Taking these findings into account, we hypothesize that nematode infection interfered with the humoral response to vaccination already at the stage of TFH induction. Reduced numbers of vaccine-induced TFH will result in reduced provision of co-stimulation for vaccine-specific B cells. As a consequence reduced numbers of vaccine-specific B cells expand in the draining lymph node leading to reduced titres of vaccine-specific IgG in the peripheral circulation of nematode-infected mice. This hypothesis is supported by our previous study where we reported reduced proliferation of ovalbumin-specific TCR transgenic OT-II T cells as a simplified model for accessory T cells upon adoptive transfer into L. sigmodontis-infected mice [31].
Regarding the mechanism, we provide evidence that suppression of humoral response did not reflect direct competition between nematode- and vaccine-specific lymphocytes by separating the sites of nematode- and vaccine-specific immune responses. Suppression of OT-II T cell proliferation during acute L. sigmodontis-infection of C57BL/6 mice was shown to be Treg-independent and partially mediated by IL-10 [31]. Early depletion of CD4+CD25+ Treg improved host defence in L. sigmodontis-infected BALB/c mice, suggesting an implication of Treg in immune evasion for this genetic background [36]. Since we recently described a central role for Foxp3+ Treg in gastrointestinal nematode-induced immune evasion in BALB/c mice that was not functional in C57BL/6 mice [39], it was conceivable that Foxp3+ Treg would contribute to the observed suppression of TD vaccination in L. sigmodontis-infected BALB/c mice. However, in the current study we show that suppression of IgG response during chronic infection of BALB/c mice was clearly established in the absence of Foxp3+ Treg. This suggests that immune modulation during acute and chronic L. sigmodontis infection in C57BL/6 and BALB/c mice is established by similar mechanisms such as IL-10 induction [31]. Potential mediators of suppression in addition to Foxp3+ Treg are IL-10 producing Foxp3− T cells [50], IL-10 producing regulatory B cells [51], [52], alternatively activated macrophages [53], and tolerogenic dendritic cells [54] that have been shown to mediate suppression during nematode infection in several murine systems. We are currently testing the function of these regulatory cell populations in suppression of OT-II T cell proliferation during acute L. sigmodontis-infection of C57BL/6 mice.
Using fully susceptible BALB/c mice we dissected the impact of different L. sigmodontis life stages on vaccination efficacy. Suppression was induced by L4, immature and mature adults, but not by recently transmitted L3. Injection of isolated MF, in contrast, elevated IgG2a responses to vaccination, thus delivering a pro-inflammatory stimulus. A comparable pro-inflammatory stimulation was mediated by isolated MF in a model of LPS-induced sepsis [55]. In line with one previous study [56], we demonstrate that this pro-inflammatory effect of MF was dominated by the anti-inflammatory effect exerted by L. sigmodontis adults. By contrast, the MF-mediated aggravation of LPS-induced sepsis was not rescued by presence of adults but also induced by implantation of mature adults releasing MF [55]. The different outcome may reflect the impact of long exposure (i.e. 60 days) of the host to L. sigmodontis before MF occurred in the circulation in our study in contrast to sudden implantation of MF releasing adults. Prolonged exposure may be needed for complete immune modulation to silence the strong pro-inflammatory effect of MF. In line with this reasoning we observed that suppression of DNP-specific IgG response to vaccination clearly increased with duration of infection.
As our study focused on the first response to vaccination we used the early humoral and cellular response as indicator of efficacy. In order to model vaccination efficacy for the human situation more precisely, also the magnitude of memory responses several months after initial vaccination in absence and presence of helminth infection will be compared in future studies.
Suppression of vaccination responses, once established, was observed several months after immune-mediated termination of infection. We cannot exclude that remaining helminth-derived material in the thoracic cavity contributed to suppression in the mice that had cleared microfilaria from the circulation 16 weeks before vaccination was performed. However, comparable remnants of large parasites are likely to be present in humans with a history of previous filarial infection as well. Thus, the prolonged suppression of vaccination-induced responses reported in this study may have implications for health policy. While some murine studies suggest that drug-induced termination of helminth infection may improve vaccination efficacy after 1–3 weeks of recovery [21], [22], our results show that the immune status does not immediately return to normal responsiveness in a setting modelling chronic infection. Prolonged suppression of vaccination responses after drug-induced termination of infection have been described in other mouse models. However, in these studies responsiveness was eventually achieved after recovery periods of 8 and 16 weeks, respectively [57], [58]. Regarding the human population in the tropics where nematode infections are endemic a re-infection during such a recovery period is likely to occur. Despite the limits of murine models to reflect every aspect of the human situation, combined evidence gained in different mouse models for helminth infection can be informative. Since first murine studies demonstrated successful vaccination despite concurrent nematode infection by improved vaccination strategies [21], [30], [59], [60], we suggest that development of vaccination regimes that are functional despite pre-existing nematode infection would be more promising and should be considered in addition to deworming programs before vaccination.
|
10.1371/journal.pgen.1001069 | A Global Overview of the Genetic and Functional Diversity in the Helicobacter pylori cag Pathogenicity Island | The Helicobacter pylori cag pathogenicity island (cagPAI) encodes a type IV secretion system. Humans infected with cagPAI–carrying H. pylori are at increased risk for sequelae such as gastric cancer. Housekeeping genes in H. pylori show considerable genetic diversity; but the diversity of virulence factors such as the cagPAI, which transports the bacterial oncogene CagA into host cells, has not been systematically investigated. Here we compared the complete cagPAI sequences for 38 representative isolates from all known H. pylori biogeographic populations. Their gene content and gene order were highly conserved. The phylogeny of most cagPAI genes was similar to that of housekeeping genes, indicating that the cagPAI was probably acquired only once by H. pylori, and its genetic diversity reflects the isolation by distance that has shaped this bacterial species since modern humans migrated out of Africa. Most isolates induced IL-8 release in gastric epithelial cells, indicating that the function of the Cag secretion system has been conserved despite some genetic rearrangements. More than one third of cagPAI genes, in particular those encoding cell-surface exposed proteins, showed signatures of diversifying (Darwinian) selection at more than 5% of codons. Several unknown gene products predicted to be under Darwinian selection are also likely to be secreted proteins (e.g. HP0522, HP0535). One of these, HP0535, is predicted to code for either a new secreted candidate effector protein or a protein which interacts with CagA because it contains two genetic lineages, similar to cagA. Our study provides a resource that can guide future research on the biological roles and host interactions of cagPAI proteins, including several whose function is still unknown.
| Most humans are infected with Helicobacter pylori. The H. pylori cag pathogenicity island (cagPAI) encodes a secretion apparatus that can translocate the CagA protein into host cells. Humans infected with cagPAI–carrying H. pylori are at increased risk of severe disease, including gastric cancer. We analyzed the nucleotide sequences and functional diversity of the cagPAI in a globally representative collection of isolates. Complete cagPAI sequences were obtained for 29 strains from all known H. pylori biogeographic populations. The gene content and arrangement of the cagPAI and its function were highly conserved. Diversity in most cag genes consisted in large part of synonymous polymorphisms. However some genes—in particular those that encode proteins predicted to be secreted or located on the outside of the bacterial cell—had particularly high frequencies of non-synonymous polymorphisms, suggesting that they were under diversifying selection. Our study provides evidence that the cagPAI was only acquired once and provides an important resource that can guide future research on the biological roles and host interactions of cagPAI proteins, including several whose function is still unknown.
| Helicobacter pylori persistently infects more than one half of all humans, and can cause ulcer disease, gastric cancer, and MALT lymphoma [1]. The H. pylori cag pathogenicity island (cagPAI) is an intriguing virulence module of this obligate host-associated bacterium [2]–[4]. H. pylori strains that possess a functional cagPAI are particularly frequently associated with severe sequelae, notably gastric atrophy and cancer [4]–[7]. The cagPAI is ∼37 kb long, and contains ∼28 genes [3]. These genes encode multiple structural components of a bacterial type IV secretion system (t4ss) as well as the 128 kDa effector protein, CagA [7]. After H. pylori has adhered to a host cell, the Cag t4ss translocates CagA into that cell. CagA is subsequently phosphorylated by host cell kinases and interacts with multiple targets (e.g. SHP-2, Grb2, FAK), profoundly altering host cellular functions [8], [9]. The alterations induced by the cagPAI are thought to ultimately contribute to malignant transformation [4], [10], and CagA has been designated a bacterial oncoprotein [11].
H. pylori has a high mutation rate, which has resulted in extensive genetic diversity [12], and also recombines frequently with other H. pylori [13]. H. pylori isolates have been subdivided into distinct biogeographic populations and subpopulations with specific geographical distributions that reflect ancient human migrations [14]–[16]. The global population structure of H. pylori is now well understood based on multilocus haplotypes from seven housekeeping genes. However, very little is known about the biogeographic variation of virulence factors, such as the cagPAI, nor has the impact of genetic variation on disease outcome and host adaptation been adequately addressed. Previous analyses on the basis of comparative genome hybridization have demonstrated marked differences between biogeographic populations with respect to the cagPAI [17]. Microarray analysis of 56 globally representative strains of H. pylori revealed that the cagPAI was present in almost all strains from some biogeographic populations and subpopulations in Africa and Asia, while it was variably present in other populations [17]. The cagPAI was lacking in all isolates of hpAfrica2, which is distantly related to the other populations [17]. Currently, nine complete cagPAI sequences are publicly available [2], [18]–[22], whose isolates belong to hpEurope (7 sequences), hspWAfrica (1) and hspEAsia (1) (see Results), and no sequence data is available for the cagPAI in the other six populations and subpopulations where the cagPAI is present.
Here we analyze complete cagPAI sequences from 38 isolates representing all known H. pylori populations and subpopulations and compare their genetic polymorphisms with measures of functional expression. Our data show that the cagPAI has shared a long evolutionary history with the H. pylori core genome, and displays a remarkable global conservation of gene content, structure and function, with minor exceptions. We provide evidence that the cagPAI was acquired by ancestral H. pylori in a single event that occurred before modern humans migrated out of Africa. Sequence comparisons identified domains in multiple components of the t4ss that are likely to be under diversifying selection, and these findings can guide future research into the function of t4ss components.
In order to define the occurrence of the cagPAI in H. pylori, we screened a globally representative collection of H. pylori isolates from 53 different geographical or ethnic sources [15], [16] (Figure 1). 877 isolates were tested for the presence of the cagPAI by a PCR approach. Strains were classified as cagPAI-positive if we succeeded in separate PCR amplifications for the 5′ and 3′ ends of the cagPAI, or as cagPAI-negative if we succeeded in amplifying an empty site with primers from the flanking regions. The cagPAI was present in at least 95% of strains assigned to the hpAfrica1 (hspWAfrica plus hspSAfrica), hpEastAsia (hspEAsia, hspMaori) and hpAsia2 populations. In contrast, none of the hpAfrica2 strains possessed the cagPAI, and it was only variably present in strains from the populations hpEurope (225/330 strains; 58%), hpNEAfrica (58/72: 81%), and hpSahul (32/49; 65%) or the hspAmerind subpopulation of hpEastAsia (5/18; 28%).
Based on their multilocus sequence typing (MLST) haplotypes, seven strains with published cagPAI sequences belong to the hpEurope population (NCTC11638 from Australia [2]; 26695 from England [18]; and DU23, DU52, Ca52, Ca73 [20] and HPAG1 [21] from Sweden). J99 from the U.S.A. [22] belongs to hpAfrica1, and F32 [19] from Japan belongs to the hspEAsia population of hpEastAsia. None of these published cagPAI sequences were from strains of the hpNEAfrica, hpSahul, or hpAsia2 populations, from the hpEastAsia subpopulations hspAmerind or hspMaori, or from the hpAfrica1 subpopulation hspSAfrica, although those populations are also potentially important for our understanding of the evolutionary history of H. pylori. We therefore selected 29 strains from our global strain collection to supplement these nine published cagPAI sequences and provide a globally representative sample of cagPAI diversity (Figure 1). These strains included all known biogeographic populations, except for the cag-negative hpAfrica2. The entire cagPAI, approximately 37 kilobasepairs in length, was sequenced and annotated from each of the 29 strains, either after shot-gun cloning of overlapping long-range PCR products or via direct amplification of multiple, smaller PCR products.
The 38 complete cagPAI sequences were compared by pairwise sequence alignments and by a multiple alignment in Kodon relative to the cagPAI from J99 used as a scaffold sequence (Figure 2). The general pattern of gene content and gene order (signifying macrodiversity) was similar in most sequences, with only limited variation due to changed synteny or deletions. Synteny changes resulted from genomic rearrangements, horizontal genetic exchange (e.g. replacement of HP0521 by HP0521b), possibly in conjunction with IS (insertion sequence) element insertion, or gene inversions, such as for HP0535. Insertions, deletions, point mutations, frameshift mutations or disruption through insertion elements (Figure S1) were also observed in some of the cagPAI sequences, some of which should have resulted in pseudogenes. We therefore tested all strains for their ability to induce interleukin-8 (IL-8) in gastric epithelial cells (Figure 2, Figure 3), as an indicator of PAI function [23]. Most of the strains containing a cagPAI were able to induce IL-8, indicating that many of the mutations did not drastically reduce the general function of the cagPAI (Table 1).
Most new mutations are deleterious, whether associated with single nucleotide polymorphisms, mobile elements or genomic rearrangements, and will be removed by purifying selection. However, mutations without a drastic effect on fitness, so-called neutral or nearly neutral mutations, can remain as rare variants within a population for long time periods. The vast majority of such mutations remain at low frequency until they are (usually) lost due to genetic drift. Rare neutral mutations can become more frequent over time, or even become fixed, also due to genetic drift [24]. Still other mutations are under positive selection. These rapidly become frequent or fixed due to Darwinian selection. In isolated clonal populations, Muller's ratchet can even result in some deleterious mutations rising to high frequency [25] and the same is true of extreme bottlenecks, which can fix deleterious mutations immediately. These basic evolutionary principles indicate that the demographies of rare versus frequent mutations differ and should be examined separately.
A number of frequent cagPAI macrodiversity variants were found, some of which were present in all isolates of at least one sub-population, or almost all isolates (Table 1). These included insertion events due to one of three variants of IS606 [26] or of a mini-IS605 insertion [27], [28], an inversion of gene HP0535 plus its flanking non-coding DNA, a deletion of either the complete HP0521 ORF (Δ2; Figure 2) or part of that ORF, or the replacement of HP0521 by the unrelated ORF HP0521B (Figure 2, Table 1). Additionally, most of the 3′ (right) half of the cagPAI is lacking in all three hspAmerind strains due to one of two similar 11.2 kb deletions with distinct 3′ ends (Δ4, Δ5; Figure 2). These large deletions terminate within HP0546, and are associated with a second (intergenic) deletion of 410 bp or a 620 bp deletion that terminates within the N-terminal part of HP0547 (cagA). In strains V225 and HUI1769, a copy of the deleted segment plus the HP0546 and HP0547 ORFs have translocated to a separate, currently unidentified, location of the chromosome, leaving a shortened version of HP0546 at the original location (Figure 2). It is interesting to note that IL-8 induction was not eliminated by any of these frequent mutations (Figure 2, Figure 3, Table 1), suggesting that they are not deleterious to cagPAI function, and might be neutral or even under positive selection.
Rare variants were present in only one or two strains, are probably transient, and will tend to disappear during genetic drift [29]. The rare variants included frameshift mutations in multiple ORFs within three single isolates (CC42C, HPAG1 and L72) and IS elements (mini-IS605, IS605, IS606, IS607 or IS608 [26]) that have integrated at distinct locations in 7 other isolates (Table 1; Figure S1). Our dataset consisted of only 38 isolates, and it was possible that these rare mutations might be more widely distributed. We therefore screened 95 other globally representative strains for the presence of IS605, IS606, IS607 or IS608 at those locations, but only identified two additional strains with IS element insertions, one each for IS605 (MOR3055 – hspWAfrica) and IS607 (BASQ9523 – hpEurope) (data not shown). Thus, strains carrying these particular insertion mutations really are rare.
We also found two rare, distinct genomic rearrangements (Table 1). One of these was in strain NCTC11638 from Australia and has been reported previously [2]. It splits the cagPAI between ORFs HP0534 and HP0535 into two segments, one of which is translocated elsewhere in the genome, and is distinct from the split of the cagPAI in the hspAmerind strains. Previous analyses identified the same rearrangement in 4/40 strains from Italy [2], but it was not found in any of the other 38 cagPAI sequences analyzed here nor in any of the 95 other, globally representative strains that we investigated by PCR. The other rearrangement separated HP0547 (cagA) through HP0549 plus flanking DNA from the rest of the cagPAI. It has been previously described for two hpEurope strains from Sweden and one from Australia [20]. We found the same pattern in a fourth hpEurope strain isolated in Palestine (PAL3414). Both of these rearrangements were present in less than 5% of isolates.
The 17 rare mutations were identified in a total of 12 isolates. Only three of those, CC42C, HUI1692 and L72, did not induce IL-8, indicating that the majority of the rare sequence changes also did not cause a severe loss of cagPAI function. This observation is compatible with most of the rare mutations being selective neutral or near-neutral.
Three overlapping small deletions (Δ1, Δ2, Δ3) that removed the HP521 ORF were found in all but one hpEastAsia isolate, one hpEurope isolate and the hpSahul strain (Figure 2; Table 1), but those did not abolish cagPAI function (see above). Eight other deletions were found in four individual strains (Figure 2). Two of these isolates were unable to induce IL-8: CC42C (hspSAfrica) contains multiple frameshift mutations and an insertion of IS606 as well as deletion Δ11, which removes part of cagA (HP547). Δ4 and Δ6 deleted half of the cagPAI in hspAmerind strain HUI1692. The cagPAI is clearly decaying in both CC42C and HUI1692. In contrast, although deletions Δ5 and Δ7–Δ10 also removed large parts of the cagPAI in hspAmerind strains V225 and HUI1769, these deletions occurred in a segment that has been duplicated to a separate location (see above) and these two isolates remain able to induce IL-8. Thus, with one exception (Δ1), these deletions are rare and seem to be associated with accelerated decay of non-functional cagPAI genes. In addition, the cagPAI in non IL-8-inducing strain L72 also contained one frameshift and one premature stop codon in a coding region, and seems to be undergoing decay.
Darwinian selection for variation in coding regions can also be exerted at the nucleotide or protein level. We therefore analyzed sequence polymorphisms (microdiversity) in individual cagPAI genes for traces of such selection (Materials and Methods). Similar to housekeeping genes [30], almost all alleles of each cagPAI ORF were unique to one isolate among the 38 strains. Exceptionally, we identified duplicates of a single allelic sequence in six genes; in each case, the strains possessing the duplicate alleles were from a common population (Table S4). Occasional duplicate alleles within populations have also been described for housekeeping genes [30] and are considered to represent homologous recombination. Again, similar to housekeeping genes, most cagPAI genes seemed to be under purifying selection because their Ka/Ks ratios were ≤0.2 (Table 2). However, five genes (HP0534-0535, HP0538, HP0546-0547) showed signs of positive or diversifying selection because their overall Ka/Ks ratios were greater than 0.2; of these, cagA (HP0547) had the highest proportion of non-synonymous polymorphisms (Ka/Ks = 0.45). However, Ka/Ks ratios are relatively insensitive indicators of Darwinian selection, which can act at the level of single protein epitopes or conformational domains. We therefore used a Bayesian method (PAML/CODEML [31]) to search MLST and cagPAI genes for codons that might be under diversifying selection (indicated by ω >1). Only two of the seven MLST housekeeping genes (trpC, yphC) contained an appreciable frequency (3.9%; 5.3%) of codons with posterior probabilities of ω >1 being above 0.95 (Table 2). In contrast, >5.3% of the codons matched this criterion in 10 of the 28 cagPAI ORFs (Table 2), including four of the five ORFs with high overall Ka/Ks ratios (HP0535, HP0538, HP0546, HP0547).
We also tested eleven cagPAI ORFs, including nine with high frequencies of codons under selection according to PAML, and two with lower frequencies (HP0524, HP0525) with a second Bayesian program, OmegaMap [32], [33], which unlike PAML also takes into account the occurrence of recombination (ρ) between different alleles (Table S5). OmegaMap detected fewer codons with high probabilities of positive selection, but the codons that it identified often overlapped with codons that had been identified as being under positive selection by PAML (Table S5). Finally, we employed a sliding window along codons of PAML posterior probabilities of ω to identify clusters of sites with signs of diversifying selection (Figure 4). The combination of three forms of analysis (criteria: Ka/Ks >0.2, or likelihood of at least 95% for ω >1 in ≥5.3% of codons, or at least two clusters of two or more adjacent amino acids (aa) predicted under diversifying selection in PAML) identified 13 cagPAI genes that are likely to have evolved under diversifying selection: HP0520, HP0522, HP0523, HP0527, HP0528, HP0534, HP0535, HP0536, HP0538, HP0539, HP0540, HP0546 and HP0547. Of these, functions or structural contributions are known only for HP0523 (virB1), HP0527 (virB10), HP0539 (virB5), HP0546 (virB2) and HP0547 (cagA) [7], [34]–[38]. The percentage of codons with high likelihood of positive selection was highest in cagA (26.9%), followed by cagY (15.5%) and a gene of unknown function, cagQ (HP0535; 9.9%) (Table 2).
In addition to a high frequency of putative codons under diversifying selection, HP0527 (cagY) and HP0547 (cagA) also exhibited variable gene lengths. This was due to variable numbers of repetitive modules within the genes, as previously reported [35], [39]. In the CagA protein, the number of phosphorylation sites (C-terminal EPIYA repeat motifs) differed, as did the types of these repeats (Figure 3). As previously described [39], the third EPIYA motif of CagA was type D in most (13/17) Asian strains whereas type D was not found in isolates from any other population. This reflected the preponderance of type D EPIYA in isolates assigned to the hpEastAsia and hpAsia2 populations. If the EPIYA type D motif were ancestral in Asian populations, this finding might reflect horizontal acquisition of cagA by the four exceptional Asian strains from Western strains. Homologous recombination involving the cagPAI has also been reported in isolates from Mestizos in Peru [40] and might reflect selection due to functional differences that are related to ethnic specificity.
We next asked whether the phylogeny of cagPAI genes was similar to that of housekeeping genes. Concatenated sequences of the cagPAI genes yielded a tree (Figure 5B) that is very similar to the tree based on a concatenate of the seven MLST housekeeping genes (Figure 5A). Similarly, matrices of pairwise genetic distances of the concatenated cagPAI genes were highly correlated with corresponding matrices of pairwise distances of concatenated housekeeping genes (R = 0.65, p<0.001) (Figure 5C). These data show that 42% of the variance among cagPAI genes can be attributed to a linear relationship with housekeeping genes. The correlations for individual cagPAI genes ranged from R = 0.17 to R = 0.74 (Table 2). While most cagPAI genes thus fell into the range observed for the individual housekeeping genes (0.46 to 0.69), the correlations were lower for particular cagPAI genes (e.g. cagL, R = 0.17), which might reflect selection and/or recombination between cagPAIs from different bacterial populations. These observations indicate a generally similar genealogy of cagPAI and housekeeping genes, which would imply that the cagPAI has accompanied H. pylori since before human migrations out of Africa some 60,000 years ago [17]. In agreement, the genetic diversity of the cagPAI genes per population decreased significantly with distance from Northeast Africa (data not shown).
Only five of the strains tested here were not able to induce IL-8 (Figure 3). The same five strains did not translocate CagA into AGS cells, a second marker of t4ss function (Figure 3B). For three of the five strains (CC42, L72 and HUI1692), a lack of function can be explained by sequence features of coding sequence (CDS) decay. The cagPAI of CC42C contains multiple pseudogenes, some of which are crucial for t4ss function [3]. Half of the cagPAI including numerous essential t4ss genes is lacking in strain HUI1692. For strain L72, a point mutation results in a premature stop codon in gene HP0530, which is essential for t4ss function. In contrast, the cagPAI sequences did not offer obvious explanations for the lack of induction of IL-8 by strains M49 and D3a. We therefore investigated the transcript abundance of all 14 genes involved in IL-8 induction and of cagA for 28 sequenced strains as well as for the reference strains 26695A and J99 (Figure 3C; Table S3). The inability of strain M49 to induce IL-8 can be accounted for by very low transcript levels for 7/15 cagPAI genes (Figure 3C; Table S3); the cause of this low transcription is unknown. However, we are unable to explain the inability of strain D3a to induce IL-8, because it was not impaired in cagPAI transcription (Table S3). We are also not readily able to explain the considerable variation of transcript levels among the other strains that did induce IL-8 (Table S3), except that it did not correlate with the macrodiversity patterns described above (data not shown).
Similar to the variable transcript levels, the levels of IL-8 induction also varied dramatically (Figure 3). This variation did not correlate with strain assignments to biogeographic populations or with the type and number of EPIYA motifs within CagA (Figure 3A; [39]). Nor did they correlate with quantitative values for adhesion of the strains to AGS or MKN28 gastric epithelial cells (data not shown).
Since its discovery in 1996 [2], the cagPAI has probably been the most intensively studied segment of the H. pylori genome. The virulence functions of the Cag t4ss and its translocated effector, CagA, have been investigated in great detail, and numerous studies have correlated cagPAI-associated polymorphic markers with disease risk. However, all these studies focused on one or only few genes within the cagPAI (such as cagA), and were performed with strains from one or few geographic regions. We therefore anticipated that a comparative analysis of complete cagPAI sequences from a globally representative and well characterized collection of strains would provide valuable information about the evolutionary history of the cagPAI and its variability within a phylogeographic context. The complete cagPAI sequences of 29 strains were determined and combined with 9 published complete sequences to yield a large and comprehensive dataset of cagPAI diversity, which was analysed at the levels of both macrodiversity (differences in gene content, synteny and function), and microdiversity (sequence polymorphisms).
It has previously been noted from limited samples that different populations of H. pylori differ in the frequency of possession of the cagPAI [14], [17]. Our data on 877 isolates from all known H. pylori populations and subpopulations provide unambiguous evidence for this variability. Carriage of the cagPAI varies from almost universal presence in hpEastAsia and hpAfrica1 through intermediate presence (hpEurope) to complete absence (hpAfrica2) (Figure 1). The cagPAI is also absent in the related species H. acinonychis [17], which resulted from a host jump from humans to large felines [41]. The absence of the cagPAI from hpAfrica2 and H. acinonychis has been interpreted as the ancestral state, i.e. H. pylori acquired this genomic island by horizontal gene transfer from an unknown source after H. pylori had established itself in humans [17]. But when was it acquired, and on how many occasions?
The data presented here indicate that the cagPAI was only acquired once because its microdiversity correlated with microdiversity within housekeeping genes (Figure 5). That acquisition was prior to 60,000 years ago, the time when H. pylori accompanied modern humans during their migrations “out of Africa” [16], because cagPAI sequence microdiversity diminished with distance from North East Africa. An important implication of this conclusion is that, with the exception of hpAfrica2, the variable presence of the cagPAI in H. pylori populations usually reflects secondary loss, rather than inheritance of the ancestral virgin state.
Previous analyses have shown that strains that circulate within the same communities, and even within the same stomach, can be mixed in respect to possession of the cagPAI [30]. This observation indicates that cag positive bacteria do not outcompete cag negative bacteria in all environments. Nevertheless, our data support the inference [17] that a functional cagPAI provides a fitness advantage to H. pylori in most human populations: macrodiversity variants that inactivated t4ss function through deletions or insertion of IS elements were rare, whereas macrodiversity variants that were frequent did not affect t4ss function. For instance, shortening, complete loss or replacement (by HP0521b) of gene HP0521 was observed in almost all populations but this did not reduce cagPAI functionality, suggesting that this gene is not important for t4ss functions. Similarly, the genetic organization of the cagPAI was in general strongly conserved, and insertion elements did not play a decisive evolutionary role for the cagPAIs, unlike previous conclusions [2]. Even separation of the cagPAI in two parts did not lead to loss of function, except when a deletion was involved.
High variation at the level of sequence microdiversity was found along the cagPAI, but this is also true of housekeeping genes, and might possibly result from the high frequencies of mutation and recombination in H. pylori [14], [16]. However, unlike most housekeeping genes, multiple cagPAI ORFs showed signs of Darwinian diversifying selection, as indicated by higher Ka/Ks values and codon-based analyses, which identified specific amino acids or regions of particularly high non-synonymous diversity in 13 cagPAI genes (Figure 4, Table 2). In the following we attempt to interpret these measures of selection by mapping them onto known components including structural features of the t4ss encoded by the cagPAI.
Seventeen of the cagPAI genes are essential for the known t4ss functions (IL-8 induction, CagA translocation [3]), of which 12 have been characterized in structural or functional terms (virB1,2,4,5,6,7,8,9,10,11 and virD4 orthologs, cagA). In Figure 6, we present a schematic structural model of the cagPAI t4ss apparatus including all known structural Cag proteins plus the effector CagA. Different shades of grey indicate the proportion of amino acids which are likely to have undergone diversifying selection according to PAML.
The translocated effector protein CagA (HP0547), which interacts with various host proteins [42], had the highest proportion of such amino acids of the entire cagPAI. These were distributed along its entire length, suggesting functional adaptation or modulation. CagA binds to host cell integrins [42] and is translocated into host cells by the cagPAI t4ss. Within the host cell, individual domains of CagA interact with intracellular proteins such as SH-2 proteins and protein kinases (e.g. Src, Abl [19], MARK2/PAR1b kinase family [7], [9]). These interactions render it potentially subject to diversifying or positive selection due to host polymorphisms which could even result in modified host protein interactions. A prominent example of amino acid diversity noted previously are the EPIYA motifs in the C-terminal half of CagA, which differ between Asian (hpAsia2; hpEastAsia) (type D) and all other populations [43]. The D type EPIYA repeat binds SHP-2 phosphatase more avidly than other types [19]. A clear bipartite “Eastern”/“Western” separation in the present global dataset was not only observed in phylogenetic trees based on the C-terminal half of CagA containing the divergent EPIYA repeat motifs, but also in its less well-characterized N-terminal moiety. Interestingly, CagA from the ancient and isolated hpSahul population [15] localised in between the Eastern and Western type CagA clusters (not shown).
The global strain selection provided further evidence of functional adaptation in a different CagA motif. Recently, structural analyses of a second CagA subdomain (CM domain, aa 885 to 1005) in complex with its interaction partner from the human host, the cellular kinase MARK2, were performed [44]. This analysis revealed the crucial contribution of specific residues in CagA (MKI motif; [44]) to the physical interaction with the kinase. The short CagA peptide that could be mapped in the cocrystal (Phe948–Lys961) is characterized in our strain collection by high amino acid variability (Figure 7A and 7B). Superposition of the amino acids under selection (according to PAML) onto the structure of the peptide [44] revealed that all but five of the 14 amino acids in this MARK2 binding domain of CagA have a high posterior probability of being under diversifying selection (Figure 7A). Interestingly, Arg952 and Val956, which both strongly influence MARK2 binding [44], have a likelihood of 1.0 and 0.81, respectively, of being under positive selection whereas two other MARK2 binding residues, Leu950 and Leu959, were not under diversifying selection. This result suggests that, although some specific MARK2 binding sites in CagA do have a lower propensity of being under positive selection, the binding strength of CagA to MARK2 can still be influenced by H. pylori protein variation, indicative of functional fine-tuning. These predicted functional implications of global variation in the MKI motif are in agreement with an earlier study by Lu et al. [9] who observed differences in CagA PAR1b binding and function when they exchanged two Western and Eastern phylogeographic variants of the CagA MARK2/PAR1b binding region within CagA chimeras. We therefore expect that other regions of CagA that are under selection (Figure 4) also warrant detailed structural and functional analyses. The observed CagA diversity, which is proposed to allow functional fine-tuning, may not only be associated with different host ethnicities but also with niche-dependent intrahost diversification during long-term colonization (e.g. stomach antrum versus corpus) [45], [46].
A prior general comparison of component diversity in type III and IV secretion systems from different bacterial species [47] found that core structural proteins located in the bacterial cytoplasm or the inner membrane exhibit significantly lower diversity than do structural proteins exposed on the surface of the bacteria or secreted effector proteins [47]. Two well-characterized cag genes whose gene products are exposed on the cell surface have experienced strong selection: cagY (HP0527), which encodes a VirB10 ortholog that is a structural component of the cagPAI t4ss [36], and cagC (HP0546), which encodes a VirB2 pilin subunit ortholog [35], [38]. CagY is under selection due to host antibodies and/or direct host interactions [35], [36]. In cagC, those codons with the highest likelihood of diversifying selection (amino acids 21 to 42; Table S5) overlap with codons forming surface-exposed and highly strain-specific epitopes in the N-terminus of mature CagC [38]. The virB2 (HP0546) and virB5 (HP0539) orthologs of the cagPAI show signatures of diversifying selection in the present study; they encode surface-exposed pilin and pilus tip structural components of the Cag apparatus [48] and their sequence homology with functionally related VirB2 and VirB5 proteins from other bacteria is so low that they had to be identified by non-sequence-based approaches [37], [38]. We also find that 9 other cagPAI genes are under diversifying selection but their function is largely unclear. These include HP0520, HP0522 (part of the Cag outer membrane subcomplex [49]), HP0523 (cagγ; proposed to code for a virB1 orthologous peptidyglycan hydrolase [34], [50]), HP0528 (virB9), HP0534, HP0535, HP0536, HP0538 (encodes a membrane protein [50], [51]), and HP0540 [52]. Of these, HP0535 exhibits extensive non-synonymous variation and a clear bipartite Eastern-Western subdivision, similar to cagA. This gene is not involved in IL-8 induction or CagA translocation and is not predicted to possess a signal peptide. It may be a non-canonical secreted protein (score of 0.48 by SecretomeP). Based on the signs of selection and high diversity, we hypothesize that the HP0535-encoded protein interacts closely with CagA or is a novel effector protein that is translocated into host cells by the Cag t4ss. Of the other genes under diversifying selection whose function is unknown, HP0520 might be a non-canonical secreted protein because its SecretomeP score was also high (0.92).
In contrast to the genes just described, genes encoding cagPAI proteins that are not thought to be exposed on the bacterial surface [3] should be subject to purifying selection. In agreement with this expectation, other cagPAI genes including virD4 (HP0524) and virB11 (HP0525) orthologs [36], [50], displayed lower non-synonymous diversity and fewer codons under positive selection (Figure 6; Table S5).
In conclusion, the present work reports a genetic and functional approach within a global population genetic perspective to study diversity in a complex secretion system. This comprehensive library of data allowed the identification of genes with a high probability of having undergone diversifying selection. cagPAI genetic diversity is accompanied by modulations in functionality, but rarely by complete loss of function. Functional modulation of the t4ss appears to be an important feature in vivo and is predicted to rely not only on protein diversification but also on strain-dependent transcript level diversity in the cagPAI. These data will be a resource for future research on the biological roles and variable host interactions of individual cagPAI proteins. It will also foster research on the phylogeographic variability and evolution of determinants of host interaction in other microbes. The diversity in this dataset will also be useful to evaluating predictions by recent evolutionary models based on the structure of proteins, such as neutral networks of protein folds [53], [54]), which might be able to distinguish selection processes that favor structural versus functional conservation.
Bacterial isolates and sequences of seven housekeeping gene fragments (atpA, efp, mutY, ppa, trpC, ureI, yphC) have been described previously [13], [16], [55]. Strains were checked for the presence of the cagPAI by PCR, amplifying the 5′ (Primers O2872 + O2902) and 3′ (O2899 + O3326) flanking regions, or for absence (empty site) (primers O2872 + O3326). Primer sequences are provided in Table S1. Strains were chosen to represent all currently defined H. pylori populations possessing the cagPAI (Figure 1, Figure 2). The complete cagPAI was amplified for sequencing as two overlapping long range PCR products of ∼20 kb each with primers O2903 + O3048 and O3047 + O2904 (Table S1), respectively in 50 µl reactions with the EXL long range polymerase kit (Stratagene) using the following conditions: bacterial DNA 20 ng, Primers 20 µM each, 6 µl of 2mM dNTPs, 5 µl Buffer 1, 1 µl stabilizing solution, 1 µl EXL Polymerase, H2O to 50 µl. An initial denaturation for 1 min at 94°C was followed by 30 cycles of 45 sec at 94°C, 1 min at 65°C and 17 min 30 sec at 68°C. Long range PCR fragments were subjected to shotgun cloning. DNA fragments ranging from 0.8 to 1.2 kb were end repaired and cloned into the pGEM T-Easy vector (Promega), inserts were sequenced to 10-fold coverage by MWG Biotech. Alternatively, the cagPAIs were amplified as overlapping PCR products of ∼5 kb each with additional primers listed in Table S1 (primer combinations available on request) and sequenced with an extended set of primers (Table S1) by gene walking. The cagPAI sequence of strain PNGhigh85 was obtained by shotgun 454 sequencing of the whole genome (unpublished). Sequences were assembled with Gap4 (Staden Package, GCG Wisconsin). The individual cagPAI sequences have been submitted to the EMBL Nucleotide Sequence Database (accession numbers FR666825 - FR666857). Details for RNA preparation and RT-PCR are given in Text S1. RT-PCR primers and cycling conditions for transcript analyses of the cagPAIs are listed in Table S2.
CDSs were annotated in ACT and in KODON (Applied Maths BVBA, Sint-Martens-Latem, Belgium), automatic multiple sequence alignment of individual cagPAI genes was performed in BIONUMERICS (Applied Maths BVBA, Sint-Martens-Latem, Belgium) and corrected manually after visual inspection, where necessary. Sequence comparison and graphical output of multiple complete cagPAI sequences was performed in KODON. We only included one of eleven cagPAI sequences (F32) available from Japanese strains [19] because information is lacking on the phylogeographic population assignment of the remaining 10 strains. Pairwise genetic distances, phylogenetic trees and FST were calculated in MEGA3 [56] and in Arlequin [57], respectively. Pairwise geographic distances and distance from North East Africa (Addis Ababa, Ethiopia), as well as confidence intervals were calculated as previously described [16]. For analyses of increasing diversity with geographic distance from East Africa, the dataset was stripped of recent migrants [16] which resulted in the use of 33 out of the 37 cagPAI sequences. Pseudogenes were excluded from the dataset in all phylogenetic analyses.
Ks/Ka ratios were determined in DnaSP4.0 [58] and SWAAP, including a sliding window analysis. The number and location of potential codons under selection (ω) in each cagPAI gene were determined using the program CODEML in PAML 3.15 [59], implementing a sliding windows graphic representation. This software calculates the ratio of maximum likelihood of different evolutionary algorithms (models) for each codon (site) of a coding sequence to be under positive selection (ω>1), followed by Naive Empirical Bayes (NEB) and Bayes Empirical Bayes (BEB) analyses of posterior probabilities. Sites with a posterior probability P>0.95 by the CODEML codon substitution models M3 (discrete) or M8 (beta and ω) of ω>1 were considered as being under positive or diversifying selection. The likelihood of codons under diversifying selection in the presence of recombination was further analyzed using OmegaMap (V 0.5; [32]). This software uses a Bayesian modeling algorithm to calculate the probability of codons to evolve under diversifying selection (ω>1) in the presence of recombination (ρ). By explicitly modeling recombination, this method has a low rate to detect false positives. The settings used in the program were: norders = 100, thinning = 100, rhoprior = inverse, omegaprior = inverse, block length = 3 and 100,000 or 250,000 iterations. 5,000 iterations were deduced after each calculation as the burn-in phase. The model type used for both ω and ρ was “variable”. Three repetitions of the calculations with different settings were initially performed for control genes of defined structural properties and where some information is available about their function (e.g. HP0546), to exclude high variations in the calculations due to inadequate settings. Pseudogenes were excluded from the dataset.
Fragments of the housekeeping genes atpA, efp, mutY, ppa, trpC, ureI, and yphC were amplified and both strands were sequenced from independent PCR products as described [55]. Alternatively, comparable sequences were extracted from the published genomes (26695, HPAG1, J99). These sequences were assigned to populations and subpopulations by STRUCTURE [14].
IL-8 induction assay using the human gastric epithelial carcinoma cell line AGS (isolated from adenocarcinoma from a Caucasian patient) was performed for all strains of the sequencing project. Strain 26695A [60] was used as a reference. Cells were cultured in RPMI 1640 medium (buffered with 25 mM HEPES, supplemented with 10% heat-inactivated fetal bovine serum (medium and serum: Biochrom, Berlin, Germany). Details for bacterial culture conditions are given in Text S1. Cell infection experiments for IL-8 secretion measurement were performed on subconfluent cell layers (70%–90% confluence) in 24-well tissue culture plates. Cells were washed three times and preincubated in fresh medium with serum for 30 min prior to infection. By the addition of exponentially growing bacteria that were resuspended in cell culture medium (RPMI 1640, 25 mM HEPES, 10% heat-inactivated serum), the infection was started (MOI of 50). To synchronize the infection, the incubation plates were centrifuged at 500 x g, 20°C, for 3 min. The coincubation was carried out for 20 h. Non-infected cells (mock coincubated) were used as negative control. Supernatants were harvested, cleared of cell debris by centifugation, immediately frozen and stored at −20°C until use. Release of IL-8 into the cell supernatants was quantified by using BD OptEIA IL-8 enzyme-linked immunosorbent assay kit (BD Pharmingen; San Diego, USA) according to the company's instructions, using appropriate dilutions. The assays were performed in triplicate and the means and standard deviations of at least six independent coincubations were calculated. Adherence of the strains was tested in a high throughput assay, but no correlation was found between adherence and the IL-8 induction (data not shown).
To study CagA translocation, AGS cells were cultured in six-well plates and infected with H. pylori at a multiplicity of infection (MOI) of 100. After 4 h of coincubaction, non-adherent bacteria were removed by washing twice with PBS-Dulbecco (pH 7.4; Biochrom, Berlin, Germany). Cells were harvested with a cell scraper and resuspended in 1 ml PBS (pH = 7.4; Biochrom, Berlin, Germany). After centrifugation (250 x g, 4°C, 5 min), cells were resuspended in 300 µl of modified RIPA buffer (20 mM Tris-HCl [pH 7.5], 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM β-glycerol phosphate, 1 mM sodium orthovanadate, 1 protease inhibitor tablet per 10 ml buffer (Complete, Roche, Mannheim, Germany), 1 mM PMSF). During lysis, cells were incubated on ice for 30 min. Lysates were cleared by centrifugation (10 min, 21,900 x g, 4°) and the pellets were carefully separated from the supernatants. The pellet fraction was resuspended in 100 µl RIPA buffer and the fractions were immediately frozen at −80°C. To determine the amount of protein, a BCA protein assay was performed using the BCA Protein Assay kit (Pierce, Rockford, IL, USA) according to the manufacturer's instructions.
Equal amounts of cleared cell lysates (see above; corresponding to 10 µg of protein) of infected cells were resuspended in 5 x SDS loading buffer (0.31M Tris-HCl, pH6.8, 37.5% glycerol, 10% SDS, 0.05% bromophenol blue, 20% β-mercaptoethanol) and boiled for 10 min. For determination of molecular mass, BenchMark pre-stained Protein Ladder (Invitrogen, Karlsruhe, Germany) was used. Samples were separated on 10.4% denaturing SDS-polyacrylamide gels and transferred to nitrocellulose membranes (Protran BA 85, Whatman, Dassel, Germany) by semi-dry blotting. Membranes were blocked with 5% non-fat dried milk in TBS-T (20 mM Tris-HCl, 13.7 mM NaCl, 0.1% Tween 20, pH 7.4) for 1 h and subsequently incubated with specific primary antibody. Anti-CagA-antibody (Rabbit anti-H. pylori Cag antigen IgG fraction [polyclonal], Austral Biologicals, San Ramon, USA) was used at a dilution of 1/1,000 for the detection of CagA protein. To detect phosphorylated CagA, PY99-antibody (Santa Cruz Biotechnology, Heidelberg, Germany) was used (dilution 1/250). Goat-anti-Rabbit-HRP antibody (dilution 1/10,000, Jackson Immunoresearch Laboratories, Suffolk, Great Britain) or Goat-anti-mouse-HRP-antibody (dilution 1/5,000, Dianova, Hamburg, Germany) were used as secondary antibodies. Signal detection was performed with Enhanced SuperSignal West chemiluminescence substrate (Pierce, Rockford, IL, USA), and detection was on X-ray film (Hyperfilm, Amersham Biosciences, Buckinghamshire, UK).
|
10.1371/journal.pntd.0005271 | High Incidence of Human Rabies Exposure in Northwestern Tigray, Ethiopia: A Four-Year Retrospective Study | Rabies is a fatal zoonotic disease that has been known in Ethiopia for centuries in society as “Mad Dog Disease”. It is an important disease with veterinary and public health significance in the North western zone of Tigray where previous studies have not been conducted. Frequent occurrence of outbreaks in the area led the researchers to carry out a four year retrospective study to estimate the incidence of human rabies exposure in Northwestern Tigray, Ethiopia.
A referent study was conducted on human rabies exposure cases recorded from 2012 to 2015 at Suhul hospital, Shire Endaselase, Northwestern Tigray, Ethiopia. Exposure cases included in this research constituted victims bitten by unprovoked dogs and who received post exposure prophylaxis (PEP) at the hospital. Two thousand one hundred eighty human rabies exposure cases retrieved from the rabies case database were included in this study.
The majority of the exposed cases were males (1363/2180, 63%). Age wise, the most exposed age group was ≥15 years in all the study years: 166 (58%), 335 (65%), 492 (66%) and 394 (63%) in 2012, 2013, 2014 and 2015, respectively. Similarly, exposure cases for human rabies increased with age in both males and females across the study years. The incidence of human rabies exposure cases calculated per 100,000 populations was 35.8, 63.0, 89.8 and 73.1 in 2012, 2013, 2014 and 2015, respectively. Binary logistic regression analysis revealed that being male was a risk for human rabies exposure in all the study years.
The study discovered the highest annual human rabies exposure incidence in Ethiopia. This suggests an urgent need for synergistic efforts of human and animal health sectors to implement prevention and control strategies in this area.
| Rabies is a deadly disease of human and animals. The disease has been recognized in Ethiopia for centuries, but its impact remained underestimated. This limitation masks the true magnitude of rabies incidence and has been a stumbling block for its prevention and control. The aim of the study was to determine the incidence of human rabies exposure in Northwestern Tigray, Ethiopia, where studies have not been conducted, although outbreaks of the disease were common. A health facility-based four year referent study was conducted on human rabies exposure cases recorded from 2012 to 2015 at Suhul hospital, located in Shire Endaselase, Northwestern Zone of Tigray. Majority of the exposed cases were males. The most exposed age group was ≥15 years in all the study years. Similarly, exposure cases increased with age in both male and female individuals across the study years. Incidence of human rabies exposure cases per 100,000 populations also showed a continuous increment with the highest being recorded in 2014. Moreover, the study discovered the highest annual human rabies exposure incidence in Ethiopia. This suggests an urgent need for synergistic efforts of human and animal health sectors to implement prevention and control strategies.
| Rabies is a fatal and one of the most important reemerging zoonotic diseases throughout the world, caused by RNA viruses that affect the central nervous system of all warm-blooded animals, including humans [1, 2]. Carnivores are one of the primary virus reservoirs and rabid dogs pose the greatest hazard of rabies worldwide [3]. Transmission of the virus usually occurs by the bite of rabid animals. Under unusual circumstances, inhalation of aerosolized virus and organ transplantation from rabid patients may occur [4]. Rabies is incurable once the clinical signs of the disease appear [5]. The virus penetrates the body through wounds or by direct contact with mucosal surfaces, replicates locally, gains centripetal access to the peripheral and central nervous system via motor endplates and motor axons [6] and undergoes centrifugal spread to major exit portals, such as the salivary glands [3] and excretion in saliva [2] which contains abundant virus and is the main source of infection.
Globally, rabies is estimated to cause more than 1.9 million disability-adjusted life years (DALYs) and 6 billion in annual monetary losses [2]. Although effective vaccines are widely available for humans and animals [7], rabies remains the most deadly neglected disease in developing countries [8, 9]. In Africa, an estimate of the rabies burden in humans is about 23,800 deaths and 609,000 DALYs [2]. It is also a cause of substantial livestock losses [10] and is a constant threat to rare carnivores such as the Ethiopian wolf (Canis simensis) [11] and the African wild dog (Lycaon pictus) [12]. In Ethiopia, rabies has been known for centuries in society as “Mad Dog Disease” [13], and has been recorded scientifically since 1903 [14]. To date, rabies is an important disease in Ethiopia both in human and animals [5, 15–19].
Conducting regular epidemiological surveillance programs, designing and enforcing laws for registration, certification and regular vaccination of owned dogs, creating public awareness as well as provision of easily accessible, effective and affordable post exposure human vaccines could result in efficient prevention and control of the disease [20]. In Ethiopia, however, lack of comprehensive national epidemiological data on rabies in animals and humans [21] results in underestimating the disease, which is a stumbling block for its control and prevention. This masks the true magnitude of disease incidence and reduces the efficiency of the notification system as well as surveillance potential [22] and hence reduces the concern of policy-makers and funding agencies. Historically, the incidence of human rabies exposure in Ethiopia ranges from 1.3 to 18.6 per 100,000 populations [13, 17, 21, 23].
However, most of these studies are 20 years old, which typically represent national data, confined in and around the capital city [5, 15]. Severe under-reporting of human rabies cases as well as lack of record keeping of the disease were reported in Eastern Ethiopia [19]. The latest national surveillance data of human rabies exposure cases (national incidence of 3.4/100,000), collected from all regional states, indicated that the highest incidence was registered in Addis Ababa (14.5/100,000), followed by Tigray (12.6/100,000) and Oromia (3.8/100,000) [21].
There is lack of scientific information on the status of human rabies and little is known about the disease to apply effective control measures in Northwestern Tigray. Frequent occurrence of outbreaks of human rabies and lack of documented information regarding the epidemiological situation of the disease led us to conduct this study. The objective of the research was to estimate the incidence of human rabies exposure in the Northwestern administrative zone, Tigray National Regional state, Ethiopia.
A health facility-based study was conducted from May to July, 2016 at Suhul hospital, located in Shire Endaselase, the capital city of the Northwestern Zone of Tigray (Fig 1). Shire Endaselase is located 1,087 km north of Addis Ababa, the capital city of Ethiopia. Suhul hospital is the only referral hospital that manages rabies cases and is the center for administration of post exposure prophylaxis (PEP) in Northwestern Tigray. The hospital accommodates patients from Shire and Sheraro towns as well as other six Woredas in the zone: Medebay Zana, Tahtay Koraro, Laelay Adiabo, Tahtay Adiabo, Tselemti and Asgede Tsimbla. The altitude ranges from 645 to 2852 meters above sea level. The total area is 18,325.1 km2. It receives an annual average rainfall of 877.6 mm, mainly from June to September, and temperature ranges from 18 to 34.6°C [24]. The main vegetation includes deciduous acacia species, Eucalyptus and riparian trees. Forest coverage of the area is estimated to be 40.5% (Zekarias, Agricultural advisor of Northwestern zonal administrator, personal communication, 2016). The Northwestern zone of Tigray is known for its livestock production potential. The Kafta Sheraro national park is in the study area, where 42 mammalian and 95 avian wildlife species reside. Elephants, spotted and striped hyenas, wild and serval cats, jackals, foxes, red fronted gazelle and greater kudu are common in the area.
The study was conducted on human rabies exposure cases who registered from January 1, 2012 to December 31, 2015 for PEP. For convenience, we included only these four years because data of earlier years were not recorded appropriately and not easily accessible. Retrospectively, registered human rabies exposure cases of these four-year data were collected from the hospital’s annual disease report database. Moreover, discussion was held with the disease surveillance and control/prevention coordinator of Suhul hospital. Patients included in this study were kept anonymous to protect their medical confidentiality rights. Information retrieved comprised factors such as patient age and sex; species and vaccination history of rabies-suspect animals; history of exposure and previous rabies vaccination of the human cases; and recommended PEP for individuals. Two research team members retrieved the data from the hospital’s annual disease report database.
Prior to PEP administration for humans, a decision was made on suspect dogs which inflicted unprovoked bites through quarantine and follow up for 10 days, usually at the owner's premise and sometimes at the hospital compound. The dogs were diagnosed clinically as stated in the Standard Treatment Guidelines for General Hospitals [25] that unvaccinated dogs which cause unprovoked bites must be suspected to be rabid. The underlying principle is that rabies virus-infected animals could only transmit the virus shortly before and after clinical signs have developed. After quarantine, the dog might have either displayed a significant change in behavior or signs of illness suggestive of rabies and/or died during the observation period. Then, bite victims may have adequate time to receive PEP and prevent the disease development. By using the above principles, dogs suspected as rabid during the observation period have been positive for rabies virus by laboratory diagnosis [5]. ‘Suspected’ refers to a case that is compatible with a clinical case definition [3]. Unfortunately, confirmation of suspected dog brain samples by laboratory diagnosis was not practiced in the hospital due to a lack of facilities.
After clinical diagnosis of the dog, the type of exposure was identified and human PEP was recommended. To identify the type of exposure, the definition of exposure used by the hospital was consistent with the European Centre for Disease Prevention and Control (ECDC) recommendation [26] and included both category II (minor exposure) and category III (severe exposure) exposures. Category II refers to minor scratches or abrasions without bleeding or licks on broken skin and nibbling of uncovered skin, while category III refers to single or multiple transdermal bites or scratches by infected animals. To collect history of human exposure, the hospital used a similar format developed by WHO [2] for cases of possible rabies exposure (S1 Text). A final decision for PEP was made based on history of exposure, typical clinical signs of the disease in dog and/or its death. When victims were bitten by stray dogs, it was difficult to follow up and the persons received PEP depending on the type of exposure. Exposure cases included victims bitten by unprovoked dogs and who received a complete PEP course at Suhul hospital. The human populations at risk (for the respective years, 2012–2015) of this study were estimated by projecting the results of the National Population and Housing Census of Ethiopia conducted in May 2007 [27].
Ethical approval was obtained from Aksum University, Research and Ethical Review Committee (S2 Text). Permission was sought from the hospital administration before data collection. Moreover, an official letter was issued to Suhul hospital that the findings would be used for scientific purposes.
Data were entered, checked and analyzed using STATA statistical software (version 11.0,Stata Corp, college station, Texas 77845 USA). To ensure quality, data were entered and cross-checked independently by two members of the research team. Age, sex and time (year) based distribution of human rabies exposure cases was analyzed using descriptive statistics. With 95% confidence intervals, crude odds ratio was employed to compare the association between the outcome (human rabies exposure) and potential predictor variables (male and female) using binary logistic regression model. A P-value < 0.05 was considered statistically significant.
In total, 2180 human rabies exposure cases were registered and followed for their PEP at Suhul hospital from 2012 to 2015. The results showed that the number of human rabies exposure cases sharply increased in the study years with the highest being recorded in 2014. Dog bite was the only described cause of human rabies exposure in the study area.
Demographic data of the registered human rabies exposure cases (n = 2180) originated from two towns and six other administrative divisions (e.g., Woredas) of Northwestern Tigray were obtained and included in the study. Most of the exposed individuals were males (1363/2180, 62.5%). In all the four years, human rabies exposure cases were higher in males than females (Table 1). In the hospital rabies database, victim age was obtained clustered in three categories: ≤ 4, 5 to 14 and ≥15 years. Accordingly, the greatest exposed age group was ≥15 years in all the study years: 166 (58%), 335 (65%), 492 (66%) and 394 (63%) in 2012, 2013, 2014 and 2015, respectively. Similarly, it was observed that exposure for human rabies increased with age in both males and females across the four years (Table 1).
The results showed that human rabies exposure cases increased across the study years with the highest recorded in 2014. The four-year trend of human rabies exposure cases is shown in Fig 2. The incidence of human rabies exposure cases calculated per 100,000 populations was 35.8, 63.0, 89.8 and 73.1 in 2012, 2013, 2014 and 2015, respectively. Binary logistic regression analysis revealed that being male was a risk for human rabies exposure in all study years (Table 2).
Data on the coverage of preventive dog vaccination and demography were not evident in the study area. In this study, however, it was confirmed that all dogs inflicted unprovoked bites were unvaccinated. Moreover, inadequate supply of preventive dog vaccination was observed in the study area and was irregular in application. Victims who visited the nearest health centers were always sent to Suhul hospital and allowed to get the recommended doses of PEP. The PEP doses given, for category II and III exposures, were 14 doses for the first 14 days, consecutively, and then 3 doses at 10 days interval (i.e. 24th, 34th and 44th days). No rabies immune globulin (RIG) was administered. The attenuated Fermi vaccine produced by the Ethiopian Health and Nutrition Research Institute, Addis Ababa was used for PEP. Besides vaccine, standard wound management and administration of Tetanus antitoxin (TAT) was practiced. According to the manager of the hospital, Northwestern zone of Tigray was among the areas where rabies was the most prevalent. He also added that out of the total annually allocated PEP to Tigray National Regional State, nearly 60% was utilized by Suhul hospital.
Rabies is an entirely preventable infectious disease if regular epidemiological surveillance programs are practiced accompanied with enforcing laws for dog registration, vaccination and certification, and provision of easily accessible, effective and affordable vaccines and RIG [20]. It is an important disease with significant public health concern in the Northwestern zone of Tigray region. Human rabies exposure cases (n = 2180) registered for and followed up their PEP at Suhul hospital during the respective study years (2012, 2013, 2014 and 2015) showed a sharp increase with the highest recorded in 2014. This observation might be either from the improvement of the coverage of dog vaccination or under reporting conditions occurred in 2015. Binary logistic regression analysis revealed that being male was a risk for human rabies exposure in all the study years. This might be associated with activities of males, in that they are engaged in outdoor activities while females are more likely to remain indoors due to sociocultural and religious reasons. The same findings were reported in other locations of Ethiopia [5, 17] and Nigeria [28]. Age wise, most of the cases were found in people aged 15 years old and above (Table 1). The result was discordant with study conducted at Gondar, Ethiopia [17]. This might be explained by the wide range of clustering age. This suggests the extent of the impact of the disease on the national economy by affecting the most productive age group.
The incidence of human rabies exposure cases calculated per 100,000 populations was 35.8, 63.0, 89.8 and 73.1 in 2012, 2013, 2014 and 2015, respectively. This incidence was much higher than previously reported studies undertaken in Gondar areas, Ethiopia [16, 17]. So far, the incidence of human rabies exposure cases were commonly estimated from health facility records based on provision of PEP, which showed increases in the number of recorded human rabies exposure cases from 1986 onwards, ranging from 1.3 to 18.6 per 100,000 populations [13, 17, 21, 23]. The high human rabies exposure incidence of the present study reveals that the number of human rabies exposed individuals visited health facilities has increased. This might be due to growing awareness of the communities in the study area and/or increased risk of the disease resulted from ecological changes in the area.
However, recent studies conducted on rabies in different settings of Ethiopia showed that community awareness (knowledge) towards the disease was improved despite medical care seeking behavior was low. For instance, immediate follow up modern medication after bite and vaccination of dogs are not satisfactory. In addition, strong believe in traditional medicine for the treatment of rabies has been discovered. This indicates low level of community attitude towards the disease [16, 18, 29, 30]. In Eastern Ethiopia, one study confirmed that majority of the respondents have heard about the disease from their family in both urban and pastoralist households [19] which implied government based awareness creation might not be adequate or practiced. In the same study, overall poor knowledge about the disease has been reported in pastoralists. It has been demonstrated that investment in mass vaccination of owned dogs [8] and control of stray dogs [31]) are the most effective way of reducing the burden of rabies in humans. This is because canine rabies causes around 59,000 human deaths globally [8] and 95.6–96.6% of deaths from rabies occur in Asia and Africa, where canine rabies is enzootic [2]. In the present study, dog bite was the only source of rabies exposure. Similarly, 92% of humans who received post exposure anti-rabies treatments in Ethiopia were due to dog bites [15]. Likewise, other studies reported dog bite was the most important sources of human rabies [5, 16, 17, 29, 30].
In Africa, owner charged vaccination coverage is generally low though the percentage of dogs vaccinated under free of charge vaccination schemes is high [32]. According to the WHO recommendation [20], 70% of the dog population must be regularly vaccinated to control rabies. In Ethiopia, however, data on dog demography is not yet established [5]. Besides, the coverage of anti-rabies dog vaccination was not sufficient to put the disease under control [5, 21]. In the Awash Basin of Eastern Ethiopia, preventive dog vaccination was non-existent due to lack of availability of the vaccine [19]. The supply of preventive dog vaccination was also inadequate and irregular in the study area. This might signify that rabies, the tropical neglected zoonotic disease, has also not given attention in Ethiopia.
In the present study, only one fatal case was recorded at Suhul hospital in 2014 which seems to be insignificant. However, this was hospital based study and it is assumed that considerable numbers of the victims might have not visited health care posts because of the socio-cultural influences. Tschopp et al. [19] reported that individuals exposed to rabies could not visit health centers due to distance to health facility, mistrust in the medical system and poor knowledge about the disease. Thus, it is believed that the burden of rabies may be beyond the reported incidences because active surveillance has rarely been conducted. In Tanzania, it has been predicted that the incidence of human rabies, on the basis of active surveillance is 100 times greater than that of officially recorded [33]. In Ethiopia, severe under-reporting of human rabies cases as well as lack of record keeping of the disease were reported [19].
In Ethiopia, rehabilitation and biodiversity protection and management are among the strategies designed for poverty reduction [34, 35]. The study area (Northwestern Tigray) is known for its extensive uncultivated land with considerable forest coverage, which is estimated to be about 40.5% (Zekarias, agricultural advisor of Northwestern zonal administrator, personal communication, 2016). In Kafta Sheraro National Park wildlife might be reservoirs of the virus, as suggested in the Ethiopian Health and Nutrition Research Institute proceedings [21]. Due to wildlife concerns, most farmers/residents prefer to have more than one dog to guard their house. Opportunities for spillover infection at the wildlife-livestock/pet animal interface are a common phenomenon. This may facilitate circulation of viruses in the area. Wildlife are often believed to play a major role in transmission [2, 21]. In Ethiopia, one study identified that domestic dogs and cattle exhibiting clinical signs consistent with rabies, and people bitten by suspected rabid dogs, were reported in communities adjacent to Bale Mountains National Park, when rabies outbreak had occurred in wolves of this park [11]. Dog and livestock vaccination would reduce this concern for all species.
This study incorporated data of human rabies exposure case records inflicted by suspected dogs which had clinical information of their signs of illness suggestive of rabies. Nevertheless, sample submission for confirmation was not undertaken and the study was based on retrospective data. Nonexistence of laboratory confirmed cases in the hospital’s record was due to lack of facilities in the region. Moreover, the study did not include suspect or probable human rabies cases. Despite having these limitations, we believe that exposed humans (with exposure category II and III) had a high probability of developing rabies as all suspected dogs were not vaccinated. This is supported by research findings discovered at the Ethiopian Health and Nutrition Research Institute, in that dogs confirmed as suspect cases during the observation period have been found positive for rabies virus by laboratory findings [5]. In addition to the lack of rabies confirmation in suspect animals, no RIG was administered in situation of category III exposures, in conflict with current WHO recommendations for human rabies PEP [2].
In conclusion, the present study showed that human rabies exposure has shown a continuous increase in Northwestern Tigray across the study years (2012–2015) with the highest recorded in 2014. Moreover, the study discovered the highest annual human rabies exposure incidence in Ethiopia. This suggests an urgent need for synergistic efforts of human and animal health sectors to implement prevention and control strategies in this area.
|
10.1371/journal.pntd.0001271 | Male Mating Competitiveness of a Wolbachia-Introgressed Aedes polynesiensis Strain under Semi-Field Conditions | Lymphatic filariasis (LF), a global public health problem affecting approximately 120 million people worldwide, is a leading cause of disability in the developing world including the South Pacific. Despite decades of ongoing mass drug administration (MDA) in the region, some island nations have not yet achieved the threshold levels of microfilaremia established by the World Health Organization for eliminating transmission. Previously, the generation of a novel Aedes polynesiensis strain (CP) infected with an exogenous type of Wolbachia has been described. The CP mosquito is cytoplasmically incompatible (i.e., effectively sterile) when mated with wildtype mosquitoes, and a strategy was proposed for the control of A. polynesiensis populations by repeated, inundative releases of CP males to disrupt fertility of wild females. Such a strategy could lead to suppression of the vector population and subsequently lead to a reduction in the transmission of filarial worms.
CP males and F1 male offspring from wild-caught A. polynesiensis females exhibit near equal mating competitiveness with F1 females under semi-field conditions.
While laboratory experiments are important, prior projects have demonstrated the need for additional testing under semi-field conditions in order to recognize problems before field implementation. The results reported here from semi-field experiments encourage forward progression toward small-scale field releases.
| Aedes polynesiensis is the primary mosquito vector of lymphatic filariasis (LF) in the island nations of the South Pacific. Control of LF in this region of the world is difficult due to the unique biology of the mosquito vector. A proposed method to control LF in the Pacific is through the release of male mosquitoes that are effectively sterile. In order for this approach to be successful, it is critical that the modified male mosquitoes be able to compete with wild type male mosquitoes for female mates. In this study the authors examined the mating competitiveness of modified males under semi-field conditions. Modified males were released into field cages holding field-collected, virgin females and field collected wild type males. The resulting proportion of eggs that hatched was inversely related to the number of modified males released into the cage, which is consistent with the hypothesized competitiveness of modified males against indigenous males. The outcome indicates that mass release of modified A. polynesiensis mosquitoes could result in the suppression of A. polynesiensis populations and supports the continued development of applied strategies for suppression of this important disease vector.
| Lymphatic filariasis (LF) is a mosquito-borne disease that can lead to gross disfigurement (lymphedema and elephantiasis) and disability. In addition to the severe pain that often accompanies LF, many affected individuals suffer psychological distress due to associated social stigmas. In severe cases, individuals may become physically incapacitated [1]. Thus, filariasis can place a significant socioeconomic burden on individuals, communities, and healthcare systems [2].
Because there is currently no vaccine available for LF, current control of this disease is based on the regular administration of anti-filarial compounds to the entire at-risk population. Although the drugs do not kill adult filarial worms, they are microfilaricidal and decrease the level of infectious larvae in the blood. In theory, mass drug administration (MDA) campaigns that last longer than the fecund life span of adult filarial worms, or approximately five years [3], should eliminate LF altogether. However, experience has shown that the strategy can be complicated within some systems [4].
The South Pacific region has a longstanding history of public health campaigns directed toward the control of filarial transmission, including some areas that have practiced mass drug administration since the 1950′s [5]. In the case of Maupiti, a small, relatively isolated island in French Polynesia, low-level transmission persists despite more than three decades of MDA [6], suggesting that MDA alone may be inadequate for the elimination of LF in some areas. In such cases, integration of complementary vector control strategies may be required [4].
Throughout much of the South Pacific, the primary vector of human filariasis is Aedes polynesiensis, a mosquito that exhibits higher transmission efficiency when microfilaremia is low [7], [8]. This pattern of negative density-dependent transmission has been hypothesized to contribute to the inability of MDA to eliminate LF in some regions of the Pacific. Control of A. polynesiensis is difficult because the mosquito is exophilic and breeds in both artificial containers and natural sites, such as tree holes, crab burrows, shells, and leaves [9], [10]. Multiple attempts have been made to control A. polynesiensis, using a variety of measures, including the competitive replacement of A. polynesiensis with a refractory species A. albopictus on the atoll of Taiaro [11]. Additional control strategies based on the manipulation of vector breeding site have utilized polyester beads, larvivorous fish Gambusia affinis and Poecilia reticulata, and the copepod Mesocyclops [12] as well as land-crab burrows baits [10], [13], [14]. These efforts have been met with limited success [4], [10], [15] as the wide range and number of available breeding sites, coupled with the often rugged and inaccessible terrain of South Pacific island nations, makes it unfeasible to sustain vector control strategies across the numerous, widely dispersed islands.
An additional strategy for A. polynesiensis control in the Pacific is a variation of Sterile Insect Technique (SIT). SIT is based upon the release of sterile males in order to suppress and eliminate an insect species. A frequently noted example is sterile male releases used to eliminate the screwworm fly, Cochliomyia hominovorax, from the United States, Mexico, Central America and Curacao in the 1950′s [16], [17], [18]. Weekly releases of 40 million sterile males are ongoing in Panama, to prevent the reinvasion of C. homnivorax from South America [19].
Success with the screwworm encouraged research into the broader use of SIT in additional insects of both economic and medical importance. The technique has been successfully employed in the eradication of the melon fly, Bactrocera cucurbitae, from Southwestern Japan [20], [21] as well as for the eradication the tsetse fly, Glossina austeni, from Zanzibar [22]. SIT is also a critical component in controlling and eliminating the Mediterranean fruit fly, Ceratitis capitata (Wiedemann), from California [23].
The earliest attempt at employing SIT for mosquito control was made between 1959-1961, when the USDA used ionizing radiation to sterilize Anopheles quadrimaculatus pupae [24]. Subsequent attempts by a variety of researchers involved irradiation of pupae from Aedes aegypti [25], Culex quinquefasciatus [26], [27], and Culex tarsalis [28]. These attempts were met with limited success, in part because the process of irradiation affected the male fitness in terms of locating and mating with wild females [29]. Recent efforts at radiation-based SIT of mosquitoes is focused on the release of sterilized Anopheles arabiensis in the Sudan [30], [31] as well as on the irradiation of Aedes albopictus in Italy [32].
As an alternative to irradiation, several control programs based on chemical sterilization of male A. aegypti, Cx. quinquefasciatus, and Anopheles albimanus were initiated in the 1970′s and 1980′s [33], [34], [35], [36]. However, these have not been extended, primarily due in part to environmental concerns associated with residual chemosterilants on the released mosquitoes [35].
Incompatible Insect Technique (IIT) is similar to SIT, but IIT relies upon embryonic lethality resulting from cytoplasmic incompatibility (CI) induced by the maternally transmitted intracellular bacterium Wolbachia pipientis [37]. Wolbachia is a naturally occurring endosymbiont of arthropods that renders mosquitoes reproductively incompatible when mated to individuals with a differing infection type [38]. Because it does not rely on modifying the males through irradiation or chemical treatment it may avoid some of the male fitness problems associated with SIT programs in the past [29]. The IIT approach provides a relatively rare example of a successful mosquito field trial, when Laven used this technique to successfully eliminate Culex pipiens fatigans from a region of Burma in 1967 [39].
In 2008, Brelsfoard et al proposed a vector control strategy for the South Pacific that is based on IIT [37]. Field surveys to date have shown that natural populations of A. polynesiensis are infected with a single Wolbachia type [40], [41], [42]. An artificially infected A. polynesiensis strain (CP) was generated by introgressing a Wolbachia type from Aedes riversi into the A. polynesiensis genotype. Laboratory tests demonstrated that the CP strain was bidirectionally incompatible with naturally infected mosquitoes. Subsequent tests also demonstrated that CP and wild type males exhibited near equal mating competitiveness under laboratory conditions [43].
Previous SIT programs have repeatedly demonstrated the importance of confirming laboratory results within field conditions prior to large scale implementation [44]. Specifically, laboratory strains typically have lower relative fitness compared to wild type mosquitoes, and the difference in fitness may not become apparent until the mosquitoes are moved from the stable environment of the laboratory and placed under more natural conditions. Here, we describe male mating competitiveness assays between CP and wild-type males, performed under semi-field cage conditions.
Two mosquito strains were compared in this study: the bidirectionally incompatible CP strain [43], which has been maintained in the laboratory for over twenty generations and the wild-type A. polynesiensis Atimaono strain (APA). APA was collected from a coconut grove in Atimaono, Tahiti (17°46′41.44"S 149°27′14.23"W). In order to minimize the effects of laboratory maintenance, eggs from field-collected APA females were reared. The resulting F1 APA adults were used for experiments.
To minimize differences caused by immature rearing conditions, CP and APA were reared under identical laboratory conditions. Larvae were maintained on a 60 g/L liver powder solution (MP Biomedicals LLC, Solon, OH). Adult mosquitoes were maintained on 10% sucrose. Ambient temperature ranged from 23–31°C. Relative humidity was maintained at or above 80% using a humidifier.
Field cages were 223×127×102 cm tents (Aura, Marmot Mountain LLC, USA) placed on a platform with its legs in a water moat were used to prevent ants from entering the field cages. Each field cages was covered by a 365×275×215 cm screen house (Ozark Trail, Model WMT-1290S, USA). The two-cage design was employed in order to reduce the potential for accidental escape of laboratory-reared mosquitoes or the accidental introduction of wild mosquitoes. Mosquitoes observed inside the external screen house were killed before opening the inner cage. A 3×5 m tarpaulin was suspended over each field cage, to provide shading and protection of cages (e.g., during periodic heavy rainfall). Each field cage contained a black plastic flowerpot as a resting area and containers of a 10% sucrose solution as a carbohydrate resource. A Hobo data logger (U12-012, Onset Computer Corp., USA,) was placed within cages to record temperature, relative humidity and light intensity. Rainfall was monitored using a rain gauge located within 500 meters of the cages.
The field study was conducted on the campus of the Institut Louis Malardé, Paea-Tahiti, French Polynesia. Field cages 1–3 were placed under a Hibiscus tiliaceus canopy, while field cages 4–6 were surrounded by Wedelia trilobata, Spathodea campanulata, Citrus, Musa, and Acacia. The natural vegetation provided protection from direct sunlight and heat. However, some of the surrounding plants, especially the W. trilobata growing underneath and the vines growing directly on the platforms, were trimmed periodically to prevent ants from accessing the cages.
In order to ensure virginity of adult mosquitoes, pupae from each strain were individualized into 10 ml tubes with water and allowed to eclose. Mosquitoes were sexed at the adult stage. Individuals were then released into 30.5×30.5×30.5 cm cages (Cat. No. 1450; Bioquip Corp., USA). Males and females were held in separate cages in the laboratory until sexually mature. At the time of release, males were approximately 48 hours post-eclosion, and females were approximately 24 hours post-eclosion. Fifty virgin APA females and fifty virgin males were released into cages.
For the mating competitiveness trials, two experimental designs were performed. The first design (Experiment A) compared three APA:CP male ratios (0∶50, 25∶25, 50∶0). Experiment A was performed on two different days and each treatment was repeated in two different tents on each of those days (4 treatment replicates). The second design (Experiment B) compared five APA:CP male ratios (0∶50, 12∶38, 25∶25, 38∶12, 50∶0). Experiment B was performed on three different days and each treatment was represented once on each of those days (3 treatment replicates). In both experiments the female:male sex ratio was 50∶50. For both designs, the different treatments were randomly assigned to different cages, to avoid a potential bias due to environmental variation between cage locations. Males were released into field cages first, followed by virgin females.
Twenty-four hours after releasing mosquitoes into field cages surviving mosquitoes were removed from cages using a backpack aspirator (Model 1412, John Hock Co., USA), and male and female mortality was recorded. Males were separated from females to avoid subsequent mating events, and both sexes were placed into separate Bioquip cages as described above and held in the insectary.
Female mosquitoes were blood fed on laboratory mice (Mus musculus) at the Institut Louis Malardé (Tahiti, FP), with the authorization of the “Commission permanente de l'assemblée de la Polynésie Française (Tahiti)” [Deliberation N°2001-16/APF] and in accordance with French regulations. Engorged mosquitoes were individualized into oviposition cups and provided with a sugar source. Following embryonation, eggs were hatched by flooding and placed under a vacuum for one hour. Hatch rates were determined by examining eggs using a Leica EZ4D dissecting microscope (Leica Microsystems GmbH).
Females that produced egg batches resulting in less than ten larvae were dissected in Ringer Lactate B. Braun buffer (B. Braun Medical SA, Spain), and the insemination status was determined by direct visualization of sperm in the spermathecae under 60x magnification using a Leica Diaplan compound light microscope (Leica Microsystems GmbH, Germany). In addition to qualitative insemination status, the number of inseminated spermathecae was also recorded. Broods in which more than 10% of the eggs produced larvae were considered to be from a compatible cross.
Male mortality and egg hatch data were arcsine transformed, and Analysis of Variance (ANOVA) was used to compare male mortality for the different treatments. Male mating competitiveness was analyzed using a Chi-square goodness of fit test to compare observed and expected numbers of hatching broods for each APA:CP male ratio. An additional estimate of CP male competitiveness was determined using the method of Fried [45]. Briefly, this statistic is derived through the equation
Where Ha = the % egg hatch of normal (N) males x normal females, E = the % egg hatch of a mixed ratio of normal and sterile males, and Hs = the % egg hatch of sterile (S) males x normal females.
Egg batches laid by non-inseminated females were excluded from the analysis. Egg hatch data was compared using the Kruskal-Wallis test. Pairwise comparisons for egg hatch between treatments were performed using Wilcoxon Rank Sum test with Bonferroni correction. All statistical tests were performed using JMP 8.0.1 (SAS Institute, Cary, NC).
This study was conducted over the course of 3 months (April-June) in 2009 on the island of Tahiti, French Polynesia. Temperatures during the course of the experiments ranged from 21–33°C. The percent relative humidity ranged from 64–97%. Precipitation levels were negligible during the study except for the final replicate of experiment B, when 29.3 mm of precipitation was recorded.
The mean 24-hour mortality by cage treatment for male mosquitoes in Experiment A and experiment B is shown in Table 1. There was no significant difference in male mosquito mortality between the three treatments in experiment A (ANOVA; P>0.05) or between the five treatments in experiment B (ANOVA; P>0.05). The pooled mean 24-hour male mortality for experiment A was 7.7%±0.5 % (SEM) while the pooled mean 24-hour mortality for males in Experiment B was 7.1%±0.3% (SEM).
Assuming equal mating competitiveness, one would expect the proportion of females mating with CP males, and therefore producing inviable egg broods, to equal the proportion of CP males present. Figure 1 illustrates that no significant difference was observed between the expected and observed number of hatching broods for any of the treatments in Experiment A (Chi-square; P>0.75). The observed brood hatch rate decreased from 91% to 1%, inversely proportional to the number of CP males present (R2>0.99). Again in Experiment B, no significant difference was observed between the expected and observed brood hatch rates (Chi-square; P>0.10; Figure 1B). The observed brood hatch rate decreased from 72% to 4%, inversely proportional to the number of CP males (R2 = 0.96). The mean competitiveness value (C) for CP mosquitoes was calculated for both experiments, at 0.84±0.04 and 0.92±0.48 for Experiments A and B, respectively.
Broods were considered compatible when hatch rates were greater than 10%. No significant difference was observed in egg hatch rates from compatible broods (i.e., hatching) of the three treatment groups from Experiment A (Figure 1A; P = 0.07) nor from the five treatment groups from Experiment B (Figure 1B; P = 0.18). Compatible broods were further stratified into two groups; those with an intermediate hatch rate (11%–69%) and those with a high hatch rate (>70%). There was no significant difference observed in egg hatch rates for broods from the 5 treatments in the intermediate category from experiment B (Figure 2; P = 0.16) nor was a significant difference observed in egg hatch rate for broods from the 5 treatments in the high category (Figure 2; P = 0.12). There was also no difference between the 5 treatments in experiment B in the number of hatching broods in the intermediate category (χ2 = 3.62; P = 0.46).
In order to confirm that inviable broods were due to CI and not to a failure of the females to mate, all females that produced incompatible brood broods were dissected, and their insemination status was determined. Of the examined females, five of 610 were unfertilized (Table 2). These females were excluded from the analyses. The majority of dissected females (96.1%) had two spermathecae inseminated. The remaining females were observed to have sperm present in a single spermathecum (1.8%) or all three spermathecae (1.3%).
One of the factors determining the success of an IIT vector control strategy will be the ability of the released males to compete with indigenous males. Colonization and extended maintenance in the insectary can select for inappropriate mating behaviors adapted to the unnatural conditions found in the insectary (e.g. cage size, lighting, temperature, humidity). For example, while wild Anopheles form mating swarms at dusk, their laboratory counterparts may be forced to swarm in the dark [46]. The resulting released males that attempt to mate in the dark would be unlikely to find mates. An additional example is provided by a control program focused on the release of Culex tritaeniorhynchus, where mating behaviors were selected which resulted in assortative mating in the field [47]. Extended colonization may also allow for the masking of mating barriers that exist between release males and females found in the wild [48].
Therefore, it is critical to perform intermediate tests under semi-field conditions to identify potential problems before proceeding to field implementation. Here, we report that CP males are sexually (but not reproductively) compatible with field collected A. polynesiensis females and that under semi-field conditions, CP males exhibit mating competitiveness that is indistinguishable from field collected A. polynesiensis males.
Based on male competitiveness estimates (C) of 0.84 and 0.92 for experiments A and B respectively it is estimated that the number of CP males released in any control program would need to be increased by 1.25 to 1.3 times the number that would be needed if CP males had a competitiveness value of 1. This compares favorably with the estimated (C) of 0.785 reported for the MACHO strain of Anopheles albimanus in field tests assessing male mating competitiveness in El Salvador in the 1970s [49].
APA females are inseminated at equal rates by both CP and APA males, as less than 1% of the examined APA females had no inseminated spermathecae. The predominance of females with two inseminated spermathecae (96%) is consistent with a prior report [50]. This is additional evidence that the low brood hatch rates observed in treatment cages with increasing ratios of CP:APA males is due to CI and not due to the lack of successful matings of APA females with CP males.
Within the anophelines there is evidence for multiple insemination of female mosquitoes [51], [52] and previous reports indicated that Aedes aegypti is typically inseminated only once [53]. The results from this study also support the hypothesis that female A. polynesiensis only utilize sperm from a single mating. A similar finding was observed in laboratory studies evaluating the mating competitiveness of CP mosquitoes with laboratory strains of A. polynesiensis [37]. Although this study does not preclude the possibility that females are mating with more than one male, the lack of a reduction in egg hatch rate among treatments with mixed ratios of CP and APA males points to preferential utilization of a single inseminated spermathecae.
Females exposed only to incompatible CP males produced 145 broods, of which four produced an egg hatch greater than 10%. A potential explanation is that one or more wild A. polynesiensis males were accidentally introduced into the field cage as researchers entered the cage. Wild type males were commonly noted in close proximity to field cages. Additional explanations include the inadvertent introduction of either female CP mosquitoes (via a failure to completely separate females from males) or gravid wild A. polynesiensis mosquitoes into the experimental field cage. Close examination of the four hatching broods reveals that three were collected from the same field cage replicate in Experiment B. The hatch rate resulting in these three broods was >80%, which is most congruent with accidental entry of a wild type male. The remaining example occurred in Experiment A, and the hatch rate was 75%, which is congruent with the accidental introduction of a CP or previously-inseminated female. It is emphasized that broods with hatch rates <10% were considered to be from incompatible matings. CI does not necessarily equate with perfect sterility, as the strength of cytoplasmic incompatibility can be affected by the host species as well as by the strain of Wolbachia involved [54].
Although this study demonstrates that CP males are highly competitive with the APA field strain males, it should be noted that the true effectiveness of a release program is based upon a number of factors beyond just that of male competitiveness. Release program effectiveness can be impacted by the frequency and distribution of male releases, the ability of released males to locate mates, and the longevity of released males within the field environment [29]. The latter will require open field releases. The results of this study support the progression to small-scale field releases to test the efficacy of incompatible CP male releases as a vector control strategy for the South Pacific.
|
10.1371/journal.pntd.0003392 | Transcriptome Profiles of the Protoscoleces of Echinococcus granulosus Reveal that Excretory-Secretory Products Are Essential to Metabolic Adaptation | Cystic hydatid disease (CHD) is caused by the larval stages of the cestode and affects humans and domestic animals worldwide. Protoscoleces (PSCs) are one component of the larval stages that can interact with both definitive and intermediate hosts. Previous genomic and transcriptomic data have provided an overall snapshot of the genomics of the growth and development of this parasite. However, our understanding of how PSCs subvert the immune response of hosts and maintains metabolic adaptation remains unclear. In this study, we used Roche 454 sequencing technology and in silico secretome analysis to explore the transcriptome profiles of the PSCs from E. granulosus and elucidate the potential functions of the excretory-secretory proteins (ESPs) released by the parasite.
A large number of nonredundant sequences as unigenes were generated (26,514), of which 22,910 (86.4%) were mapped to the newly published E. granulosus genome and 17,705 (66.8%) were distributed within the coding sequence (CDS) regions. Of the 2,280 ESPs predicted from the transcriptome, 138 ESPs were inferred to be involved in the metabolism of carbohydrates, while 124 ESPs were inferred to be involved in the metabolism of protein. Eleven ESPs were identified as intracellular enzymes that regulate glycolysis/gluconeogenesis (GL/GN) pathways, while a further 44 antigenic proteins, 25 molecular chaperones and four proteases were highly represented. Many proteins were also found to be significantly enriched in development-related signaling pathways, such as the TGF-β receptor pathways and insulin pathways.
This study provides valuable information on the metabolic adaptation of parasites to their hosts that can be used to aid the development of novel intervention targets for hydatid treatment and control.
| The successful infection establishment of parasites depends on their ability to combat their host's immune system while maintaining metabolic adaptation to their hosts. The mechanisms of these processes are not well understood. We used the protoscoleces (PSCs) of E. granulosus as a model system to study this complex host-parasite interaction by investigating the role of excretory-secretory proteins (ESPs) in the physiological adaptation of the parasite. Using Roche 454 sequencing technology and in silico secretome analysis, we predicted 2280 ESPs and analyzed their biological functions. Our analysis of the bioinformatic data suggested that ESPs are integral to the metabolism of carbohydrates and proteins within the parasite and/or hosts. We also found that ESPs are involved in mediating the immune responses of hosts and function within key development-related signaling pathways. We found 11 intracellular enzymes, 25 molecular chaperones and four proteases that were highly represented in the ESPs, in addition to 44 antigenic proteins that showed promise as candidates for vaccine or serodiagnostic development purposes. These findings provide valuable information on the mechanisms of metabolic adaptation in parasites that will aid the development of novel hydatid treatment and control targets.
| Cystic hydatid disease (CHD) is a serious parasitic zoonosis that is caused by the larval stages of Echinococcus granulosus, a cestode that poses a threat to public health as well as significant economic losses [1], [2], [3]. At present, more than 3 million people are infected with this parasite [4], [5], and the prevalence reaches 10% in some areas [6], [7]. The disease is difficult to control because appropriate diagnostic procedures are lacking and the available drugs are inefficient [8].
E. granulosus has a complex developmental cycle, involving eggs, oncospheres, protoscoleces (PSCs), and adult stages. Adult parasites live in the small intestine of dogs. After sexual maturation, numerous eggs are produced by the adult parasites and are then excreted with the dog feces. Infections occur in an intermediate host, when eggs containing larvae are ingested. Hydatid cysts (the larval stage or metacestode) develop in the internal organs (primarily in liver and lungs) of intermediate hosts. The larval stages of E. granulosus are comprised of two layers of cyst wall: cyst fluid and PSCs [9].
As the only infectious form of the larval stages, PSCs can interact with both definitive and intermediate hosts. They mature into adult parasites when the hydatid cysts are ingested by the definitive host. They can also differentiate into new cysts when released into the body cavity of intermediate hosts upon cyst rupture [10]. Mouse models of CHD are often established via the intraperitoneal inoculation with PSCs, a method that has been widely applied to drug screening and vaccine development [11], [12]. Overall, the PSC is an important infectious reagent that contributes to the transmission of CHD and also an excellent model system in which many aspects of the host-parasite interaction can be studied.
Understanding the elaborate immune evasion strategies and mechanisms of physiological adaptation of the PSCs is critical to ascertain effective intervention targets to control the prevalence of the parasite. In this study, we focus on the role of excretory-secretory products (ESPs) that are released by parasites, as these compounds are exposed directly to the immune system of the hosts and are engaged at the host-parasite interface [13]. The mechanism by which PSCs can subvert the immune environment via ESPs is the key to successful infection. Recently, we found that ESPs from adult E. granulosus could downregulate host immune responses by preventing dendritic cells (DC) from maturing, by impairing DC function and by inducing the generation of CD4+ CD25+ FoxP3+ T cells (unpublished data). Previous studies have shown that cystic fluids produced in the intermediate hosts can modulate DC differentiation and cytokine secretion [14], while antigen B released by the germinal cells of E. granulosus can direct immature DCs towards the maturation of a Th2 cell response [15]. Moreover, the ESPs from E. multilocularis larvae have been found to induce apoptosis and tolerogenic properties in DC in vitro [16]. To date, studies have focused primarily on the immune regulation of ESPs by the host, with little work undertaken to investigate the influence of ESPs on the physiological adaptation of parasites to their hosts. Interestingly, several intracellular proteins that were not previously thought to be exposed to the immune system of hosts have recently been identified in the ESPs of PSCs [9], [17]. This finding suggests that parasite-derived ESPs are incorporated in the metabolites of the host [18], [19].
Further investigations into the mechanisms of physiological adaptation of ESPs released by PSC have been hampered due to the paucity of information regarding ESPs. Although studies have utilized proteomics to identify the constituents of ESPs [9], [20]–[22], very few have been identified. This is largely because of interference from host proteins [20]–[21] and because of technical limitations of the methodologies used. In recent years, however, the combination of transcriptomics and proteomics has enabled the identification of an increasing number of parasitic proteins [23], [24].
In this study, we used Roche 454 sequencing technology and in silico secretome analysis to explore the transcriptome profiles of E. granulosus PSCs and to elucidate the potential functions of the ESPs released by the parasite.
This study was performed in strict accordance with the recommendations provided in the Guide for the Care and Use of Laboratory Animals of the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention. The protocol was approved by the Laboratory Animal Welfare & Ethics Committee (LAWEC), National Institute of Parasitic Diseases, Chinese Center for Diseases Control and Prevention (Permit Number: IPD 2011-006).
Hydatid cysts were collected from the livers of a naturally infected sheep in a slaughterhouse in Qinghai, China. Cyst fluids containing PSCs were sucked out of the cysts using a sterile syringe. After natural sedimentation for 10 min, PSCs were carefully collected from the sediment of cyst fluids and washed 10 times with saline solution. We then added 2 mL of Trizol reagent (Invitrogen, USA) to the well-washed PSCs. After continuous mixing with a pipette, the PSCs were stored at −80°C prior to use.
Genomic DNA from the PSCs was extracted using the DNeasy tissue kit (Qiagen, Hilden, Germany) and used as a template for a polymerase chain reaction (PCR) [25]. The following two primer pairs were used to amplify the mitochondrial genes of Echinococcus species: cytochrome coxidase subunit 1 (cox1) gene (F: 5′-TTGAATTTGCCACGTTTGAATGC-3′; and R: 5′-GAACCTAACGACATAACATAATGA-3′) and cytochrome b (cytb) gene (F: 5′-GTCAGATGTCTTATTGGGCTGC-3′; R: 5′-TCTGGGTGACACCCACCTAAATA-3′). Each 25-µL reaction mixture contained 1 µL of template DNA, 12.5 µL Premix Taq® mix (TaKaRa Biomedicals, Tokyo, Japan), l µL of 10 µM of each primer, and 9.5 µL nuclease-free water. The procedure of PCR amplification consisted of 94°C for 1 min, 30 cycles of 94°C for 30 s, 56°C for 30 s, and 72°C for 1 min, followed by 72°C for 10 min, with a final holding step at 4°C. The PCR products were directly sequenced with a Dye Terminator Cycle Sequencing Kit (Amersham Biosciences, Tokyo, Japan) and ABI 3730 DNA Analyzer (Applied Biosystems, Foster City, USA).
The total RNA was extracted from the PSCs in TRIzol reagent, and RNA quality was performed by gel electrophoresis with a 2100 BioAnalyzer (Agilent Technology, Santa Clara, USA). The sequencing protocol followed that described in Liao et al. [26], and was carried out at the Shanghai OE Biotech Company. cDNA was synthesized using 2 µg of total RNA with the SMART cDNA synthesis kit (Clontech Laboratories, Mountain View, USA) according to the manufacturer's instructions. The cDNA library was constructed using a GS-FLX Titanium General Library Preparation Kit (Roche, Branford, USA) without normalization [27], and then sequenced using a half run on the Roche 454 GS-FLX Titanium platform. The modules built-in Newbler 2.5.3 (a de novo sequence assembly software, Roche, USA) was used to remove low quality sequences and assemble the remaining sequences. Briefly, the quality score trimming filter trims back from the 3′ end of reads and was based on estimated quality scores (not the final quality scores) derived from an internal calibrated signal histogram. The error rate in a sliding window (default size of 40 bp) was calculated from the estimated quality scores and multiplied by an empirical scaling factor (default of 1.1). The window was moved leftwards until the estimated error rate in the window was <1.0% (by default). If the resulting read was less than 40 bp (default), the read was discarded and not counted (numTrimmedTooShortQuality metric). After removing low quality sequences and sequencing adaptors, the remaining sequencing reads were assembled using the Newbler 2.5.3 with the ‘extend low depth overlaps’ parameter. All of the ESTs from the Roche 454 were used to run the final assembly. The resulting isotig consensus sequences and singletons were referred as ‘unigenes’ in the following study.
The software SOAP2 was used to map the raw sequence reads to the nonredundant sequence data [28]. Briefly, raw reads were aligned to the assembled, nonredundant transcriptomic data, to ensure that each read was mapped to a unique transcript. Reads mapped to more than one transcript were randomly assigned to one unique transcript, to ensure that they were recorded only once. Reads per kilobase per million reads (RPKM), the evaluation index of relative assessment of transcript abundance, was calculated using the standard formula [29].
Unigene sequences were compared (using BLASTn with a cutoff E-value of 1e-5) to public sequences available in NCBI non-reductant (Nr) and STRING databases, and to five entire genome sequences (E. multilocularis [30], E. granulosus [31], Schistosoma hematobium [32], S. japonicum [33], S. mansoni [34]).
After conceptual translation from the predicted coding domains of individual transcriptomic sequences, the functions of the potential proteins were predicted using InterProScan [35], employing the default parameters. According to their homology with conserved domains and with protein families, proteins inferred for E. granulosus PSC (EgPSC) were assigned to three gene ontology (GO) categories, including molecular function, cellular component and biological process [36]. The pathway analysis of inferred proteins was carried out using the KEGG (Kyoto Encyclopedia of Genes and Genomes) database [37].
Excretory-secretory proteins (ESPs) were predicted according to the methods described by Garg and Ranganathan [38], [39]. Briefly, the secretory proteins were predicted utilizing the following five tools: ESTScan 3.0.3 [40] to translate the unigenes into putative proteins; SecretomeP 1.0 [41] for non-classical secreted proteins; SignalP 4.1 [42] for classical secreted proteins; TargetP 1.1 [43] for trimming mitochondrial proteins; and TMHMM 2.0 [44] for trimming transmembrane proteins. The predicted proteins with no transmembrane helices were thought to be ESPs.
In addition to traditional computational approaches for ESPs prediction, we also predicted E. granulosus ESPs (EgESPs) using BLASTP [45]. Based on their homology, a list of ESP sequences that included 478 nucleotides and 1,126 proteins was obtained to extract ESPs from the proteins that were predicted to be non-secretory by SecretomeP. Those ESPs had been identified in experiments in other species (S. mansoni, S. japonicum, Brugia malayi, Ancylostoma caninum, Teladorsagia circumcinta, Fasciola hepatica and Clonorchis sinesis) [46]–[59]. In this approach, a correct match for protein (Query) to protein (Subject) was designated when the query ratio was>80% of their length and identity ≥60, while a correct match for protein (Query) to nucleic acids (Subject) was designated when the query ratio was>80% of their length and identity ≥90.
All potential ESPs were blasted with known ESP sequences from E. granulosus (including nucleotide and protein sequences [9], [7], [20]–[22] and our unpublished data) to validate the in silico secretome analysis. They were then annotated against GO, KEGG, Reactome (http://www.reactome.org/ReactomeGWT/entrypoint.htm1) and Panther (http://www.patherdb.org/) databases to identify functional groups and pathway annotations. Enrichment of KEGG pathways for genes with significant expression was calculated utilizing a classical hypergeometric distribution statistical comparison of the query gene list against all predicted E. granulosus genes. Caenorhaditis elegans pathways were used as a reference. Calculated P-values were subjected to FDR correction, with p<0.05 taken as the threshold for significance.
The transcriptome data is stored in Sequence Read Archive (SRA, No. SRP040541, http://www.ncbi.nlm.nih.gov/sra/?term=SRP040541).
The genotype of E. granulosus PSCs used in this study was sheep G1, as the PCR fragment amplified from cytb gene showed the highest identity (99%) to the E. granulosus G1 genotype referenced in GenBank (accession AF297617, S1 Figure). This was consistent with the fact that sheep G1 strain is the most common strain worldwide [60].
A total of 330,188 raw reads (mean length = 411.8 bp) were generated. The data is stored in Sequence Read Archive (SRA, No. SRP040541). After trimming to remove adaptors, low quality reads and polyN tail sequences, 329,927 clean reads remained (mean length = 400.3 bp; Table 1). Clean reads were assembled and produced about 26,514 unigenes ranging in size form 150–3,357 bp (mean = 501.5 bp). These included 4,175 isotigs ranging in size from 154 to 3,357 bp and 22,339 singletons of 150 to 1,710 bp. Approximately 84% of the isotigs were>500 bp, while most singletons (85.97%) were between 300 and 800 bp in size (Table 1, S2 Figure). The numbers of EgPSCs unigenes matching known sequences are listed in Table 1. In summary, 26,514 unigenes were inferred from our transcriptome. The large majority of these (17,861, 67.4%) exhibited the highest level of homology to proteins in E. multilocularis, followed by proteins from E. granulosus (17,732; 66.9%), Caenorhabditis elegans (8,946; 33.7%) and S. mansoni (2,159; 17.5%). Moreover, 22,910 (86.4%) contigs were mapped to the E. granulosus genome and 17,705 (66.8%) of these were distributed within the coding sequence (CDS) region, which suggested that our results were reliable.
Proteins predicted from EgPSCs transcriptome were categorized using Blast2Go [61]. A total of 5,846 were assigned at least one GO term involved in 56 GO assignments. The predominant terms for ‘biological process’ were ‘cellular process’ and ‘metabolic process’ (19.69% and 17.42%, respectively), for ‘cellular component’ were ‘cell part’ and ‘cell’ (21.65% and 21.65%, respectively), and for ‘molecular function’ were ‘catalytic activity’ and ‘binding’ (43.41% and 40.89%, respectively) (S3 Figure).
Of the proteins predicted for EgPSCs, 5,657 proteins were assigned to 306 biological pathway terms in the KEGG database (Table S1), including ‘endocytosis’ (n = 144 molecules), ‘oocyte meiosis pathway’ (n = 120), and ‘focal adhesion pathway’ (n = 118). We obtained 25 KOG clusters (S4 Figure), with 1,590 of the identified unigenes involved in at least one cluster. The largest functional group represented ‘translation, ribosomal structure and biogenesis’ (n = 214, 13.45%), followed by proteins associated with ‘post-translational modification, protein turnover, chaperones’ (n = 206, 12.95%). We also identified a further 220 (13.84%) peptidases and proteins that were linked to metabolism in eight functional categories.
PSCs are an important, infectious component of the larval stages of E. granulosus that can interact with both definitive and intermediate hosts [10]. The adaptive mechanisms that facilitate this interaction between host and parasite is of great interest to our understanding of the transmission of this widespread disease. Preliminary investigations suggest that parasites secrete certain molecules to assist in host tissue colonization [13]. We therefore focused on the components of ESPs released by PSCs and their potential roles in the physiological adaptation to their hosts and/or themselves.
Of the 26,514 unigenes identified, 19,576 were translated into proteins by ESTScan, 437 proteins were predicted to be classical secreted proteins using SignalP, while 592 were predicted to be non-classical secreted proteins according to SecretomeP. The classical and non-classical proteins were then analyzed using TargetP software for mitochondrial proteins, which resulted in the removal of 25 proteins. A further 123 transmembrane proteins were removed from the secretory protein dataset by TMHMM. In total, we obtained 881 ESPs using the four tools. A further 1,399 proteins that showed a high degree of similarity to experimentally identified secreted proteins were added by the Blast program. Thus, a total of 2,280 proteins were finally predicted as secretory proteins (Table 2).
To validate the in silico secretome analysis, we compiled a list of all experimentally identified ESP sequences of E. granulosus from the NCBI database and from previous studies (47 nucleotides and 77 proteins) [9], [17], [20]–[22], and then blasted the putative ESP sequences with the known ESP sequences (see Table S2). Ninety-one proteins were successfully mapped to the known ES proteins, of which 18 shared 100% identity and 33 shared 95%–99% identity. In addition, most known ESPs from other parasites [62] were matched successfully to those identified in our study. More importantly, domains in ESPs of Teladorsagia circumcincta (including metridin-like ShK toxin, lectin, proteinase inhibitor I29, and allergen V5/Tpx-1) were also found in the ESPs of EgPSC, which strengthens the concept that parasites employ universal ESPs to mediate parasite-host interplay [55]. Overall, these data suggest that the ESPs of EgPSCs identified in this study were reliable.
To date, there have been five proteomic studies regarding E. granulosus that have identified just 157 ESPs among them [9], [17], [20]–[22]. In this study, approximately 500 ESP domains were found, including known proteins (Table S3), a result that significantly expands the known ES components of EgPSCs. For example, WD40 repeats [63], [64], G-protein-coupled receptor (GPCR) [65] and Cadherin [66] all presented novel ESPs that were involved in parasite development-related processes. Recent studies using genome-wide and transcriptome data provide comprehensive information about the growth and development of E. granulosus [31], [67]. The results of this study extend this information and pave the way to a greater understanding of how PSCs utilize ESPs to survive in hosts.
The putative ESPs were allocated to functional categories based on InterPro domains and GO categories. Of the 2,280 proteins predicted from EgPSC, the largest functional group represented ‘binding’ (n = 201, GO: 0005488), followed by ‘catalytic activity’ (n = 196, GO: 0003824) for ‘molecular function’, ‘metabolic process’ (n = 190, GO: 0008152) and ‘cellular process’ (n = 181, GO: 0009987) for ‘biological process’, and ‘cell part’ (n = 200, GO: 0044464) and ‘cell’ (n = 200, GO: 0005623) for ‘cellular component’ (S5 Figure).
The pathway enrichment analysis for identified ESPs was performed using KOBAS v2.0 software and more than 400 pathways were identified, of which 33 were statistically significant (Table 3). The term for ‘Huntington disease’ represented the most significant group (39, corrected p<0.0001), followed by Phagosome (37, p<0.0001), Protein folding (22, p<0.0001) and Chaperonin-mediated protein folding (16, p<0.0001).
Of the 2,280 putative ESPs, only 1,406 were mapped to known functions (Table S3). These proteins included not only many common and abundant ‘house-keeping proteins’ (e.g., ribosome proteins, cytochrome subunit proteins, and enzymes involved in carbohydrate and protein metabolism), but also some rare but interesting proteins (e.g., putative receptor and antigenic proteins). This highlights the important roles of ESPs in parasite survival and development within hostile host environments. Below, we characterize these potential ESPs in greater detail.
The interaction of pathogens with mammalian hosts leads to a variety of physiology responses that drive the adaptation of the interacting partners to their new environments and conditions [19]. The ESPs released by parasites might be important actors in this process of adaptation, because they are involved in the metabolism of carbohydrates [68]. We identified a total of 122 domains (summarized in Table S4), of which, 32 proteins were identified to have a higher level of expression in the parasite (Table 4).
E. granulosus has evolved an optimal strategy to gain energy and nutrition from its host using ESPs (Fig. 1). Firstly, the parasite can regulate glycolysis (GL). We identified nine enzymes associated with GL, including the rate-limiting enzymes PFK1 and pyruvate kinase. Through GL, non-essential amino acids (e.g., glutamine, aspartic acid, arginine, proline, histidine, alanine, tyrosine and cysteine), fatty acids, adenine and hypoxanthine nucleotides, as well as pyrimidine, could be synthesized to support parasite development and growth. Alternatively, glucose and other carbohydrates could be synthesized via gluconeogenesis (GN) when alternative carbon sources (e.g., glucogenic amino acids, lactate, and glycerol) were available. In addition to the reversible enzymatic GL steps, several reactions are essential in the GN pathway from pyruvate via oxaloacetate to glucose: the reactions catalyzed by pyruvate carboxylase, phosphoenolpyuvate carboxykinase (PEPCK), fructose-1, 6-bisphosphatase, and glucose-6-phosphatase leading to oxaloacetate, phosphoenolpyruvate (PEP), fructose-6-phosphate, and glucose. Finally, tricarboxylic acid (TCA) enzymes, such as aconitate hydratase, succinate dehydrogenase complex, malate dehydrogenase, were identified in the TCA cycle. Other enzymes involved in carbohydrate metabolism are shown in Table 4.
Certain enzymes have been recognized to play key roles in the development of parasites. Phosphoglucose isomerase (PGI), one of glycolytic enzymes, has been found to stimulate parasite growth and the formation of novel blood vessels nearby the developing metacestode [69]. Vaccinating mice with recombinant PGI increases their resistance towards a secondary infection challenge [69]. Similarly, PEPCK is a novel egg antigen of S. mansoni [70] and an abundant protein in adult parasites that is related to numerous metabolic pathways (e.g., endocrine function, excretion and carbohydrate metabolism [22].
To date, only five ESPs have been identified to participate in this metabolic process [17]. The results of this study support the role of these proteins in metabolic adaptation to their hosts and, more importantly, demonstrate that many more ESPs may be used by E. granulosus to regulate carbohydrate metabolism. Further work is required to identify these additional ESPs and establish their functions.
Following infection with E. granulosus, the intermediate host produces a significant immune response that affects the growth and development of parasites [71], [72], while the parasites initiate effective evasion mechanisms to counteract adverse host environments.
In this study, we found that 36 ESP domains were molecular chaperones (Table S5), and identified a further 25 proteins that were present with high levels of abundance (Table 4), including several novel molecules (heat shock proteins, HSP90 and HSP40, universal stress protein [Usp], calreticulin, calcineurin B, GrpE in the HSP60 family and Gp96). HSP90 was the most strongly expressed of all the molecular chaperons (Fig. 2), suggesting it is one of the key molecules in mediating parasite development. This is supported by the fact that nitration of HSP90 is known to induce cell death [73], and HSP90 has been used as a drug target in protozoa intervention [74]. Previous studies have also shown that UspA and Usp8 are associated with stress resistance and growth in bacterial species [75]. ESPs might disrupt the expression of intracellular 70 protein in the host immune cells, while the parasite itself might release HSP70 to prevent damage from those same cells [76]. These molecular chaperone-like proteins may be released to regulate the stress responses that arise in the extremely harsh intestinal environments of definitive hosts (e.g., numerous highly active proteases, variable pH levels).
E. granulosus may secrete proteases or inhibitors to digest host proteins, or to protect itself from digestion by endogenous or host-derived proteinases. In this study, 39 proteases, including serine, aspartic, metallo- and cysteine proteinases, and five inhibitors, were inferred among the set of ESPs (see Table S6). Several of these (serine, cysteine, and the proteinase inhibitors) are likely to be important targets for parasite intervention and control [77]–[79]. However, only three proteases and two protease inhibitors were strongly expressed in the set of ESPs (Table 4). More sensitive technologies will therefore be required to identify other proteases that were expressed at lower levels of abundance.
In contrast, the action of antioxidant enzymes is a key component of parasite survival during infection. In this study, seven ESPs were identified as antioxidant enzymes, including glutathione transferase, peroxiredoxin, thioredoxin, Cu2+/Zn2+ superoxide dismutase, and neuronal nitric oxide synthase protein inhibitor. These molecules might be utilized by the parasite to detoxify the reactive oxygen species produced by the host environments [80].
In previous experiments we demonstrated that following infection with EgPSCs the microenvironment of the murine peripheral immune system undergoes several changes. These included T cell activation and the accumulation of immunosuppressive cells, such as myeloid-derived suppressor cells (MDSC) and CD4+CD25+FoxP3+ T cells (Treg) [71]. Such alterations might occur via the action of ESPs as many ESPs have been found to redirect host immune responses [13], [17]. In this study, we found several ESPs that contribute to immune regulation following infection (Table 4). Tegument protein (Teg) is known to induce a biased Th2 cell immune response related to chronic infection [81], while 14-3-3 proteins are associated with resistance to the immune responses mediated by local cells [82]. In addition, the antigen B (AgB) family are important in immune evasion because the antigen is secreted at variable amounts [83], and have also been demonstrated to direct immature DC maturation towards a preferential Th2 immune response [15].
Notably, cysteine proteinases have been reported to inhibit Th1 immune response via the induction of IL-4, which is the main cytokine responsible for Th2 differentiation [84]. HSP70 has been shown to stimulate both of types of response in CHD patients [85]. Also, the intraperitoneal injection of calreticulin (CRT) significantly influences Th1/Th2 balance [86]. Hence, these proteins might be novel immunoregulatory molecules that contribute to immune evasion.
We found that EgPSC possesses many signaling pathways such as P13K-Akt, mitogen-activated protein kinase (MAPK), Wnt, calcium, HIF-1, insulin, estrogen and chemokine signaling (Table S1). However, in the putative set of ESPs, only G-protein, calcium, IFN-α/β, TGF-β receptor and apoptosis signaling pathways were dominant (Table S7), which indicated their importance in parasite-host interactions and physiological processes.
Notably, we found that G-protein-coupled receptors (GPCRs), TGF-β and insulin signaling pathways might closely associate with the development of EgPSCs. For example, GPCRs can activate the G-proteins located within the cell. They work cooperatively to deliver varied signals, which in turn regulate various physiological processes [87]. However, the exact function of G-protein signaling in parasites remain unclear.
Studies have shown that TGF-β and insulin signaling pathways in C. elegans can trigger an ‘alternative’ developmental pathway, and can regulate and transit the environmental stresses on the first larval stage of the parasite [88], [89]. In particular, the disruption of both signaling pathways leads to arrested development in this species [90], [91]. Indeed, the TGF-β pathway is speculated to regulate developmental events in parasitic nematodes [92], as molecules involved in the TGF-β pathway have been found in several parasitic nematodes including Brugia pahangi, Brugia malayi and Parastrongyloides trichosuri [93]–[95]. The role of TGF-signaling in E. granulosus development and growth warrants further investigation. A recent study revealed that host insulin acts as a stimulant for parasite development within the host liver and that E. multilocularis senses the hormones of hosts through an evolutionary-conserved insulin signaling pathway, which demonstrates the importance of insulin signaling for parasite survival [96].
CHD has a global distribution and causes high rates of morbidity and has a high socio-economic burden in several countries [97]. The Eg95 vaccine induces a high antibody titer in sheep and goats, which protects them against CHD [98]. However, due to antigenic variation caused by genotypic diversity [99], the common Eg95 vaccine does not bind the antibodies of all E. granulosus species, which limits its utility. We suggest that the ESPs of EgPSCs are an excellent alternative candidate for a vaccine, as they are easy to prepare and safer for human health. More importantly, the ESPs obtained by in vitro culture have shown a 92.07% protection rate against a high dose of egg infection in sheep (1,000 eggs per sheep) [100].
Using in silico secretome analysis, we identified 44 antigenic proteins present at high abundance in our set of ESPs (Table 4). Of these, elongation factor 1 alpha, antigen B8/1, myophilin, thioredoxin peroxidase, phosphoglycerate mutase, heat shock protein 90a and actin, were the most abundant. In addition, HSP70, enolase, 14-3-3, phosphate glucose isomerase, malate dehydrogenase, glutathione S-transferase were also present at high abundance in the set of ESPs (Table S8). These abundant proteins hold enormous potential as diagnostic markers or intervention targets. Indeed, malate dehydragenase (MDH) has been tested for the immunodiagnosis of E. granulosus, while thiredoxin peroxidase (TPx) has been used for the immunodiagnosis of human CHD [101]. Likewise, the 14-3-3 molecule has been demonstrated to be a candidate vaccine against E. granulosus in mice [12], while recombinant GST protein has been used in the diagnosis of echinococcosis [102].
Proteins that are present at lower levels of abundance might also be relevant as diagnostic markers or target molecules for vaccine development. In this study, these include antigen 5 (Ag5), calreticulin, calcineurin B, thioredoxin, phosphoglucomutase, fructose-bisphosphate aldolase and gp96 (Table S8). Many of these have already shown promise for serodiagnostic purposes. For example, Ag5 is a dominant immunogenic and diagnostic antigen of the E. granulosus metacestode in both adults and PSCs [22]. Similarly, calcineurin B has been previously identified as a candidate for a vaccine or drug target [103]. Surprisingly, the E. granulosus-specific protein domain antigen B (EgAgB) family, which are well known as diagnostic targets, were undetectable in this study. This result was consistent with previous observations that little or no AgB is secreted by in vitro cultured PSCs [17], [104]. Previous studies have demonstrated that the germinal layer, but not the PSC, contributes to the primary secretion of AgB [17]. Thus, serological examination based on the AgB antibody would not be useful in early-stage PSC infection as only minute amounts of AgB antibody are produced at that time.
There are currently just two methods for the treatment of hydatid disease: surgery and the use of benzimidazole, both of which give unsatisfactory results. Hence, novel treatment compounds are urgently needed. In this study, we have identified several secretory drug targets for echinococcosis (Table 4, Table S3), including GPCRs, threonine and tyrosine protein kinase and nuclear hormones, which have been the targets of successful new drug discoveries [65]. Insulin signaling [96], thyrotropin-releasing hormone receptor, pancreatic hormone-like or transforming growth factor-β (TFG-β) families have been linked to the larval developmental of E. multilocularis. Thus, interventions that utilize these molecules could also arrest parasite growth. In addition, GL enzymes could be drug targets for parasites that rely on the GL pathway for growth and development [22]. Finally, HSP90 has been used as a drug target in protozoa intervention programs [74].
The larval stages of E. granulosus are pathogenic to human, which therefore have become the research focus of CHD. Parkinson et al. [2012] first reported genes with features that reflect physiological adaptations of different parasite stages, including PSCs, and revealed abundant long non-protein coding transcripts, upregulated fermentative pathways, candidate apomucins and a set of platyhelminth-specific gene products, which greatly increased the quality and the quantity of the molecular information regarding E. granulosus [67]. The most newly published genome of the parasite also uncovered several key events of the parasites, including the species-specific genes AgB family, bile salt pathways and Cavβ1 gene variation associated with praziquantel sensitivity [31]. Those studies have provided a molecular understanding of the growth and development of E. granulosus. In this study, we focused on the transcriptome of PSCs, which is the only infective component of the larval stages. We present novel and urgently needed information regarding the components of ESPs released by PSCs and their potential roles in the metabolic adaptation of parasites to their hosts. We suggest that intracellular ESPs are essential to the metabolism of carbohydrates within their hosts and that various molecular chaperones with a high level of expression may play a role in resisting harsh host environments. We also reveal a set of antigenic ESPs that show promise as candidates for vaccine development or in the development of serodiagnostic markers. Such findings will encourage more novel strategies for the treatment and control of CHD.
Although the coverage of the transcriptome data in this study was not deep as the genome-wide study [31], [67], these findings are novel and hold importance for understanding the mechanisms of parasite metabolic adaptations within their hosts. Overall, this study adds supplementary knowledge regarding the genomics of E. granulosus, and deepens our understanding of host-parasite interactions.
|
10.1371/journal.pntd.0003921 | A Spatio-temporal Model of African Animal Trypanosomosis Risk | African animal trypanosomosis (AAT) is a major constraint to sustainable development of cattle farming in sub-Saharan Africa. The habitat of the tsetse fly vector is increasingly fragmented owing to demographic pressure and shifts in climate, which leads to heterogeneous risk of cyclical transmission both in space and time. In Burkina Faso and Ghana, the most important vectors are riverine species, namely Glossina palpalis gambiensis and G. tachinoides, which are more resilient to human-induced changes than the savannah and forest species. Although many authors studied the distribution of AAT risk both in space and time, spatio-temporal models allowing predictions of it are lacking.
We used datasets generated by various projects, including two baseline surveys conducted in Burkina Faso and Ghana within PATTEC (Pan African Tsetse and Trypanosomosis Eradication Campaign) national initiatives. We computed the entomological inoculation rate (EIR) or tsetse challenge using a range of environmental data. The tsetse apparent density and their infection rate were separately estimated and subsequently combined to derive the EIR using a “one layer-one model” approach. The estimated EIR was then projected into suitable habitat. This risk index was finally validated against data on bovine trypanosomosis. It allowed a good prediction of the parasitological status (r2 = 67%), showed a positive correlation but less predictive power with serological status (r2 = 22%) aggregated at the village level but was not related to the illness status (r2 = 2%).
The presented spatio-temporal model provides a fine-scale picture of the dynamics of AAT risk in sub-humid areas of West Africa. The estimated EIR was high in the proximity of rivers during the dry season and more widespread during the rainy season. The present analysis is a first step in a broader framework for an efficient risk management of climate-sensitive vector-borne diseases.
| African animal trypanosomosis (AAT) is a major constraint to sustainable development of cattle farming in sub-Saharan Africa. The habitat of the tsetse fly vector is increasingly fragmented owing to demographic pressure and shifts in climate, which leads to heterogeneous risk of transmission both in space and time. In Burkina Faso and Ghana, the most important vectors are riverine species that are more resilient to human-induced changes than savannah and forest tsetse species. Therefore, understanding the spatio-temporal distribution of AAT risk remains an important task in order to design effective disease management approaches. The model developed in this research provides a fine-scale picture of the dynamics of AAT risk in sub-humid areas of West Africa. The output of the model is a risk index, the entomological inoculation rate, and it was validated against bovine trypanosomosis data using regression analysis. Parasitological status of cattle was accurately predicted, serological status was positively correlated but less accurately, whereas clinical case was not related to EIR. Our results show that the risk was high in the proximity of rivers during the dry season and more widespread during the rainy season.
| In sub-Saharan Africa, African animal trypanosomosis (AAT) is one of the main constraints to the sustainable development of cattle farming [1]. In recent years, the habitat of tsetse fly vector (genus Glossina) has undergone significant modifications due to demographic and climatic pressures. Landscape fragmentation is progressively reducing the geographic distribution and densities of tsetse, and is also affecting the epidemiology of the disease by reducing host, vector and parasite diversities [2]. In Burkina Faso and Ghana, climatic and human factors, such as cattle keeping and crop-farming, have altered the riverine landscapes over the last decades, leading to a fragmentation of gallery forests [3]. Two tsetse species remain in most of this region, namely Glossina palpalis gambiensis Vanderplank and Glossina tachinoides Westwood (Diptera: Glossinidae). Their presence and densities heavily depend on the ecotype of riverine vegetation and its degree of disturbance [4].
In Burkina Faso, several studies have investigated the impact of fragmentation on tsetse distribution and densities [4], as well as on population structure and dispersal [5,6]. A longitudinal survey investigated seasonal dynamics of tsetse and mechanical vectors of trypanosomoses in landscapes at various levels of fragmentation [6]. Environmental factors, namely temperature and relative humidity, appeared to structure tsetse distribution and densities quite differently to those of most species of mechanical vectors. Mean maximum temperature was also found to be highly correlated to the tsetse infectious rates [7]. Finally, the cyclical risk of AAT transmission was mapped during the dry and rainy seasons of the year 2005 using the entomological inoculation index, i.e. the product of tsetse apparent densities and their infection rate [8,9].
A spatio-temporal model of tsetse apparent densities was also developed in a few sites along the Mouhoun river, where a longitudinal monitoring of the parasitological status of cattle was conducted [10]. Finally, two recent national eradication initiatives with a regional dimension were undertaken in south-western Burkina-Faso and north-western Ghana under the umbrella of the Pan African Tsetse and Trypanosomosis Eradication Campaign (PATTEC), within which extensive baseline data on vector distributions and disease prevalence were generated [11,12].
By building on the above body of information, the present paper focuses on AAT risk assessment by developing a spatio-temporal statistical model of the entomological inoculation rate (EIR). EIR is a simplified index derived from vectorial capacity, which is directly correlated to the rate of transmission (R0) of a vectorial disease (see [13] for a detailed explanation). This index does not give the prevalence in cattle, but the risk for cattle that would enter a given area to become infected from a bite by cyclical vectors. Since we used cattle parasites only to calculate and model the infection rate in tsetse, the risk that we map here is specific to cattle. The use of a simplified index presents the benefit to avoid the multiplication of uncertainties for each parameter that finally reduces the predicting power of such an index [14]. A number of authors have demonstrated previously that EIR (or tsetse challenge) is well correlated to the incidence of trypanosomosis in animals (see [15–19]. This is the first time however that this risk index is mapped in space and time and linked to climatic variables. This index will help designing some future climate risk management mechanisms to control AAT. In particular, early warning system and potential index based insurance can be built using the output of this spatio-temporal modeling of AAT risk.
The study area in south-western Burkina Faso and north-western Ghana is located between latitude 9°23'- 15°5' N and longitude 0°29'- 5°31' W. The area is approximately 372,000 km2, and the main river is the Mouhoun/Black Volta. Mean monthly temperatures vary between a minimum of 18°C and a maximum of 36°C and annual precipitation between 250 and 1,170mm. The study area is constituted of Sudano-Guinean savannah in the south, Sudanian savannah in the central part and Sahelian savannah in the north [20].
In Burkina Faso, tsetse eradication efforts, targeting the northern part of the Mouhoun river basin started in 2008 (http://www.pattec.bf). In Ghana, the eradication project started in 2010 [12]. During the feasibility studies of these projects, baseline entomological surveys were carried out and generated an important amount of data, such as tsetse apparent density and their trypanosome infection rates. Biconical traps were used in all surveys [21].
In Burkina Faso, for the PATTEC baseline survey, all traps were set for three days [12] and they were deployed following a grid-based approach, within grid cells of 10x10 km [22]. Within each grid cell, 13 traps were set in the most suitable sites, in particular along the rivers and riparian thickets.
We also used longitudinal data on tsetse densities and infection rates originating from a longitudinal survey conducted in Burkina Faso. In this survey, 13 traps were spaced by 100m and set along three sections of the Mouhoun river. Traps were kept in place for three days a month, for total duration of 18 months in 2006 and 2007 [23].
The last dataset from Burkina Faso are from a recent study in the southern party of the country where entomological surveys were conducted in Moussodougou and Folonzo [24]. In these surveys, 25 traps were deployed in each site for 5 days during the rainy and dry seasons 2011–2012.
In the PATTEC baseline entomological survey done in Ghana, traps were deployed every 200m for 24h along the main rivers in dry seasons of 2008 and 2009 [12].
In addition to this entomological data from Burkina Faso and Ghana, 25 biconical traps were set in Kalofo, in northern Côte d’Ivoire. This survey was conducted during the dry and rainy seasons 2012 [25]. The space between each trap was 200m along a transect and they were set for 5 days and collected daily [26].
Finally in Mali, we used data from a PATTEC baseline entomological survey conducted in 2000–2002 for the habitat suitability model only. In this study, traps were set every 1km for 24h along the rivers (Djiteye, personal communication, and data in S1 File).
S1 and S2 Figs present the entomological data used in the models that are also provided as supplementary materials (data in S1 and S2 Files).
Data on bovine trypanosomosis originated from various sources and studies. In particular, the parasitological and serological statuses and the packed cell volume (PCV) of surveyed bovines were assembled. PCV is the proportion of red cells in the blood; it allows measuring the level of anemia in cattle. We used a threshold of 25% below which the animal was considered anemic [27]. Anemia is one of the main symptoms of AAT and it is considered to be correlated with most cattle productivity parameters [28]. In addition to data on trypanosomosis and anemia (PCV), information on sex, breed, and age of animals was also available. Three sources were used to generate the final dataset.
The first source is a cross-sectional survey carried out in the Boucle du Mouhoun region in Burkina Faso: 47 villages were selected and 2,650 cattle were sampled between September 2007 and November 2007. The study and experimental design have been previously described [11].
The second dataset is from a longitudinal survey conducted in southern Burkina Faso. Six villages were sampled and a total of 363 cattle were monitored every four weeks between June 2003 and June 2005 [23].
The last survey was performed between February 2008 and March 2008 in the Upper West Region of Ghana. In this cross-sectional study, the area was divided into 180 grid cells of 10x10 km and 36 cells were randomly selected. In each cell, 50 cattle were sampled giving a total of 1,800 cattle for the whole area [12].
For all of the above surveys, blood samples were obtained from each animal and the level of parasitaemia was scored using to the phase contrast buffy coat technique [29]. For the serological status, antibodies against Trypanosoma vivax, T. congolense and T. brucei were detected using the antibody enzyme-linked immunosorbent assay (ELISA) [30]. Finally, the PCV, a measure of anaemia, was recorded after centrifugation of blood samples. S3 Fig presents the location of the sampled herds and their serological prevalence. Data are provided in S3 File.
For the present study, a series of remote sensing data at high spatial and temporal resolution was used to assess the spatio-temporal risk of AAT (Fig 1 and Table 1). Firstly, Moderate-resolution Imaging Spectroradiometer (MODIS) data from the Terra and Aqua satellites were downloaded (http://e4ftl01.cr.usgs.gov/MOLT for Terra and http://e4ftl01.cr.usgs.gov/MOLA for Aqua). Daytime (DLST) and night-time land surface temperature (NLST) were extracted from MOD11A2/MYD11A2 temperature and emissivity MODIS products. DLST and NLST are used as proxies for both soil and air temperature, which play an important role in the epidemiology of AAT.
Both DLST and NLST data have a temporal resolution of eight days for each satellite (same composite daily data patched for both Terra and Aqua) and a spatial resolution of 1km. Low quality pixels were removed using the accompanying quality assessment layer and outliers were filtered using a variant of the boxplot algorithm [31]. The cleaned time series of DLST and NLST data was finally averaged monthly.
Monthly vegetation indices at 1km spatial resolution and monthly temporal resolution (MOD13A3/MYD13A3) were also downloaded and processed using a quality assessment layer. In particular, the Normalized Difference Vegetation Index (NDVI) and Middle Infrared (MIR) reflectance were selected to describe the vegetation condition in the study area.
Finally, a time series of dekadal gridded (11km spatial resolution) precipitation data product from FEWS-NET called Rainfall Estimator version 2 (RFE2) [32] were downloaded, downscaled (using bilinear downscaling) to match MODIS-based covariates spatial resolution (1km) and temporal resolution (monthly cumulated precipitation). We thus ended up with 11 years of monthly environmental variables (DLST, NLST, NDVI, MIR, and RFE) for the period up to—December 2013. However, the resulting dataset were still missing a few values due to cloud contamination, failure of some satellite instruments, and the data pre-processing scheme used (filtering of outliers and low quality pixels). Therefore, a spatio-temporal spectral analysis was used to fill the gaps. In particular, multivariate singular spectrum analysis was used because of its ability to capture the spatio-temporal dependence in the data and its excellent performance in comparison to other gap-filling routines when using similar spatio-temporal data [33,34]. In addition to these time series of remote sensing data, a digital elevation model (DEM) from the Shuttle Radar Topography Mission (SRTM) was used. The SRTM product at 1km spatial resolution was acquired through the CGIAR-CSI GeoPortal (http://srtm.csi.cgiar.org/). Lastly, a recently enhanced FAO cattle density layer was used [35]. This layer matches fairly recent statistics (2006 FAOstat data) and is characterized by a spatial resolution of 1km. This data was downloaded from the FAO Geonetwork website (http://www.fao.org/geonetwork/).
The main goal of the modelling exercise was to estimate the EIR in the study area using climatic and environmental data. EIR represents the number of infectious bites a host receives during a given period of time.
EIR, which is also known as tsetse challenge, is one of the most widely used and effective indicators of risk for tsetse-borne trypanosomosis [15].
The indicator is well known and widely used by malariologists to measure the intensity of malaria transmission [36]. Some efforts have been recently made to map EIR for malaria using similar spatio-temporal entomological data [37].
EIR is calculated as the product of tsetse apparent density and trypanosome infection rates of tsetse.
For a location s at a time t, we thus have:
EIRs,t=ADTs,t×IRs,t
(ADT and IR are the apparent density per trap and infection rates respectively).
For this study, the statistical models were fitted separately for each one of the two layers constituting the EIR. This was necessary because the input data used originate from various sources, and, in particular, infection rates were not available for all samples. In order to maximize the use of all available data, we decided to compute separately the apparent density and infection rates, rather than fitting a single model for the observed EIR. For the rest of this analysis, the following components of the EIR were then considered:
Tsetse habitat suitability.
Tsetse apparent density per trap.
Trypanosome infection rates in tsetse.
The first layer, habitat suitability, is not part of the mathematical definition of EIR, but it is always implied that we measure the risk of transmission where the vector occurs. Consequently, we first analyzed the habitat of the main tsetse vectors of AAT in the study area (i.e. G. p. gambiensis and G. tachinoides) before estimating and predicting EIR where the vectors can survive and transmit AAT.
The first layer needed to map the risk index is the habitat suitability. We used this layer to determine the area where the vector of the disease can survive (the ecological niche). A statistical analysis of the habitat was carried out using correlative species distribution models. Occurrence data from already described entomological surveys were used as input. Characterization of the environment in the study area relied on the 11-year average, minimum, maximum, range and standard deviation of each spatio-temporal layer (DLST, NLST, NDVI, MIR, RFE), with the DEM added to the set of summarized variables.
The methodology used to predict tsetse habitat suitability is based on the framework developed in the Niayes areas (Senegal) using the Maximum Entropy model (MaxEnt) [38]. MaxEnt is one of the most widely-used species distribution models. It is a machine learning method based on the information theory concept of maximum entropy [39]. MaxEnt fits a species distribution by contrasting the environmental condition where the species is present to the global environment characterized by some generated pseudo-absence data, also called the background. The logistic output from MaxEnt is a suitability index that ranges between 0 (least suitable habitat) and 1 (most suitable habitat). It therefore gives us a quantitative indicator of the habitat preferences of the two tsetse species in the study area.
Moreover, to account for the sampling bias present in the entomological data, a gaussian kernel based grid that gives more weight to more densely sampled areas was constructed (S4 Fig).
In order to build this grid, a smoothing parameter is needed. Five parameters corresponding to the range of maximal dispersal distance of tsetse fly were used (2, 4, 6, 8, 10km) [40] to build five bias grids for the MaxEnt models [41,42]. Model complexity in the MaxEnt framework can be controlled using the beta regularization parameter. Five parameters (1, 1.5, 2, 3, 4) were used to fit a model for each parameter. Finally, we ended up with five regularization parameters and five bias grid (one for each smoothing parameter), resulting in twenty five models. Multi-model inference was then made using model averaging weighted by the AICc [43,44].
A model was fitted for each species and we created binary maps by setting the thresholds for presence that maximize the True Skill Score (TSS = sensitivity + specificity). These thresholds were 0.33 and 0.30 for G. p. gambiensis and G. tachinoides respectively. The final layer of tsetse habitat suitability for both species was obtained by combining the two previous layers: a pixel was considered as tsetse infested when it was infested by at least one species.
The second layer of the risk index is the dynamic of the apparent density of tsetse flies, as measured using biconical traps, considered here as substitution hosts. The number of tsetse caught per trap per day is thus considered to be correlated to the relative density of tsetse to hosts. We predicted tsetse apparent density per trap (ADT) at a monthly temporal resolution and a spatial resolution of 1km2 using spatio-temporal statistical model fitted against the monthly temperature (DLST), vegetation (NDVI) and the DEM. A negative binomial model with spatial random effects was used. Negative binomial models can be seen as an extension of the classical Poisson regression to account for over-dispersion in count data.
Covariates were chosen on the basis of the available literature on tsetse population dynamics and ecology [45]. In particular, thermal- and vegetation-related covariates impact on tsetse population dynamics through their direct effects on demographic parameters (birth, mortality, etc.). Moreover, because of the sampling bias and clustering of the observations in such entomological data, a spatial random effect using the Matern correlation structure was used [46]. The correlation structure was further altered to account for the temporal effects and thus resulted in a fully spatio-temporal correlation structure. Finally, model selection and, in particular, the optimal temporal lag between environmental data and tsetse apparent density was carried out by means of a likelihood-based information criterion (corrected Akaike information criterion, AICc) [43]. Each species was modelled separately and the final layer of tsetse apparent density was obtained by summing the fitted apparent densities of both G. p. gambiensis and G. tachinoides.
The infection rate of tsetse flies represents the third and last layer in our risk index. A fly was considered infected if any major trypanosome species was detected (Trypanosoma vivax, T. congolense and T. brucei). The infection rate was modelled irrespective of tsetse species, unlike the two other models, since previous studies in the area indicated that the two species have similar infection rates [7]. It was also analyzed in the flexible framework of a generalized linear mixed model. In particular, the infection rates were investigated using a logistic regression with a random effect on the trapping site to account for spatial heterogeneity in the data. We considered temperature and host density as the main factors that influence trypanosome infection in tsetse in our study area [7]. Consequently, the model was fitted using DLST and cattle density as principal covariates and a sinusoidal function of the month when the infection status was recorded was added to the regression to account for seasonality. We also tested the same spatio-temporal correlation structure used for apparent density, which did not improve the model and was thus discarded for the sake of simplicity.
For the tsetse distribution models, we used the area under the ROC curve (AUC), the specificity and the sensitivity to assess the accuracy of the fitted models. For the apparent density model, we kept one tenth of the trapping sites for each species as testing sets and we computed the percentage of variance explained by the predicted values for each models. Finally, the infection rates model was validated by computing the McFadden pseudo-R2 [47].
The expected apparent density of tsetse (ADT) was multiplied by the tsetse infection rates (IR) and projected into the suitable habitat (HS) to estimate the EIR (tsetse challenge). In order to have an external validation of the index and to assess its ability to predict the relationship between EIR and bovine trypanosomosis, sero-prevalence, parasitological prevalence and percentage of clinical cases were explored for various temporal (1 to 4 months) and spatial (3 to 10 km around the cattle pens) lags. It must be noted that positivity to the ELISA test corresponds either to active infections or cured, past infections. Antibodies detected with this method persist for three to five months for T. vivax [48], and about two to four months for T. brucei [49,50]. Low level of haematocrit (PCV) combined with the results of the serological status was also used as a proxy of bovine trypanosomosis: an ill animal was thus defined as a sero-positive animal with a PCV below 25%.
NDVI-related covariates and cumulative rainfall estimates that describe health of vegetation (greenness relative density) and humidity were positively correlated with the presence of both G. p. gambiensis and G. tachinoides, whereas high values of temperature-related variables (DLST and NLST) lead to a low suitability index for both species. S5 Fig presents the variable contributions and response curves for the different variables. The most important variables for G. p. gambiensis were (in order on decreasing importance) minimum LST, minimum NDVI and mean LST whereas for G. tachinoides, altitude, mean LST and the standard deviation of NDVI were the most influential. The responses curves showed that overall, the response of both species to the different environmental variables were similar in shape. Mean LST, mean MIR and altitude were negatively correlated to suitability whereas minimum NDVI was positively correlated to suitability. However, the response of G. tachinoides to minimum NDVI was clearly less pronounced than that of G. p. gambiensis, confirming that the former is more xerophylous.
Fig 2 shows that hydrological network is in general highly suitable, with a wider distribution for G. tachinoides than for G. p. gambiensis. With the exception of a small area in western Burkina Faso, the uncertainty in the predictions was low (S6 Fig). The predictive power of each model was high with an average AUC of 0.95 (resp. 0.91) for G. p. gambiensis (resp. G. tachinoides) (Fig 3). The average sensitivity of the model for G. tachinoides (0.85) is higher than for G. p. gambiensis (0.76). The kappa statistic follows the same pattern, whereas average specificity for the habitat suitability model of G. p. gambiensis is higher (0.84) than that of the G. tachinoides one (0.80).
The abundance of G. p. gambiensis was positively correlated to DLST (Table 2, p = 0.002), and the suitability index (p = 0.02) and negatively correlated to NDVI (p = 0.02). The abundance of G. tachinoides was not affected by DSLT (Table 2, p = 0.22), whereas NDVI (p<0.01) and the suitability index (p<0.01) had a positive impact. Finally, percentage of variability explained by the covariates on a test dataset was high for G. p. gambiensis (94%) and moderate for G. tachinoides (39%).
High infection rates of tsetse were associated with high temperatures (Table 3, OR = 1.10, p < 0.01). However, a negative correlation with cattle density (domestic host) was observed in the study area (OR = 0.97, p = 0.03). The generalized linear mixed model captured the seasonality of infection rates in tsetse although with a low pseudo-R2 of 11%.
The computed EIR was high around rivers during all dry season and more widespread during the all rainy season between 2003 and 2013 (Fig 4). Optimal spatio-temporal lag for the regression of serological prevalence against EIR was obtained for a time lag of one month and a radius of 5km, which was then kept for the predictions (lowest AICc, Table 4).
For the model of seropositivity, fitted at the animal level (cattle), we used the breed (zebu/taurin/cross) of the animal and its age as co-variables. EIR had an important positive impact on sero-positivity probability (OR = 1.5, CI = 1.3–1.7) with a positive marginal effect (Fig 5 and Table 5). Observed serological prevalence aggregated at the village level showed a positive correlation with the predicted sero-prevalence (Fig 6, r2 = 22%). Optimal spatio-temporal lag for the regression of parasitological prevalence against EIR was obtained for a time lag of one month but there was no difference between the three distances tested (Table 4). For homogeneity with the serological prevalence model, we kept the 5km radius model.
EIR was also significantly associated to the parasitological status (Table 5, p = 0.02) and marginally to the illness status at the individual level (Table 5, p = 0.1). Older animals were also less probable to be positive to the buffy-coat test (p = 0.02). Model quality and accuracy was assessed by comparing predicted disease metrics against observed metrics aggregated at the village level on a testing dataset (25% of all data). The model had a good level of accuracy with a correlation of 67% between the observed and predicted parasitological prevalence at the village level whereas predicted illness rate at the village level was not correlated to observed values (Fig 6, r2 = 2%).
Overall, EIR allowed a good prediction of parasitological and serological status. Environmental parameters were shown to have an impact on both the apparent densities and infection rates of the two riverine species considered (G. p. gambiensis and G. tachinoides). Both species responded in a similar way to environmental parameters but with various intensities. The more important impact of minimum NDVI on G. tachinoides confirmed that this species is more xerophilous. From the spatial standpoint, the most visible and arguably predictable pattern is that AAT risk is linked to the river network, but, interestingly, a few river sections are much more risky than others, and as such they might offer priority targets for control efforts—as also previously proposed [8,9,51]. The spatio-temporal risk map of AAT presented in this study was generated and validated at a high spatial resolution and concerns a wide area. In this area, the exercise was made less challenging by the scarcity of wild fauna, leading to an endemic cycle where trypanosomes circulate mainly among livestock [52]. This cycle leads to the selection of less virulent strains that can be controlled by the combined use of curative and preventive trypanocidal drugs [53,54]. The prediction area still includes a zone where wild fauna is abundant, around the protected forest of Diéfoula, where cattle are not supposed to enter. Our model succeeded to predict a high EIR in this area and indeed, a herd that was monitored by [23] in Ouangolodougou, very close to Folonzo, during 2 years, had a very high AAT incidence (up to 20% monthly). This incidence dataset was part of the validation process. Even if farmers are not supposed to enter the protected areas, they still do so in search of better grazing areas which leads to the contact between tsetse and cattle [54]. In fact, our model probably underestimate the severity of the disease in this situation, since the strains of trypanosomes that are transmitted from wild fauna to cattle are more virulent [52].
Our analysis confirmed that higher temperatures lead to increased infection rates in tsetse. This has been attributed to increased physiological stress of tsetse associated to a higher sensitivity to infection by trypanosomes [7].
EIR was best at predicting sero-prevalence when a time lag of one month and a radius of 5km were used. The one month time lag is probably related to the time of seroconversion [55] but the best correlation with the smallest time lag tested show that the risk is quite variable in time and that our model succeeded in capturing this temporal pattern. The distance of 5km is generally considered as the ray of grazing of local sedentary herds [56], which were targeted as a priority during the various surveys. EIR is best associated to parasitological than sero-prevalence and illness status. Both parasitological and serological results can suffer from various biases. Serological diagnostic is far more sensitive than BCT and less affected by other factors like the use of trypanocide drugs, inter-recurrent diseases that may affect PCV or low parasitaemia due to trypanotolerance. On the other hand, antibodies can persist up to 13 months [57]. Thus, the probability that the animals were sampled in the area where they were actually exposed to the risk during this period is lower, even if sedentary herds were selected, due to either commercial exchanges or to past movements of the herd not necessarily considered by the farmers at the time of sampling [56]. In our study, the second category of bias is apparently more important than the former, explaining the better prediction of parasitological infection rates. Age was positively correlated to seropositivity but negatively correlated to the infection probability [58,59], which confirms that older animals develop some immunity against trypanosomes and are able to control infections better [60]. The unexpected results vis-à-vis breeds (lower seropositivity and illness in zebu than in trypanotolerant taurine cattle) might be due to confounding factors [61]. For instance, zebu are mainly present in the northern part of the study area where EIR is lower, and farmers use trypanocides more readily on zebu than on trypanotolerant cattle. Moreover, our model does not account for parasite virulence which is higher in the vicinity of protected areas [52]. Finally, trypanotolerant are generally raised under different breeding systems [62,63]. The only way this is accounted for in our model is through the selection of the grazing range in the model predicting serological and parasitological infection in cattle. However, we used the same range for all the prediction area whereas it might differ a lot between sites with different farming systems.
The present analysis is a first step in a framework for an efficient risk management approach to control climate-sensitive diseases. The methodology described in this study is generic to be applied to mitigate the risk of other vector-borne diseases through an evidence-based design of climate service mechanisms. The development and use of climate services in public health has increased recently and continues to grow, especially in the context of a changing climate [50]. More specifically, an optimal transfer of bovine trypanosomosis risk and incentives for disease control by livestock owners can be achieved through the design of index-based animal disease insurance [64].
This analysis has also consequences for Human African Trypanosomosis (HAT) commonly known as sleeping sickness. Indeed, it has been suggested that climate change is likely to impact the risk of HAT in Africa [65]. However, despite the advocacy for a One Health approach [66,67] to control such diseases, climate services to mitigate both the risk of HAT and AAT have not been designed yet. The model developed in this study can be scaled up across Africa [68] and might thus serve as the basis for a spatio-temporal model of sleeping sickness risk in Africa.
The approach developed in this study enjoys a degree of flexibility because modelling separately each component of the risk (EIR), state-of-the-art methodology for each compartment can be used. However, there are some caveats when using this "one layer—one model" approach; which are related to the increasing uncertainty at each step of the modelling process [37]. This uncertainty impacts on the estimated EIR directly. Therefore, further work is needed to develop a more robust approach to design spatio-temporal risk maps based on sparse entomological data and to evaluate these maps so that they can potentially serve as early warning systems [69].
Another important aspect to keep in mind regarding AAT risk is the role of mechanical transmission [70]. In fact, it has been suggested that, when tsetse population become sparser or disappear, other biting flies like Tabanides or Stomoxines could maintain AAT transmission, also through episodic epidemics similar to those observed in South America for T. vivax. This can constitute a potential bias in our model that accounts only for cyclical transmission. Indeed, Tabanides, which are very common in the study area, have been shown to transmit T. vivax at incidence rates as high as 63% (Atylotus agrestis) and 75% (Atylotus fuscipes) within 20 days, and T. congolense at a cumulative incidence rate of 25% (A. agrestis) in experimental conditions [71].
|
10.1371/journal.pntd.0006889 | A highly multiplexed broad pathogen detection assay for infectious disease diagnostics | Rapid pathogen identification during an acute febrile illness is a critical first step for providing appropriate clinical care and patient isolation. Primary screening using sensitive and specific assays, such as real-time PCR and ELISAs, can rapidly test for known circulating infectious diseases. If the initial testing is negative, potentially due to a lack of developed diagnostic assays or an incomplete understanding of the pathogens circulating within a geographic region, additional testing would be required including highly multiplexed assays and metagenomic next generation sequencing. To bridge the gap between rapid point of care diagnostics and sequencing, we developed a highly multiplexed assay designed to detect 164 different viruses, bacteria, and parasites using the NanoString nCounter platform. Included in this assay were high consequence pathogens such as Ebola virus, highly endemic organisms including several Plasmodium species, and a large number of less prevalent pathogens to ensure a broad coverage of potential human pathogens. Evaluation of this panel resulted in positive detection of 113 (encompassing 98 different human pathogen types) of the 126 organisms available to us including the medically important Ebola virus, Lassa virus, dengue virus serotypes 1–4, Chikungunya virus, yellow fever virus, and Plasmodium falciparum. Overall, this assay could improve infectious disease diagnostics and biosurveillance efforts as a quick, highly multiplexed, and easy to use pathogen screening tool.
| Identifying the causative agent in an acute febrile illness can be challenging diagnostically, especially when organisms in a particular region have overlapping clinical presentation or when that pathogen’s presence is unexpected. Ebola virus, for example, was not considered in an acute febrile illness differential diagnosis in West Africa until the explosive outbreak in 2013 presented the risk of infection. Besides the cost and time of screening a single patient sample for a large number of pathogens, limited sample volumes place further restrictions on what assays can be applied. Here, we developed a broad pathogen screening assay targeting 164 different human pathogens and show positive detection of over 100 of the organisms on the panel including Ebola virus, Plasmodium falciparum, and a large number of rare pathogens. The hands on time and sample volume requirement is minimal. The assay performed well in mock clinical and human clinical samples, demonstrating the clinical utility of this assay in cases where the initial diagnostic testing results in negative results. Our results provide a framework for further validation studies that would be required for formal clinical diagnostic applications.
| Appropriate diagnostic assay selection for infectious diseases depends on multiple parameters including clinical presentation and endemic pathogens known to circulate within a specific geographic region. Rapid point-of-care PCR [1, 2] and lateral flow immunoassays [3, 4] as well as more complex PCR [5–7] and laboratory based antigen capture ELISAs [8, 9] can generate a clinically actionable diagnosis in patients presenting with an acute febrile illness. These assays are sensitive, rapid, and relatively inexpensive, making this testing approach ideal for initial diagnostic testing. If these assays are negative, however, additional testing including increasingly multiplexed assays and agnostic next-generation sequencing can be utilized.
Multiplexed assays such as the MAGPIX [10, 11] or multiplexed real-time PCR [12–15] can increase the number of targets being tested. For example, Munro and colleagues described a multiplexed PCR assay with detection on the MAGPIX or Luminex instruments capable of detecting multiple influenza viruses with performance similar to real-time PCR [11]. Similarly, a multiplexed real-time RT-PCR assay, developed by Santiago and colleagues and approved by the FDA as an in vitro diagnostic device, detects all four dengue virus serotypes in a single tube reaction [15].
In cases where the initial testing methods do not result in positive pathogen identification, next-generation sequencing (NGS) is another alternative for clinically actionable infectious disease diagnostics [16]. However, metagenomic sequencing can be challenging due to a large host background, necessitating high sequencing depth to generate sufficient on target reads for pathogen detection. Targeted NGS, in which a specific signature is amplified [17, 18] or enriched from a complex sample using hybridization [19], can increase pathogen specific reads sufficiently to allow detection on desktop sequencers such as the Ion Torrent or the MiSeq. Using these approaches, however, adds time-to-answer due to library preparation, sequencing, and analysis.
A potential solution described here is the use of the NanoString nCounter platform for highly multiplexed pathogen detection. This system utilizes direct hybridization and detection of a nucleic acid target and can be highly multiplexed (up to 800 different targets). Since this technology has been successfully implemented for quantitative gene expression studies [20–22], we investigated whether this platform could be used for broad, targeted pathogen detection in a situation where rapid testing (ex. real-time PCR) was negative. In this context, we developed and evaluated a panel containing 195 different assay targets against 164 different viruses, bacteria and parasites. Overall, this panel was not as sensitive as real-time PCR; however, this assay successfully identified multiple pathogens quickly, demonstrating utility as a pathogen screening assay.
All organisms used in this study (listed in S1 File) are maintained at United States Army Medical Research Institute of Infectious Diseases (USAMRIID) or were provided by the Unified Culture Collection (UCC) or the American Type Culture Collection (ATCC, Manassas, VA). Samples included bacterial, parasite DNA, cell culture supernatant from virus-infected cells treated with TRIzol LS (ThermoFisher Scientific, Waltham, MA) or gamma irradiation. Total nucleic acid from each unpurified sample was extracted using the EZ1 Virus Mini Kit v2.0 (Qiagen, Valencia, CA) with the EZ1 robot (Qiagen) according to the manufacturer’s instructions. Total nucleic acid was eluted in 90 μl elution buffer.
Due to a limited supply, Coxiella burnetii DNA was amplified using the REPLI-g Whole Genome Amplification Kit (Qiagen) according to the manufacturer’s instructions. The number of C. burnetii genome equivalents (GE) was approximated using the genome of C. burnetii RSA493 (GenBank# NC_002971) and the C+G (42.7%) and A+T (57.3%) genome percentages. Based on these calculations, 1 GE is approximately 2.05 fg. The approximate number of GE for Plasmodium falciparum 3D7 DNA (ATCC) was similarly determined to be approximately 23.89 fg.
A custom Broad Pathogen Detection Assay (BPDA) targeting a broad panel of medically important viruses, bacteria, and parasites was designed and acquired from NanoString Technologies (Seattle, WA). Using 195 different capture and reporter probes, this assay targeted 164 different pathogens of concern for human health (S1 File). After initial testing showed lower than desired assay sensitivity, nested primers targets were designed by NanoString using Primer3 software [23–25]; see S1 File for the sequences. These multiplexed primers were used in a multiplexed target enrichment (MTE) reaction to amplify the capture/reporter target prior to detection. Primer pairs for 4 probe targets could not initially be designed and were redesigned for incorporation into a subsequent MTE iteration.
Individual MTE primer pairs for all available pathogens were evaluated for amplicon generation using SuperScript One-Step with Platinum taq (Thermo Fisher Scientific) with the following cycling conditions: 50°C for 15 minutes, 95°C for 5 minutes, 40 cycles of 95°C for 30 seconds, 60°C for 1 minute and 72°C for 1 minute. The final reagent concentrations per 20 μL reaction were: 1X Reaction Mix, 4 mM MgSO4, 0.25 mg/mL BSA, 50 μM primers, 0.4 units of Platinum Taq. Amplicon generation was visualized on the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA) using the DNA 1000 Kit (Agilent Technologies).
After confirming successful amplification using selected individual primer pairs, all primers were combined into a single 500 nM primer mixture for the MTE reaction. Sample cDNA was generated by adding 4 μL purified total nucleic acid to 1μL of SuperScript VILO MasterMix (Thermo Fisher Scientific) and incubating at 25°C for 10 minutes, 42°C for 60 minutes, and 5 minutes at 85°C. The sample was then added to 5 μL of TaqMan PreAmp Master Mix (Thermo Fisher Scientific) and 1 μL of the 500 nM primer mixture. MTE used the following cycling conditions: 94°C for 10 minutes, then 18 cycles of 94°C for 15 seconds, 60°C for 4 minutes, and a 4°C hold. The entire enriched sample (11 μL) was used for detection on the NanoString nCounter platform.
Pathogen nucleic acid (total nucleic acid and MTE amplified nucleic acid) was initially denatured at 95°C for 5 minutes and immediately transferred to ice for 2 minutes. Next, 5 μl of each denatured sample or the entire MTE reaction (11 μl) was added to a 20 μl NanoString master mix containing the BPDA Reporter Codeset plus 130 μL hybridization buffer followed by 5 μl of the BPDA Capture Codeset. Reactions were immediately placed on a thermocycler for an overnight incubation at 65°C (for ~16 hours), loaded onto a sample cartridge using the nCounter Prep Station, and scanned using the NanoString nCounter Digital Analyzer. A sample was called positive for a specific target if the number of counts was greater than the average of the internal negative controls for that target plus three times the standard deviation of the negative controls.
The impact of the MTE reaction on sensitivity was assessed using SYBR Green real-time PCR assays using primers internal to the MTE primers [See S1 File for Bundibugyo virus (BDBV), Marburg virus (MARV), Ebola virus (EBOV), influenza B virus, and P. falciparum assay information]. Total nucleic acid for each organism was serially diluted and amplified by MTE. The levels of enrichment were measured by real-time RT-PCR using Superscript RT-PCR reagents (Thermo Fisher Scientific) and SYBR Green. The cycling conditions for the SYBR Green RT-PCR assays were: 50°C for 15 minutes, 95°C for 5 minutes, 45 cycles of 95°C for 5 seconds and 60°C for 20 seconds, and a final melt step from 60°C—95°C at a rate of 0.2°C per second. The final reagent concentrations per 20 μL reaction were as follows: 1X Reaction Mix, 3mM MgSO4, 0.25 mg/mL BSA, 1 μM primers, 2X SYBR Green, and 1 unit of Platinum Taq.
Following optimization and characterization of the MTE reaction, assay performance was determined utilizing the MTE reaction and detection on the NanoString platform. A preliminary limit of detection (LOD) for BDBV, MARV, EBOV, influenza B virus, and P. falciparum was determined with and without MTE by serially diluting organism and testing for positive detection. The preliminary LOD was defined as the lowest concentration of organism having all three replicates testing positive. Testing at the preliminary LOD was repeated (ten replicates) without MTE for BDBV and MARV Angola and with MTE for influenza B virus, BDBV, and MARV Angola to show assay reproducibility. Similarly, LODs were conducted using existing real-time RT-PCR assays as previously described [26, 27].
Mock clinical samples were prepared in order to evaluate the ability of the BPDA to detect samples that had been extracted from whole human blood. Dengue virus serotype 3 (DENV-3) in TRIzol LS and gamma-irradiated EBOV were diluted in human whole blood treated with EDTA (Bioreclamation, Westbury, NY) in 200 μl samples. Samples were extracted using the Qiagen EZ1 XL Advanced with the EZ1 Virus Mini kit 2.0 by adding an equal volume of ATL buffer (Qiagen) to each sample prior to being placed on the automated instrument for extraction. Each sample was run in triplicate with and without MTE amplification prior to being tested with the pathogen panel.
Since the BPDA assay could be a useful diagnostic tool, de-identified human clinical samples were tested. These samples were acquired through USAMRIID’s Special Pathogens Laboratory. All samples were de-identified prior to use, and all studies were conducted in compliance with United States Department of Defense, federal, and state statutes and regulations relating to the protection of human subjects, and adheres to principles identified in the Belmont Report. All data and human subjects research were gathered and conducted for this publication under an Institutional Review Board approved determination FY17-31 as defined by 32 CFR 219.102(f).This sample set comprised of Chikungunya virus positive and negative samples, as determined by real-time PCR [28]. Total nucleic acid was extracted from each sample using the Qiagen EZ1 XL Advanced with the EZ1 Virus Mini kit 2.0 and tested using the BPDA.
We developed a Broad Pathogen Detection Assay (BPDA) for use with the NanoString nCounter platform in order to quickly screen a sample for multiple pathogens in a single tube reaction. This assay consisted of 195 detection probes targeting 164 different viral, bacterial, and parasitic pathogens of concern for human health. Initial testing using the highest concentration of organism available showed positive detection of multiple pathogens including EBOV, MARV, P. falciparum, and C. burnetii (see Table 1 for a selected list and the S1 File for a full detection list). However, some pathogens such as Crimean-Congo hemorrhagic fever virus (CCHFV) were not detected even at this high concentration (Table 1). A multiplexed target enrichment (MTE) step utilizing a complex PCR to amplify the pathogen-specific probe hybridization site was used for use prior to detection with the BPDA to mitigate this issue. Incorporating this upfront target enrichment step increased the assay sensitivity as shown by the now positive detection of CCHFV and increased read counts for almost all pathogens tested (Table 1).
Having shown the effectiveness of the MTE step for increasing assay sensitivity, we wanted to further assess the target enrichment capability of this method by comparing the amount of target amplicon present with and without MTE. Comparing real-time PCR results for the MTE amplified and non-amplified reactions, MTE enrichment showed a decrease in Cq values, indicating an increase in the target amplicon (Fig 1). Statistical analysis (two-way ANOVA with Bonferroni correction) identified that all points for each virus were significantly different with the exception of the highest influenza B virus concentration.
Assay limit of detection (LOD) studies were conducted with serially diluted organism in order to define assay performance for medical relevant concentrations of the tested organism. Incorporating the MTE enrichment improved detection and lowered LODs (Fig 2). This improvement was most notable for influenza B virus which was undetectable without enrichment but tested positive following MTE (Fig 2B). Similarly, the preliminary LOD for MARV improved from 2.2 x 106 PFU/ml without MTE to 3.75 x 104 PFU/ml after enrichment (Fig 2D). Generally, the assays were highly specific for the targeted organism. For example, BPDA showed positive results for only P. falciparum while other Plasmodium species including knowlesi, malariae, ovale, and vivax were called as true negatives.
To confirm assay reproducibility at the preliminary LOD, ten replicates of BDBV, influenza B virus, and MARV were tested with and without MTE (Table 2). For influenza B virus, no replicates tested positive at the highest concentration used; however, MTE incorporation resulted in repeated detection of all viruses. In addition, comparison of BPDA performance to real-time PCR, the current gold standard for molecular based pathogen detection, showed real-time RT-PCR was the more sensitive technique (Table 2).
Testing EBOV and DENV-3 spiked into whole blood at three different concentrations showed the clinical applicability of this assay. Application of the MTE did not impact the overall testing results, but MTE increased the number of EBOV-specific counts for all dilutions and replicates (Table 3). DENV-3 tested positive with all replicates and dilutions; however, MTE did not result in increased DENV-3 counts (Table 3). Interestingly, the pre-amplification of DENV-3 signatures resulted in a lower number of counts as compared to the same samples without MTE, potentially suggesting suboptimal amplification for that DENV-3 isolate.
Further characterization of the clinical utility of the BPDA showed positive detection across 14 de-identified, human clinical samples with potential Chikungunya virus (CHIKV) infections (Table 4). Both the BPDA and the MTE-BPDA assays correctly identified the 12 real-time RT-PCR positive samples and the two negative samples (Table 4). Incorporating the MTE component increased the number of CHIKV counts by 1–2 log for all of the positive samples tested.
Positively identifying the etiologic agent for an acute febrile illness can be critical for ensuring appropriate administration of treatment and supportive care; however, proper identification can be challenging. Highlighting the importance of broad pathogen screening and appropriately fielded diagnostics, a recent study by Schoepp and colleagues found ~70% of the suspected Lassa fever patients admitted to the Lassa Fever Ward in Kenema, Sierra Leone, were negative for both Lassa virus and the malaria parasite, both hyperendemic pathogens in the region [29]. This study found serological evidence of filovirus infection (EBOV and MARV) in the years prior to the explosive Ebola virus disease outbreak in West Africa [29], and it is likely that previous infections with EBOV as well as MARV were misidentified as severe malaria or Lassa fever. Accurate testing of these acute febrile patients with multiplexed assays could have identified the risk of an EBOV outbreak earlier.
Here, we developed and evaluated a highly multiplexed, broad pathogen detection assay for use following negative detection using singleplex assays (ex. real-time PCR). This assay targets 164 different human pathogens of public health concern and includes viruses, bacteria, and parasites. We included multiple organisms with overlapping clinical presentation, such as Plasmodium, Lassa virus, and Ebola virus. We also included a large number of less common organisms in order to maximize the diagnostic potential of the assay. While the assay performed well without target enrichment, applying MTE prior to running on the NanoString platform greatly improved assay sensitivity. While extensive primer optimization was not conducted in this study, such an optimization would likely improve the overall assay sensitivity and the detection variation we observed. As we were unable to do this optimization, all testing was conducted with and without MTE. Overall, 98 of the 164 pathogens on the panel that we had available for testing were positively identified including endemic pathogens to West Africa such as EBOV, Lassa virus, dengue virus, and the malaria parasite P. falciparum. Assay run time, from start to finish, is approximately 27 hours with approximately one hour (or 30 minutes without the MTE) of hands-on time. Future efforts include the acquisition and testing of the remaining pathogens on the panel.
Characterization of the BPDA using mock clinical and clinical samples showed the efficacy of this assay for detecting pathogen in patient samples. Specifically, testing of spiked human serum showed positive detection of EBOV and DENV-3 at clinically relevant concentrations. In addition to mock clinical samples, the assays correctly identified all CHIKV human clinical samples, demonstrating the ability to correctly identify pathogens from natural infections. These studies were in agreement with real-time RT-PCR testing establishing preliminary 2-by-2 testing that would be required for regulatory use.
There are a variety of easy to use multiplexed assays described in the literature [6, 7, 30, 31]; however, there are inherent limitations in multiplexability within a single assay. Other technologies offer higher levels of multiplexing through microarray [32, 33] and next-generation sequencing [16–19, 34]. However, these assays are highly technical, and the large number of targets makes full validation of each signature highly challenging in both cost and time. Furthermore, clinical validation of these assays would be further complicated by the regulatory requirement to validate each potential organism the assay could detect [35, 36]. Preliminary limit of detection (5 dilutions in triplicate for 164 organisms) and confirmation of the preliminary limit of detection (twenty replicates of 164 organisms) testing alone would require 2,460 reactions at approximately $100 USD/reaction (~$575,000 USD). Mock clinical testing would further expand the testing numbers and cost.
Ideally, comparison of primer/primer interactions and a direct comparison of each target to a gold standard (ex. real-time PCR) would be performed for diagnostic applications. Within current regulatory paradigm, this type of validation also remains cost prohibitive similar to other highly multiplexed assays (ex. microarray and next-generation sequencing assays) as this would require independent testing of each target on the panel in a statistically robust manner. However, results presented here provide proof of concept testing results and the framework for such a validation by demonstrating the proof of concept utilization of this technology for infectious disease diagnostics.
|
10.1371/journal.pcbi.1006092 | Interactions of spatial strategies producing generalization gradient and blocking: A computational approach | We present a computational model of spatial navigation comprising different learning mechanisms in mammals, i.e., associative, cognitive mapping and parallel systems. This model is able to reproduce a large number of experimental results in different variants of the Morris water maze task, including standard associative phenomena (spatial generalization gradient and blocking), as well as navigation based on cognitive mapping. Furthermore, we show that competitive and cooperative patterns between different navigation strategies in the model allow to explain previous apparently contradictory results supporting either associative or cognitive mechanisms for spatial learning. The key computational mechanism to reconcile experimental results showing different influences of distal and proximal cues on the behavior, different learning times, and different abilities of individuals to alternatively perform spatial and response strategies, relies in the dynamic coordination of navigation strategies, whose performance is evaluated online with a common currency through a modular approach. We provide a set of concrete experimental predictions to further test the computational model. Overall, this computational work sheds new light on inter-individual differences in navigation learning, and provides a formal and mechanistic approach to test various theories of spatial cognition in mammals.
| We present a computational model of navigation that successfully reproduces a set of different experiments involving cognitive mapping and associative phenomena during spatial learning. The key ingredients of the model that are responsible for this achievement are (i) the coordination of different navigation strategies modeled with different types of learning, namely model-based and model-free reinforcement learning, and (ii) the fact that this coordination is adaptive in the sense that the model autonomously finds in each experimental context a suitable way to dynamically activate one strategy after the other in order to best capture experimentally observed animal behavior. We show that the model can reproduce animal performance in a series of classical tasks such as the Morris water maze, both with and without proximal cues, which support the cognitive mapping theory. Moreover, we show that associative phenomena such as generalization gradient and blocking observed within the navigation paradigm cannot be explained by each learning system alone, but rather by their interaction through the proposed coordination mechanism. The fact that these experimental results have for a long time been considered contradictory while they could here be accounted for by a unified modular principle for strategy coordination opens a promising line of research. We also derive model predictions that could be used to design new experimental protocols and assess new hypotheses about complex behavior arising from the interaction of different navigation strategies.
| Neurobehavioral evidence supports a prominent role for interactions between multiple anatomically distinct memory systems in the mammalian brain underlying the coordination of different behavioral strategies during learning (e.g., [1]): A cognitive memory system, relying on a network comprising the hippocampus, prefrontal cortex and associative parts of the basal ganglia (i.e., the dorso-medial striatum), would mediate goal-oriented planning strategies; While a stimulus-response/habitual memory system, relying on sensorimotor parts of the cortex and basal ganglia (i.e., the dorso-lateral striatum), would in parallel mediate the progressive acquisition of routine strategies that would take over with overtraining [2–8].
Recently, a growing computational effort has been put forward to model the coordination of such behavioral strategies, with more and more computational models employing such a dual learning systems framework to account for changes in animals’ behavioral strategies between different stages of learning during the task [4, 9–12], as well as between different subparts of the action sequence or movement trajectory during the trial [9, 13, 14].
In particular, when dealing with instrumental conditioning experimental data, these dual systems models well explain animals’ tendency to alternate between initial flexible goal-oriented strategies, where the animal is hypothesized to use an internal model to plan and infer future consequences of action (model-based), and more automatic and habitual strategies at late stages of learning, where behavior is supposed not to rely on an internal model but rather on stimulus-response associations (model-free) (see e.g., [5, 15] for reviews). In the case of navigation paradigms, the model-based / model-free dichotomy has been found to better account for the diversity of navigation behaviors than the old classical distinctions between place and response strategies, or between allocentric / egocentric strategies [8]. Moreover, such a distinction provides a possible explanation of the distinct roles of the hippocampus and different subparts of the striatum during navigation [8, 15].
Nevertheless, how learning systems dynamically interact during navigation is still little understood. In particular, it is not clear how a unified coordination principle or mechanism can explain both cases of strategy competition (when a lesion impairing one strategy but leaving another one spared can produce an improvement in the animal’s behavioral performance, e.g., [16]) and cases of strategy cooperation (when two strategies together produce a better performance than one strategy alone, e.g., [17]). Existing computational models have proposed various criteria to coordinate multiple learning systems, but each criterion has been evaluated on specific experimental paradigms. For instance, system coordination has been proposed to depend on the uncertainty in the model-free system alone [11], relying on the strong assumption that the model-based system always has perfect information; Alternatively, some models have released the assumption of perfect information [18], but they still bias the coordination towards a default model-free control, which cannot explain why some actions always remain under model-based control even after training [4]. Other models propose a coordination that can also depend on uncertainty in the model-based system [4, 19], an approach that does not scale up to tasks involving a large number of states [20]; some authors have used a fixed coordination weight per individual [12, 21], which cannot account for dynamic changes in coordination strength along training; system coordination has also been proposed to depend on working-memory load [18, 19, 21] in human experiments where it is considered that accessing working-memory has a cost, and more recently in similar experiments in monkeys [22]. Few multiple-systems models have addressed rodent navigation data. The model proposed by [10] could solve a variety of navigation tasks in a simulated robot but employed fixed pre-learned behavior in the model-free system and did not perform formal comparisons between their simulated robot and experimental results in rats. The model proposed by [9] combined place-based and cue-guided learning systems, coordinating them by choosing at each timestep the system with the smallest reward prediction errors and the largest reward expectations. However, since the two learning systems are model-free, the model cannot account for flexible strategies enabled by model-based learning. In general, most previous computational models of rodent navigation have employed what is called a Locale strategy to account for place-based behavior [9, 13, 17, 23, 24], which learns place-action associations through model-free learning and can thus not account for model-based behavior.
We previously proposed a computational model of navigation where a gating-network coordinates model-free and model-based systems with a commun currency: their measured instantaneous performance [14]. In the model the gating-network is an associative module which learns through model-free reinforcement learning which system is the most efficient in each location of the spatial and perceptual spaces, which implements a certain degree of hierarchy in learning [25]. This enables to gate the appropriate system at the right moment during performance, depending on input from place cells’ and visual cells’ activity (Fig 1). The fact that the gating-network learns to select which navigation strategy to follow based on model-free reinforcement learning is consistent with the hypothesis that the same selection mechanisms learned through dopamine reinforcement signals are employed in different striatal territories for movement selection, action selection, and strategy selection [5, 26–32]. Previous studies reported how the model could reproduce rodent experimental data in two specific navigation tasks. Nevertheless, the previous version of the model employed a hand-tuned mixture of Gaussians to model artificial hippocampal place cells with a fixed distance between place fields, and it is thus not clear how general these previous results were. Here, we extend this model by integrating a more realistic hippocampus model [33] and show how the mechanisms of the gating network within this new model can explain a wide range of experimental data during navigation paradigms involving spatial memory as well as associative phenomena such as generalization and blocking.
Specifically, we reproduce the classical reference memory experiment in the hidden water maze [34]; a delayed matching to place task [35]; cases of competition between strategies previously classified as cue-guided and place-based in a water maze [16]; a gradual competition between distal and proximal cues [36]; generalization gradient [37] and blocking [38]. In particular, we show that phenomena such as generalization gradient and blocking observed within the navigation paradigm, which to our knowledge have never been accounted for by computational models before, cannot be explained by each learning system alone, but rather by their interaction through the proposed gating network.
While the debate is still vivid in psychology and experimental neuroscience between the cognitive map theory and the associative theory of mammal navigation [39–44], our work highlights together with recent previous computational models that a variety of learning behaviors can result from a single coordination mechanism for the interaction between these two types of strategies. Moreover, while most previous computational models focus on mechanisms for the competition between learning systems, our work shows that a set of rodent navigation behaviors can be explained in terms of cooperation between systems. Finally, by proposing a common currency for learning systems coordination, our model can generalize to the coordination of N systems whose individual learning mechanisms may be of different nature. This could help predict behavior in paradigms involving more than two navigation strategies, which has so far rarely been experimentally studied.
The proposed computational model is composed of four main modules (Fig 1): an associative Direction strategy (D) which learns through model-free reinforcement learning to associate the perception of proximal cues within the environment with directions of movements; a cognitive mapping Planning strategy (P) which learns through model-based reinforcement learning a transition graph between different positions within the environment encoded in simulated place cells, and proposes directions of movement based on action plans towards the memorized goal position; and an Exploration strategy (E) which proposes random direction of movements. Finally, a Gating Network which learns through model-free reinforcement learning which strategy to select based on the system’s input (i.e., cue cues and place cells). The module dedicated to the model-based Planning strategy is itself composed of several modules dedicated to building hippocampal place representations through the integration of idiothetic and allothetic information from enthorinal cortex grid cells and sensorial cells to dentate gyrus place cells, and projections of pools of place cells to nodes of the cognitive graph within the prefrontal cortex module (S1 Fig). This results in more variability and plausibility of the simulated place fields compared to the the uniformly distributed Gaussian place cells that we used in the previous version of the model [14]. The detailed mathematical formulation of the model as well as parameter tables (S1 and S2 Tables) are given in S1 Text. We tested the model in several experimental paradigms with increasing complexity in order to show that the same associative principle for the coordination of the model-free Direction strategy and the model-based Planning strategy can account for a wide series of experimental data on rodent navigation. These simulations provide computational predictions about the way distal and proximal cues may compete for the control of behavior within such a modular architecture (examples of repartitions of cues processed by each of the modules are illustrated in S2 Fig). We also show that some experimental results previously accounted for by a model-free spatial strategy called Locale strategy (L) can be better explained in terms of the model-based Planning strategy (P) within this framework.
One of the best known experimental paradigms, the Morris water maze show that intact rats are able to learn the location of a hidden, stable platform [34] (Fig 2a). In contrast, hippocampal-lesioned animals are impaired in such tasks as shown in Fig 2b.
While previous computational models have already reproduced these classical results (e.g., [17]), we present here new simulations with our model to show that it can also reproduce them (Fig 2c), but also to analyze which variants of the model fail to do so. The simulated hippocampus-lesioned model, where only the Direction (D) strategy is operational, shows significantly higher latencies to reach the platform during the first 10 trials than the full model, where both strategies (DP) are operational (Mann-Whitney test for non-matched paired samples, p < 0.001). In the simulations, when P and D strategies are available simultaneously, the gating mechanism learns to privilege the former (S3a Fig) which uses the configuration of distal cues to estimate the allocentric position of the platform and to plan a sequence of movements towards it. In contrast, the performance of an associative model with a D strategy only, where distal cues compete against each other, was impaired as is the case with hippocampal lesioned animals. Interestingly, our simulations predict that if the experiment is performed for a sufficient number of trials, animals with impairments in hippocampal processing (i.e., D strategy alone) should eventually reach the platform with performance that is not statistically different than control animals (i.e., DP strategies together). This is consistent with more recent experimental results showing that the blocking of hippocampal sharp-wave ripples oscillations, known to be important for memory consolidation, impairs performance in a spatial memory task but still spares a slow improvement in performance in the tested animals [45]. Moreover, our simulations predict that Striatum-lesioned animals (i.e., P strategy alone) should not be impaired in this task (Fig 2c). This is again consistent with more recent experimental results in the water maze where striatum-lesioned animals had non-different espace latencies than controls [16]. So far, these results are not novel compared to the large body of computational simulations of this experiments that have been previously done [17, 23, 24]. Nevertheless, it is interesting to note that many such models have reproduced these results using a model-free allocentric Locale (L) strategy in contrast to the model-based one used here.
Simulation of a variant of our model where the Planning strategy is replaced by a Locale can also reproduce the experimental results (S3b Fig). Nevertheless, the simulation results strikingly lead to different predictions: that the performance of Hippocampus-lesioned animals should never reach that of the control animals even after a large number of simulated trials; and that the performance of Striatum-lesioned animals (i.e., L strategy alone in S3b Fig) should also be impaired (but less) compared to control animals. Hence in this variant of the model, the two strategies together produce a better performance than each strategy alone, which reveals a potential collaborative interaction between strategies that will be discussed later. These predictions constitute possible ways to disentangle the two alternative models. However, here we argue that the ability of a model without model-based strategy to reproduce these results is mainly due to the stationarity of the task: the platform always remains at the same location, which can be easily learned by a model-free strategy. The next simulated experiments will show that in non-stationary cases, a model-based strategy is necessary to reproduce rats’ ability to adapt in a few trials to each change in the platform location.
In an extension to the previous paradigm, the hidden platform was moved every session made of four trials (Fig 2d), thus allowing the animal to remember its position for a few trials once it has been found, but nevertheless requiring an adaptation to frequent goal location changes [35]. Experimental results show that escape latencies of intact animals increased on the first trial after the platfom is moved, but decreased quickly in the following ones (Fig 2e). Results of model simulations showed that this quick adaptation of behavior can be reproduced when a model-based Planning strategy is available (Fig 2f), but not with a model-free Locale one nor an associative Direction one (S4a and S4b Fig). The Planning strategy permitted the quick within-session adaptation observed in rats, while both Locale and Direction strategies were much slower at learning the platform location and hence did not display much within-session reduction in escape latency. Analysis of the evolution of the contribution of strategies within the full model shows that while the Direction strategy contributed to the model decisions of movement during the first simulated session, the model quickly learned to avoid using it during next sessions (S4c Fig). This explains why the performance of the full model shows smaller within-session reduction in escape latencies during the first session than during later sessions. At that stage, the model automatically learned that solving this task can be achieved through a combination of Planning and Exploration strategies. When the contribution of the Exploration strategy increased, such as during session #5, the model started the session with a lower escape latency because the model relied less on the Planning strategy at the first trial of the session and hence spent less time searching around the previous platform location. This suggests that an ideal combination of strategies in this type of tasks, once the structure of the task is learned by the model, would be to rely most of the time on the Planning strategy—to enable the quick within-session improvement in performance—while keeping a certain level of exploration to prevent the model from being stuck at the previous platform location. As we will see in Experiment IV, the gating network of the model can achieve this sort of cooperation between strategies—hence suggesting a way in which rats may do it—when the presence of an intra-maze cue enables the Direction strategy to be efficient at the first trial of each session. Before that, the next simulated experiment will illustrate a case where the use of an intramaze cue enables the Direction strategy to reach a good performance and hence to enter competition with the Planning strategy.
In this experiment, animals learned to reach a cued and stable platform, also identified by surrounding distal cues. During some trials, the cue was hidden, forcing the animals to learn its location also by distal cues (thus discarding the possibility of overshadowing—i.e., neglecting—distal cues because of the presence of the proximal one) [16]. In the last trial block (4 trials), the cued platform was moved at the opposite place, testing whether rats reach it following its spatial location or the cue (Fig 3a). In these trials, hippocampal-lesioned animals went directly towards the new cued platform position, as did half of the control animals—named cue-responders. In contrast, the remaining control animals—named place-responders—first swam towards the previous platform location (presumably following distal cues) and then went directly to the cued goal. This suggests a competition between both strategies taking place at these trials (Fig 3b).
Simulation results in this task with a previous version of our model have already been reported in [14]. That study focused on whether output actions in the model should have an egocentric or an allocentric frame. Here, using simulations relying on an allocentric frame for output actions, we address two questions: how the gating-network can manage inter-strategy competition in order to solve the task; what are the different experimental predictions raised when the model-free Direction strategy competes with a model-based Planning strategy versus a model-free Locale one.
As previously reported [14], the model can reproduce the experimental results both in the control case (when strategies P and D are available) and in the hippocampal lesion case (when only the D strategy is operational) (Fig 3c). Simulations reproduce the fact that control and lesion groups perform comparably in the trials where the intra-maze cue is visible (trials #1, 2, 4, 5, 7, 8) and during the competition trial #10, as well as the significantly larger escape latencies of the lesion group in trials where this cue is hidden (trials # 3, 6, 9). An interesting new prediction from the model is that lesions to the striatum (putatively impairing the Direction strategy while sparing the Planning strategy) would produce an intermediate performance (P group in Fig 3c). More precisely, the performance should not be impaired in the hidden cue case because the Planning strategy can still rely on distal cues to locate the platform. Nevertheless, the performance in the visible case should not be as good as the full model, indicating that the full model solves this case through a cooperation between strategies rather than by the Planning strategy alone. Such a cooperation is illustrated in S5d–S5h Fig where the trajectory produced by the agent during a given trial expresses a D strategy during the initial part of the trajectory and a P strategy later on. This enables to spend less time far from the new platform location by preventing the P strategy from driving the agent towards the previous platform, as is the case with the P model alone during the competition trial #10. This contributes to a better performance of the full model also in that case. Such a cooperation is nevertheless characterized by a strong dominance of the Planning strategy in the behavior of the simulated agents (S5a Fig). Separating selection rates by trial types clearly shows that the model manages to increase the contribution of the Planning strategy when the intra-maze cue is hidden (hence when the Direction strategy is inefficient) and to decrease it during the competition trial in order to reduce the time spent at the previous location of the platform (S5b Fig).
A second line of simulation results can be illustrated when repeated simulations with the same parameter-set enable the full model to exhibit behavior alike to the two distinct populations of experimentally observed rats: cue-responders and place-responders (S5c Fig). The most important prediction of the model in this case is that the behavior of both populations should at the same time reflect a dominance of each individual’s preferred strategy (Planning for place-responders and Direction for cue-responders), but in neither group this behavior results from the complete absence of the other strategy (S5c Fig). In our simulations, the Direction strategy still contributed to 20% of the choices made by the simulated place-responders. Conversely, the Planning strategy contributed to nearly 15% of the choices made by the simulated cue-responders. An important consequence of this feature is that individual simulated trajectories within the competition trial #10 reflect an alternation between movements guided by the three different strategies (Exploration included; S5d Fig). This illustrates a cooperation between strategies, the Planning strategy being the one which attracts the simulated agent towards the previous platform location in this case (obviously more strongly for place-responders than for cue-responders). These results suggest that even experimental situations of apparent competition between navigation strategies can be solved through different degrees of cooperation, the respective contribution of each strategy being dynamically adjusted by the model to achieve the task properly.
We performed additional simulations to analyze the case where the model-based Planning strategy is replaced by a model-free Locale strategy, as in previous computational models [9, 17, 24]. As before, the model shows an increased contribution of the spatial strategy (here the Locale instead of the Planning) during trials where the intra-maze cue is hidden, and a strong decrease in its contribution during the competition trial to avoid losing time at the previous platform location (S5e Fig). Nevertheless, these results reveal an overall increase in the contribution of the Direction strategy (S5f Fig). This can be explained as a mechanism of compensation for the lower flexibility of the model-free Locale strategy compared to the model-based Planning one. Interestingly, this version of the model is still able to reproduce the experimental results, both in the control and lesion cases (S5g Fig). Like in Experiment I, we argue that this is made possible because the platform location is stable during all trials except test trial #10. A different prediction from this version of the model is that, while place-responders should still perform mixed strategies relying on a cooperation between strategies, the involvement of the Locale strategy should be much weaker in cue-responders, resulting in frequent homogeneous trajectories only controlled by the Direction strategy (S5h Fig).
As in Experiment II, the platform was moved every four trials (a session), but in this paradigm a proximal cue indicating the position of the platform was held at a constant distance and direction from it [36] (Fig 3d). This was meant as a way to enable the Direction strategy to also solve the task on its own, and hence to trigger a fair competition between Planning and Direction strategies. Central questions addressed by the authors of the original study are whether hippocampal lesions would specifically impair a particular strategy and how this would bias the competition. Interestingly, they observed that both control and hippocampus-lesioned animals were able to (at least partially) learn the task since they both show a gradual improvement in performance, as illustrated by the decrease in escape latencies across sessions (Fig 3e). This session-by-session progressive improvement in performance converged to a point where both groups reached similar espace latencies in the last sessions. The important observation is that the two groups showed different performance characteristics within each session. Control animals were able to display a fast adaptation (i.e., within 4 trials) to the new position of the platform at each session. In contrast, hippocampus-lesioned animals did not show a significantly different performance between the first and the fourth trial of each session (Fig 3e). Strikingly, hippocampus-lesioned animals were nevertheless better than control animals at the first trial of the session. Further analyses reveal that this is explained by the tendency of hippocampus-intact animals to spend time at the previous platform location [36]. This experiment thus reveals a set of intrincate phenomena which support a dual learning systems approach: the hippocampus appears as necessary to enable fast adaptation to new platform location; Nevertheless, lesion of the hippocampus led to a reduction in the time spent around the previous platform location; In consequence, both groups were eventually able to learn the task. One important computational question is how the model should balance the competition/cooperation between the strategies to produce such a performance?
As for Experiment III, our previously reported results in this paradigm [14] focused on whether output actions in the model should have an egocentric or an allocentric frame. Here we show new simulations to (i) further analyze the balance between cooperation and competition mediated by the gating network, and (ii) to see whether a model-free Locale strategy could solve the task similarly to a model-based Planning strategy in the model.
The combination of Planning and Direction strategies in the full model can reproduce the behavior of intact animals in this experiment (Fig 3f). Emulation of hippocampal lesions in the model—leaving only the Direction and Exploration strategies spared—can reproduce the behavioral performance of the lesioned animals in this task: a better performance than the control group at the first trial of each session; and a lack of fast adaptation between the first and the fourth trial of each session; hence an impaired performance compared to controls at the fourth trial of each session. Interestingly, as observed in our previous work [14], an artificial lesion to the striatum in the model—leaving only the Planning and Exploration strategies spared—predicts an impaired performance in this task (S6a Fig): while the fast adaptation between the first and the fourth trial of each session is preserved, the striatum-lesioned model shows larger escape latencies than controls and hippocampus-lesioned models at the first trial of each session. This directly results from the tendency of the Planning strategy to be attracted by the previous platform location at the first trial of each session. We found this tendency to be stronger in the behavior of the simulated agent in the striatum-lesioned model (P) than in the full model (DP) (S6b Fig). Another interesting property of the striatum-lesioned model is that it does not show the progressive improvement of performance across sessions seen in the two other models, and which could be the signature of a slow model-free learning process (S6a Fig). This constitutes a strong prediction of the model which could be tested experimentally.
Importantly, simulation results of each strategy alone enable us to well decompose the overall behavior of the control group into two clearly distinct components: a model-based learning component responsible for the fast within-session adaptation and a model-free learning component responsible for the slow across-session adaptation. Importantly, these two components are clearly visible in the performance of the full model (Fig 3f). It is thus interesting that the model, which has been mainly designed to regulate the competition between navigation strategies—since it gives full control of the movement to a single strategy at each timestep –, learns to achieve some degree of cooperation between strategies so as to benefit from the advantages of each of them. Plotting the rate by which each strategy is selected by the gating network during the first and the fourth trial of each session reveals how the model learned to operate this cooperation (S6c Fig). The Exploration strategy was more selected during the first trial of the two first sessions until the gating network learned to decrease its contribution to the movement. In parallel, the gating network learned to decrease the contribution of the model-free Direction strategy which is not yet efficient at the beginning of the experiment, and to increase the contribution of the model-based Planning strategy which can lead to fast adaptation. Very interestingly, from the second session onwards, the gating network learned to progressively reduce the contribution of the Planning in parallel to the improvement of the Direction strategy with learning. This resulted in the simulated agent spending less and less time at the previous platform location (S6b Fig). After the eighth session, the Direction strategy is selected more often than the Planning strategy, because it is now sufficient to successfully solve the task. This is the explanation that the model offers relative to the hippocampus-lesioned group in the experimental data which eventually reached the same performance as the control group in the last sessions (Fig 3e). Finally, it is worthy of note that the selection rate of the Planning strategy not only decreases during the first trial of each session, but also during the fourth one (S6c Fig). This is because the input that the gating network receives only provides it with information about visual cues and activity in the planning graph (Fig 1). The gating network is thus not able to discriminate between the different types of trials. A prediction from this is that any learning occurring in one type of trial will affect behavior in the other type of trials, which contributed here in making the Direction strategy more prevalent in the behavior of the simulated agents in the late sessions.
We further evaluate the predictions of this approach when the model-based Planning strategy is replaced by a model-free Locale strategy. Interestingly, the learning of the Locale strategy is too slow to learn the new platform position within only 4 trials, making the performance of this version of the model at the fourth trial not better than the Direction group (S6d Fig). Hence, as it was the case in the previous experiment, the important message is that fast adaptations within a few trials experimentally observed in animals are more likely to be well accounted for by a model-based learning strategy than by a model-free one.
The last two experiments presented here highlight the role that associative learning processes can play in navigation paradigms. In particular, both experiments were originally designed as attempts to experimentally contradict the cognitive mapping theory—relying on localization processes based on a constellation of distal cues—by showing that individual cues could induce associative phenomena previously observed in non-navigation learning paradigms to support the associative learning theory. These associative phenomena are generalization gradient and blocking effects, which we will define hereafter while showing at the same time that only a competitive interaction of the associative Direction strategy with others (e.g., Exploration, cognitive-mapping-based Planning strategy) can reproduce these effects, not an associative Direction strategy alone.
The spatial generalization gradient effect was studied by [37] in a navigation task involving a hidden platform under opaque water but marked by a proximal cue B, where a gradient of occupancy of the zone near cue B was recorded as this cue was progressively moved away from a distal cue F (Fig 4a). The authors expected a gradual loss of response to the proximal cue proportional to the distance increase. This decrease was supposedly due to the competition between cues—leading to a specific decrease of the proximal cue’s associative strength—rather than due to a competition between strategies.
The experimental protocol was composed of two training stages followed by one test trial. During Stage 1, a training of four sessions of eight trials was performed with two cues present, the proximal cue B (for Beacon) being initially close to the distal cue F (for Frame of reference). Stage 2 was composed of 10 sessions of nine trials each. In all sessions, eight of these trials were performed as in the previous stage (hereby termed escape trials). In the 9th trial of sessions 2, 4, 6, 8, and 10 (gradient trials), the platform was removed and the proximal cue B was rotated 0°, 45°, 90° or 135° from its original position (Fig 4a). This rotation was done either clockwise or counterclockwise, but the direction was kept constant for each animal. In the remaining sessions (1, 3, 5, 7, 9), the 9th trial was conducted with the F cue only, without the cue B nor the platform (extinction trials). These extinction trials were performed to reduce overshadowing of B by F, assumed to bias the generalization gradient.
The main experimental result shows that the greater the angle of the proximal cue B rotation during gradient trials, the less time the animal spent in the vicinity of this cue (i.e., a generalization gradient) (Fig 4b). In contrast, the occupancy of the area near the distal cue F did not exceed chance level. Thus, the proximal cue may have been overshadowing the distal one, and the obtention of the gradient suggests that the strength of the proximal cue was learned in an associative way. Nevertheless, during extinction trials rats occupied the octant F above chance level, hence revealing that behavior could still be under the control of the distal cue in the absence of the proximal cue.
Our model simulations suggest that only the competitive interaction between associative and cognitive mapping strategies could produce such effects. We found that Direction or Planning alone cannot reproduce the experimental results (Fig 5a). However, the modular approach allowing the selection among these two strategies in the full model was able to do so (Fig 4c). Analysis of the session-by-session evolution of the selection rate of each strategy reveals that the model could achieve this performance by progressively learning during Stage 1 that the Direction strategy is more efficient and accurate than the Planning strategy in this task and should thus be progressively more selected (Fig 5b). Indeed, plotting the escape latencies for the simulations with the Planning strategy alone shows a progressive improvement during Stage 1 followed by degradation of performance during Stage 2 (Fig 5d), which the model tried to compensate by selecting more and more the Exploration strategy instead of the Planning one (S7a Fig). This degradation of performance with the Planning strategy alone was not observed in the experimental data (Fig 5c) and only the simulations with the Direction strategy alone or with the full model (i.e., DP) could reproduce the performance during Stage 2 (Fig 5d). This suggests that the contribution of the Direction strategy was required to reproduce the characteristics of the learning process, and that within the Direction strategy the associative strength of cue B overwhelmed that of cue F (S7b Fig).
Nevertheless, simulations with the Direction strategy alone cannot reproduce the occupancy rates above chance level in octant F during extinction trials (Fig 5e). Chance levels were here obtained by simulating a Chance group, consisting of only one Exploration strategy, in the same conditions as the other groups. Only the full model and the Planning strategy alone could reproduce the animal behavior during extinction trials. Overall, only the full DP model could reproduce the ensemble of observed results in this experiment.
The selection rates of the strategies in the full model can give a further clue about the cooperation between Planning and Direction strategies which was employed to solve the task (Fig 5b). During Stage 1, selection rates indicate that both Direction and Planning strategies contributed to locate the platform. At the end of this stage, the gating network gave an advantage to the Direction strategy, but the Planning strategy remained selected at a rate above chance (36.4%). In the simulations without the Planning strategy, the Direction strategy was mainly helped by the Exploration strategy (averaged selection rate of 31%). At the beginning of this Stage, its performance was lower than in the full model—suggesting that the performance of the full model during these first trials was due to the cooperation between Planning and Direction strategies. This suggests that even if the Planning strategy was not the most efficient in this task nor sufficient to explain the experimental data alone, reproduction of rats’ performance by the full model still relies on the cooperation of the Planning strategy with the other strategies.
Importantly, the spatial generalization gradient effect in the full DP model was mainly due to the associative rules underlying the interactions between strategies rather than the associative rules within the Direction strategy itself (as the original authors hypothesized). S7c and S7d Fig detail strategy selections during gradient trials in groups DP (left) and D (right) by distinguishing the moments before the simulated agents had reached the octant B for the first time, and the moments after, when they occupied this octant and the other ones. Strikingly, as can be seen in the first column (“Before B”), the generalization gradient was not expressed until the simulated agents reached the octant B for the first time (light grey dashed line; no significant difference between test trials for 0°, 45°, 90° and 135°). The association between the proximal cue and the response leading to the goal was however well learned, as the simulated agents were able to reach the zone of the displaced proximal cue without any gradient. Yet the gradient itself was generated after, by the recruitment of other strategies when searching for the absent platform, i.e., without getting a reward (S7c and S7d Fig, right column “After B”, dark grey dashed line). This is also depicted by typical trajectories during gradient trials 45° and 135° (S7e and S7f Fig): octant B was rapidly reached with the Direction strategy and the Planning and Exploration strategies gave their contribution after, the former attracting the simulated agents towards octant F. These results contrast with the original authors’ hypothesis considering that the gradient resulted from a gradual loss of the associative strength of B during its learning. Unfortunately, the original experiment did not analyze the octant occupancy within gradient trials, but this prediction of the model would be easily verified.
This experiment proposed another associative task [38] to investigate the expression of spatial blocking. This effect was proposed to depend on the amount of training with both distal and proximal cues, and on the change of the physical characteristics of the proximal cue. The hypothesis was that the blocking of the distal cues by a proximal cue would be due both to the presence of the same proximal cue during the experiment (which could be tested by the replacement by a different proximal cue in a different group of animals) and to the weak reliability of distal cues during training (which could be tested by changing the number of trials available to learn the position of a hidden platform based on distal cues).
In this experiment [38], four groups of animals are defined: Session-Same, Trial-Same, Session-Diff, Trial-Diff. The experimental protocol is decomposed into three different experimental stages (Fig 4d). In Stage 1, animals learned to find a cued platform (with a proximal cue A) in the presence of surrounding distal cues. For Session animals, the platform was moved every session (a session being composed of four trials). For Trial animals, the platform was moved every trial, so that rats did not anchor their learning process on the distal cues, contrary to groups Session. In Stage 2, the platform remained at the same location and was signaled either by the same proximal cue A (Same animals) or a different proximal cue B (Diff animals). Lastly, in Stage 3, the platform and its attached proximal cue were removed and the time spent near the previous platform location was recorded.
The original experimental results showed the following main phenomenon: Only Trial-Same animals exhibited blocking (i.e., the time spent near the previous platform location lasted no more than chance level), whereas in other groups, animals spent more time near the previous platform location, demonstrating their learning of the platform location with the help of distal cues [38] (Fig 4e). The proposed explanations are that the two Session groups could not express blocking since, in Stage 1, distal cues were relevant to locate the platform. In contrast, distal cues were irrelevant for the two Trial groups during Stage 1, thus susceptible of being blocked. However, the change of proximal cue in Stage 2 in Group Trial-Diff prevented blocking, leaving Group Trial-Same as the only one to express a blocking phenomenon. A second important experimental result in [38] relates to the escape latencies of the animals. In the original article, the observed escape latencies were reported only for Stage 2. Groups Session-Same and Trial-Same showed no learning improvement (with respect to Stage 1), whereas groups Session-Diff and Trial-Diff expressed a re-learning of the association between the new proximal cue and the platform, resulting in larger escape latencies than groups Same during the first session of Stage 2 (i.e., Session 13) (Fig 6a).
Simulation of the full model (containing a Direction strategy D, a Planning strategy P and an Exploration strategy E) can reproduce these experimental results (Fig 4f): In the Trial-Same group, the selection module learned that the Planning was inefficient due to its poor performance in Stage 1 and was thus no more selected in the later stages (Fig 6g). In Diff groups, the change of proximal cue discarded the selection of the Direction strategies, thus leaving room for the Planning strategy to take place in Stage 3. And the Planning strategy was not discarded in Session conditions, because of its satisfactory performance in Stage 1. The full model could also reproduce the significantly different escape latencies during Stage 2 in the case of a different proximal cue B (Fig 6d). The model predicts that these escape latencies for the Diff conditions in the first session of Stage 2 should not be as high as those in the first session of Stage 1, thanks to the cooperation of strategies P and E which enabled some generalization between these two situations with a different proximal cue. This prediction can however not be confronted with the original article which does not show such comparison between Stage 1 and Stage 2.
We have also tested three other versions of the model in this task: A DL version where the model-based Planning strategy is replaced with a model-free Locale strategy; a D version where the Planning strategy is removed, leaving only the Direction and Exploration strategies; and a P version where the Direction strategy is removed. The latter completely fails to reproduce the experimental results because: (i) the time spent in the quadrant containing the previous platform location is significantly above chance in all conditions, unlike experimental results; (ii) the escape latencies during Stage 2 do not show an improvement and are significantly different between conditions Trial and Session, unlike experimental results (S8 Fig). A standard cognitive mapping approach is thus not appropriate to explain blocking in this context.
Interestingly, while the two other versions (models DL and D) can reproduce the escape latencies profile in Stage 2 (Fig 6e and 6f), as model DP also does, they lead to different predictions. The DL model predicts smaller escape latencies in the first session of Stage 2 compared to the first session of Stage 1, for the same reasons as the DP model. In contrast, the D model predicts even larger escape latencies since the Direction strategy starts Stage 2 with a performance lower than chance because of its synaptic weights resulting from learning during Stage 1 with a different proximal cue (S9a Fig). In addition, the D model predicts a quicker learning across sessions during Stage 1 than the two other models (Fig 6d–6f) because it is not polluted by the presence of an inefficient P or L strategy anchored on distal cues. Conversely, model P, but not model D, demonstrates a difference of performance between Session and Trial conditions (S8b Fig), and thus confirms that only the Planning strategy is able to learn within-sessions rather than across-sessions. This complementarity between the Direction and the Planning strategies is confirmed in model DP by the drastic improvement of performance between the first and the fourth trial of each session of Stage 1 in the Session-same condition compared to the Trial-Same condition (S9b Fig).
Most importantly, neither the DL model nor the D model can produce a blocking effect only in the Trial-Same condition (Fig 6b and 6c). Model D also showed a blocking effect in the Session-Same condition (unlike animals) because of the absence of a Planning strategy to rely on distal cues for the localization of the platform in this condition. S9c Fig illustrates how the DP model could avoid to express blocking in this condition by learning large weights and thus high confidence in distal cues used by the Planning graph (PG) when learning with the distal cues was possible for several trials (Session conditions). This results in the prediction that animals will hippocampal lesions should also show a blocking effect in the Session-Same condition, similar to model D.
Analysis of the behavior of the DP model during the test trial (Stage 3) provides further insights on the strategy coordination dynamics that may explain animal behavior in this task. We recorded the details of strategy selection in the model before reaching the quadrant of the platform and after, when the simulated agents occupied this quadrant and the others. Because in model DP the Direction strategy was preferred just before Stage 3, reaching the goal quadrant during the test trial mostly relied on this strategy (S9d Fig, 1st column of each condition), even if the corresponding proximal cue was absent, thus at random. The lowest selection rates of the Planning strategy (0.57%) were obtained during the Trial-Same condition. However, the Planning strategy was recruited after, when searching for the absent platform (S9d Fig, 2nd and 3rd columns). According to these results, in all conditions—and not only Trial-Same—did the model avoid to mainly rely on distal cues for reaching the goal quadrant in the absence of the proximal cue. Moreover, in all conditions—Trial-Same included—could the model quickly re-use distal cues after, which constitutes an interesting prediction of the model. Thus, in our simulations, the low occupancy rate of the goal quadrant in the Trial-Same condition could not be assumed to be due to a total blocking of the learning of distal cues—since distal cues were eventually used during the test trial –, but to a decrease in the confidence in these cues, acquired early in the experiment (as attested by S9c Fig, right). In model D, the blocking phenomenon was also expressed in the Session-Same condition, as expected by the authors but not observed in animals. A competition between strategies happened very early in the experiment, giving to the proximal cue a too high relevance during Stage 2 (S9a Fig). This reinforced the use of the Direction strategy in the Same conditions during Stage 3, leading to a longer time spent out of the goal quadrant (S9e Fig, 3rd column, Same compared to Diff conditions).
An important remaining question is whether alternative models employing a model-free Locale mechanism for the place-based strategy instead of the model-based Planning mechanism used here can also reproduce these results. Strikingly, unlike animals, model DL showed no blocking effect at all. This is because this task again involves a platform with a constant location, which gives an advantage to the Locale strategy over the Planning one by enabling the former to learn with more precision. As a consequence, while the Direction strategy is also the most selected in the DL model, the Locale strategy still contributes substantially to the behavior in the Trial-Same condition, hence preventing the blocking effect. Altogether, these results highlight that the complex mechanisms underlying the blocking effect in some conditions but not others can here not be reproduced by a purely associative model containing only a Direction strategy, but can instead be reproduced only by a modular approach which coordinates an associative strategy (here Direction) with a cognitive mapping strategy (here Planning). Only such a modular approach was in our simulations capable, like animals, of expressing both blocking and its absence.
In this work, we have presented a computational model for navigation paradigms combining a model-based Planning strategy, a model-free Direction strategy and a random Exploration strategy. The three strategies are coordinated by a gating network which learns in a model-free associative manner which strategy is the most efficient in each situation (i.e., depending on visual input and planning graph activity). The model could reproduce a set of behavioral and lesion data observed in navigating rodents in six different experiments (the main results are summarized in Table 1). The model can account for these data by achieving both competition and cooperation between strategies, which results in non-trivial behavior both within and across-trials. It is a striking feature of the model to be able, with a single coordination mechanism through a gating network, to produce both cases of competition between strategies, where lesion of one strategy leads to an improvement of performance, and cases of cooperation where two strategies together produce a better performance than each strategy alone. The model moreover permits a precise quantification of this cooperation/competition trade-off by plotting the evolution of weights assigned by the gating network to different strategies at different times across learning. This permits concrete predictions that could be tested experimentally. Importantly, these behavioral properties result from dynamic activation of different strategies. These dynamics were different from those obtained by a model composed only of model-free strategies, or of only a Direction strategy. The fact that these different variants of the model could not reproduce all the aimed experimental results highlights that a combination of navigation strategies of different nature is key to account for these experimental data. The simulations moreover yield a series of predictions which could be tested in future experiments to further assess the model (Table 2). We also summarize predictions raised by the alternative model DL, which only relies on model-free strategies, in order to guide future experiments that aim at further comparing the two models (Table 3).
Several previous models have used a Locale strategy [9, 23, 24, 46, 47], which associates places to movements without really building a cognitive map (no topological graph; see detailed comparisons in [8, 48]). Here we have shown that the Planning strategy better explains behavioral results observed in protocols involving frequent changes of goal location, because a model-based strategy is more flexible than a model-free one [4]. We also found that the Locale works well when the goal location is stable, suggesting a possible co-existence of the two strategies in a modular architecture. Such a co-existence has been previously discussed in [8], arguing that model-free learning processes involved in the Locale strategy could take place in the dorsolateral striatum, while model-based learning processes involved in the Planning strategy could take place in the hippocampus and prefrontal cortex. The possible involvement of the dorsolateral in a model-free place-based strategy is consistent with electrophysiological recordings showing that activity in the dorsolateral striatum correlates with place when the task requires knowledge of spatial relationships [49]. Inactivations of these different regions could be a way to test the different predictions raised in this manuscript relative to Planning versus Locale strategies.
Several previous models have already proposed a coordination of model-based and model-free reinforcement learning mechanisms to account for various rodent behavioral data [4, 11, 12, 18] and could thus be considered as possible candidates to model the experiments addressed here. Nevertheless, the Lesaint model [12] proposes a fixed coordination of MB and MF through time: each individual has a specific weight attributed to each system determining its contribution in decision-making. The models proposed by Daw [4], Keramati [11] and Pezzulo [18] do incorporate a dynamic coordination of MB and MF, based on uncertainty. Nevertheless, these models were designed to account for the sequential shift from initial goal-directed behavior to habitual behavior after overtraining, explaining the insensitivity in the latter case to outcome devaluation, which is a specific case of the questions addressed here. In Experiment IV studied in the present work, animals progressively learn to reduce their use of the cognitive mapping strategy, which we explain in the model by the fact that the gating network learned to use less and less this strategy at the first trial of each session to avoid being attracted by the previous platform condition. The Daw and Keramati models should in principle not be able to explain this because the uncertainty associated to the MB system should be lower and lower sessions after sessions, while uncertainty in the MF system should remain high because the platform changes location every four trials. Besides, the Pezzulo model biases its system coordination towards a default model-free control, which cannot explain why some actions remain under model-based control even after training, as argued in [4] and as observed in several experiments considered here (e.g., S3a, S4c, S5a, S7c–S7e, S9d Figs). Moreover, the fact that the gating-network of our model learns to coordinate strategies (which is not the case for these three other models) also enables the model to learn to increase the contribution of the Exploration system when necessary (S5b Fig), which corresponds to a dynamic exploration rate which is absent from these other models. Moreover, the generalization gradient in Experiment V is produced by the model at the level of the associative rules within the gating network (thus at the level of strategy coordination) rather than at the level of associative rules within the model-free Direction strategy itself. The Daw, Keramati and Pezzulo models proposed a coordination criterion which depends on instantaneously measured signals (i.e., uncertainty) rather than on learned signals (i.e., their models cannot learn that strategy X is efficient in a particular part of the environment while strategy Y is efficient in another part), hence they cannot reproduce this effect. Nevertheless, these models account for a variety of other experiments involving outcome devaluation, contingency degradation as well as hippocampal off-line replays, which our model does not address. Thus it would be particularly interesting in future work to study if combining mechanisms from all these models can account for a wider array of experimental data.
Most animal experiments have aimed at distinguishing between only two strategies (place-based versus associative), without subtly distinguishing subtypes of these two categories. Our model enabled to show that different subtypes of place-based strategies (i.e., planning versus locale) are more efficient/relevant depending on the protocol. Similarly, we have previously illustrated how different subtypes of response strategies (taxon, direction) which differ in the frame of reference for actions (resp. egocentric and allocentric) can also display complementary behavioral properties [14]. Together these computational results predict that new elaborated protocols should permit to isolate more than two concurrent strategies (for instance, planning+locale+direction or planning+direction+taxon). The common currency proposed here enables in principle to coordinate any number of strategies of any different nature, because the model just needs to be able to evaluate their current performance in different states of the task. Moreover, these various subtypes of strategies should engage different parallel memory systems (for instance subterritories of various cortico-striatal loops, depending on the input-output of these territories and their respective learning mechanisms). This predicts that specific lesions of these subterritories should affect only particular subtypes of strategies.
The present computational results have important implications relative to the debate between the cognitive mapping theory and the associative theory of spatial cognition in mammals [50, 51]. These two theories propose alternative mechanisms to explain spatial learning. According to the associative theory, spatial learning is dependent of a single type of mechanism—abundantly studied within the framework of classical and operant conditioning—by which a new response is incrementally acquired by the association of a stimulus and a reward [52–55]. In the associative paradigm, stimuli or group of stimuli available in the environment are assumed to compete to control animal navigation.Those which are not favored by this competition are not going to contribute to the achievement of the task. The cognitive mapping paradigm rather attests the existence of non-associative spatial rules (i.e., not incremental, independent from reward), in which all cues participate to develop a spatial representation [56]. This theory has received a strong support from the discovery of hippocampal place cells [57], which enables the animal to quickly build a reliable spatial representation of their environment [58], independently from the reward (latent learning). The debate between the two theories is still vivid in that the cognitive mapping paradigm is not able to explain blocking or overshadowing effects, and since the actual existence of such “cognitive map” enabling animal and humans to plan shortest paths or shortcuts aroused and still arouses controversies [39, 40]. On the other hand, opponents to the associative theory highlight a number of experiments failing to display overshadowing between proximal and distal cues [41, 42] or revealing potentiation between cues during attempts to look for spatial blocking and overshadowing (for a review, see [43]).
We have tried to show here that these theories could however be reconciled by a modular paradigm which proposes that both kinds of mechanisms may cohabit in distinct neural systems and may be learned in parallel [44]. Indeed, a large amount of studies have shown that inactivation of specific neural zones in rodents selectively impair only part of their navigational capacities [2, 44, 59–69]. The modular approach is also strengthened by several experimental procedures that have shown animals shifting from one type of spatial strategy to another one, either within a navigation trial, or as learning takes place across sessions [36, 44, 70–78]. This suggests the existence of mechanisms ruling the selection among navigation strategies in distinct neural structures from those which learn each strategy. In support of this view, lesions of prelimbic and infralimbic areas of the medial prefrontal cortex prevent the shift of a place-based strategy towards a cue-guided one but does not prevent the strategies themselves to be learned or displayed [79]. Similarly, lesion and electrophysiological studies of the ventral striatum suggest an evaluative role of the structure, important for initial learning and flexibility, but not necessarily a substrate for learning a specific navigation strategy (e.g., [62, 80]; see more thorough discussions in [7, 8, 81, 82]). The computational model proposed here constitute a refutable proposition concerning the mechanisms that may underly such a modular organization combining associative and cognitive mapping memory systems.
Several criticisms of the cognitive mapping theory have argued against the assumption that a global topographical representation (i.e., a cognitive map) exists and this information is available at all times during training. Whether this is a valid assumption and whether real rats benefit from such a representation is open to debate. However, it is important to emphasize that our computational model does not assume that a global map is learned. The mapping mechanism that we used rather focuses on the representation of areas that have been extensively visited [83], and it leads to local, partial and sometimes approximative maps that can produce suboptimal planning behavior in embedded, noisy tests [84]. Such a mechanism is supported by the observation that successful “planning” of trips does not necessarily depend upon a global representation (see, e.g., [85]). Moreover, a number of studies over the past 20 years have provided empirical evidence of local, non-global maps (e.g., [86–88], the first one providing clear evidence that non-global (at the very least) representations are involved in rodent spatial navigation in the water task). Related to this, recent work, both behavioral and physiological, has emphasized the important distinction between local boundary and distal landmark control [89–92].
These results also have important implications for the understanding of the coordination of learning and decision-making systems in humans, beyond spatial navigation. While we focused here on the modeling of experimental data in rodents for consistency, the coordination of model-based and model-free learning principles has also been highlighted in humans during instrumental learning tasks [93]. Moreover, cognitive mapping models have implications beyond spatial navigation, including roles in information contextualization [94], navigation between conceptual relationships in a manner similar to that of space [95], mapping of social relationships [96], and more generally in the integration of memories to guide future decisions [97]. Within this framework, an important question relates to the nature of the interaction between brain networks that underlies these cognitive functions. As mentioned above, previous contributions have emphasized the role of different parts of the striatum in different types of learning [5, 7, 8], the hippocampus being in a position to provide transition information between places for the building of model-based information in the medial prefrontal cortex and more ventromedial parts of the striatum [8]. Interestingly, studies in humans have demonstrated the recruitment of the striatum during learning with immediate feedback in a probabilistic learning task, and increased activation of the hippocampus with delayed feedback [98, 99]. Strikingly, in these tasks human subjects with Parkinson’s disease—whose striatum is known to be degraded—were impaired in learning from immediate but not delayed feedback. Such results appear consistent with the separation within the model between dorsolateral striatum-dependent model-free learning and hippocampus-prefrontal cortex-dependent model-based learning. Nevertheless, the precise role of different subparts of the prefrontal cortex in these learning processes is probably more difficult to disentangle. One currently attractive theory proposes that the orbitofrontal cortex participates to the learning of relationships between states within the model-based system, which in humans can also be useful to learn cognitive maps of non-spatial tasks [100]. In contrast, hippocampal projections to regions homologous to the dorsolateral and anterior cingulate prefrontal cortex are thought to play an important role in performance monitoring, with increased between-regions coherence upon task learning [101]. Such a process could relate to the performance monitoring mechanisms that underlie systems coordination within our gating-network. Nevertheless, more investigations would be required to further test the hypothesized roles of different prefrontal cortex subregions with respect to the different computations in the model.
The proposed coordination of learning systems also offers an opportunity to discuss about the possible role(s) of dopamine in mediating memory formation. Here, to be conservative, one could argue that the only role of dopamine on which to postulate relates to the production of phasic model-free reinforcement signals to update action values [102]. Following previous work on the combination of model-based and model-free learning in Pavlovian conditioning [12], we could further predict that dopamine blockade would only impair model-free navigation strategies, but not model-based ones, thus predicting similar behavior to the one shown through simulations of the Planning system alone. Such a prediction, specific to the navigation domain, could be interesting to experimentally test in order to further assess the model. Nevertheless, dopamine is known to play a role beyond the learning of action values based on reinforcement: For instance, it has been shown that dopamine contributes to the successful binding between experiences that are separated in time [103], which have been interpreted in terms of inference-based processes at the time of generalization. While dopamine reinforcement signals hypothesized to subserve model-free learning in our model could in principle slowly produce some binding between delayed events, notably through the association of reward values to stimuli and places that precede it, true off-line inference in the model relies on model-based processes (which enable action planning through a tree-search process [104]). Hypothesizing that dopamine plays no role in model-based learning [12] would at first glance fail to explain the coupled changes in learning-phase activity between the hippocampus and the dopaminergic system during information binding [103]. Nevertheless, the possibility to include in the model some off-line replay mechanisms—which permit another form of systems cooperation through the transfer of knowledge from model-based to model-free [105]—could be a promising extension of the model to explain off-line hippocampus drives over model-free dopaminergic learning signals [106] without using these signals for model-based learning per se.
Finally, some simplifications and limitations of the present model should be stressed in order to highlight possible ways to improve it. A first criticism that can be raised against the model presented here is the important number of parameters needed. Some of them need to be tuned differently according to the experiment (S2 Table). As a consequence, this can weaken the explanatory power of the model, that could be seen as an unnecessarily complex mixture of experts [107], where each strategy is considered as an expert whose selection becomes then irrelevant. In order to tackle this issue, we limited ourselves to two free parameters only, and changed their values within constrained boundaries. These parameters are the model-free learning rates of, respectively, the Gating Network and the Direction strategy—thus 2 parameters among a total of 18. The neurobiological meaning of such parameters (inherent to any RL model) has been investigated [108], and could account for motivational levels like, for instance, a stress induced by the experiment [109]. Moreover, it is not unreasonable to consider that animals may have changed their learning rates between task conditions [110]. While adding mechanisms to dynamically adapt learning rates based on some measures of the statistics of each task (such as reward volatility as done in [110]) would have added unnecessary complexity with respects to the phenomena investigated here—making the interpretation of our results more difficult –, one particularly interesting continuation of this work could consist in modeling neural systems responsible for such task monitoring and motivational effects and their influence on learning rates.
A second important limitation of the model is that it does not address the question of which precise model-free learning mechanisms should be employed. Instead, it rather focuses on the comparison at the global level between learning properties of model-based and model-free families of reinforcement learning algorithms [104]. Thus here we have not tested different types of model-free (MF) learning algorithms (e.g., Q-learning, Actor-Critic, SARSA). Comparing these different MF algorithms is particularly important when examining the precise profile of neural activity in different brain regions, as done by Hagai Bergman’s group and Geoff Schoenbaum’s group [111, 112] (see [113, 114] for extensive discussions) who investigated which of these different algorithms could best explain dopamine neurons’ phasic activity in instrumental learning tasks. Such an analysis goes beyond the present work and extensions of the model would be required to account for this. Nevertheless, in previous work, we have shown that these precise MF learning algorithms do not make very different predictions in terms of behavioral adaptation [115, 116], the behavior of animals in such tasks instead appearing to also rely on a more flexible MB learning algorithm. This is why the present study focuses on the comparison between learning algorithms of different natures (MB, MF, random exploration) to account for animal behavior.
In summary, we presented a new computational model of navigation that successfully reproduced a set of different experiments involving cognitive mapping and associative phenomena during spatial learning. The fact that these experimental results have for a long time been considered contradictory while they could here be accounted for by a unified modular principle for strategy coordination opens a promising line of research to systematically assess computational predictions of this type of modular computational models of navigation. This type of model can also be used to design new experimental protocols and assess new hypotheses about complex behavior arising from the interaction of different navigation strategies. In parallel, such models could contribute in translating important inspiration from animals’ behavioral flexibility to autonomous agents having to display fast adaptation to rapid changes in the environment from a small amount of data, a paradigm which has been called micro-data learning [117], by opposition to big data learning where the data perimeter is already known in advance. The computational work presented in this manuscript thus highlights the importance of cross-talk between disciplines interested in biological and artificial cognition to contribute to a better understanding of brain and behavior.
The model is sketched in Fig 1 and described in more details in S1 Text.
|
10.1371/journal.pcbi.1002769 | Functional Analysis of Metabolic Channeling and Regulation in Lignin Biosynthesis: A Computational Approach | Lignin is a polymer in secondary cell walls of plants that is known to have negative impacts on forage digestibility, pulping efficiency, and sugar release from cellulosic biomass. While targeted modifications of different lignin biosynthetic enzymes have permitted the generation of transgenic plants with desirable traits, such as improved digestibility or reduced recalcitrance to saccharification, some of the engineered plants exhibit monomer compositions that are clearly at odds with the expected outcomes when the biosynthetic pathway is perturbed. In Medicago, such discrepancies were partly reconciled by the recent finding that certain biosynthetic enzymes may be spatially organized into two independent channels for the synthesis of guaiacyl (G) and syringyl (S) lignin monomers. Nevertheless, the mechanistic details, as well as the biological function of these interactions, remain unclear. To decipher the working principles of this and similar control mechanisms, we propose and employ here a novel computational approach that permits an expedient and exhaustive assessment of hundreds of minimal designs that could arise in vivo. Interestingly, this comparative analysis not only helps distinguish two most parsimonious mechanisms of crosstalk between the two channels by formulating a targeted and readily testable hypothesis, but also suggests that the G lignin-specific channel is more important for proper functioning than the S lignin-specific channel. While the proposed strategy of analysis in this article is tightly focused on lignin synthesis, it is likely to be of similar utility in extracting unbiased information in a variety of situations, where the spatial organization of molecular components is critical for coordinating the flow of cellular information, and where initially various control designs seem equally valid.
| The organization of cooperating enzymes into complexes is a pervasive feature of metabolism. In particular, this phenomenon has been shown to participate in the regulation of flux through the networks of both primary and secondary metabolism in plants. It remains a challenging task to unravel the organizing principles of such “metabolic channels,” which can be temporary or persistent, and to understand their biological function. In this article, we analyze metabolic channels in the biosynthetic pathway of lignin, a complex polymer that stiffens and fortifies secondary cell walls within woody tissues. This system is well suited because the present analysis can be based on a computational, experimentally validated model demonstrating that several enzymes are spatially associated into channels specific for the production of two lignin monomers. To characterize the functioning of these channels, we develop a novel computational approach that is capable of identifying interesting structural and regulatory features of metabolic channeling and permits the formulation of targeted and readily testable hypotheses. Since the spontaneous or controlled assembly of molecules into functional units is known to occur in many biological contexts where information flow is tightly coordinated, the proposed approach might have broad applications in the field of computational systems biology.
| Lignin is a phenolic heteropolymer in the secondary cell walls of vascular plants. It is derived mainly from three hydroxycinnamyl alcohol monomers, namely p-coumaryl, coniferyl, and sinapyl alcohols, which, when incorporated into the lignin polymer, give rise to p-hydroxyphenyl (H), guaiacyl (G), and syringyl (S) subunits, respectively. The development of lignin biosynthesis is considered to be one of the key factors that allowed vascular plants to dominate the terrestrial ecosystem [1]. This evolutionary advantage is in part due to the fact that lignin, when deposited in the cell wall, contributes to the structural integrity of the cell, facilitates transport of water and minerals through the tracheary elements, and serves as a defensive barrier against pathogens and herbivores [2].
In addition to lignin, the secondary cell walls of vascular plants contain several polysaccharides, such as cellulose and hemicellulose. Extraction of these polymers from lignocellulosic biomass for biofuel production has attracted extensive interest partly because exploitation of this source of fermentable sugars could minimize the competition for food, which has been criticized in the case of corn- or sugarcane-based biofuel production. However, the natural resistance of lignocellulosic biomass to enzymatic or microbial deconstruction has rendered the task of generating sustainable and cost effective biofuels from lignocellulosic feedstocks very challenging. While impressive advances have been made toward the reduction of biomass recalcitrance [3], [4], it was also shown that the amount of fermentable sugars released through chemical and enzymatic treatments is inversely proportional to that of lignin present in biomass, and that some transgenic plants with reduced lignin content yield up to twice as much sugar from their stems as wild-type plants [5]. These observations suggest that lignin biosynthesis may be targeted for generating engineered crops with reduced recalcitrance. The challenge of this task derives from the fact that a rational design of less recalcitrant varieties would require a thorough, multi-level understanding of the lignin biosynthesis in wild- type plants, which we do not yet have. Such an understanding would include a grasp of the system of interactions between the enzyme-encoding genes, proteins and metabolites involved in the biosynthesis of lignin, as well as details of the regulation of this multi-tiered system.
The metabolic scaffold for the biosynthesis of the three building blocks of lignin originally was seen as a grid-like structure [6], but this initial structure has been revised and refined and is now understood as an essentially linear pathway with only a few branch points (Figure 1). Although this generic pathway structure is now widely accepted, it has become clear that different lineages of vascular plants have evolved variants that engage distinct biosynthetic strategies. An interesting example is the model legume Medicago truncatula, where the characterization of two distinct cinnamoyl CoA reductases, CCR1 and CCR2, has suggested parallel routes from caffeoyl CoA to coniferyl aldehyde (Figure 1) [7]. A more unusual case is the lycophyte Selaginella moellendorffi. Functional analyses of the two enzymes recently discovered from this species, SmF5H and SmCOMT, support the notion that S. moellendorffi may have adopted a non-canonical pathway from that in angiosperms to synthesize coniferyl and sinapyl alcohol (Figure 1) [8], [9], [10].
Given such variations, it would appear reasonable to consider genus- or species-specific similarities and differences. However, such data are seldom available, and even if a customized pathway structure can be established, its regulation often remains obscure. This shortcoming tends to become evident with new, precise data. For instance, experiments using genetically modified M. truncatula lines with reduced CCR1 activity exhibited an unexplainable decrease in the ratio of S to G lignin over wild type [7]. Such discrepancies between expectation and observation suggest that the currently accepted pathway diagrams may require further revisions that include regulatory mechanisms affecting the physiological outcome when the pathway is perturbed.
The focus of this article is an assessment of such a regulatory system associated with lignin biosynthesis in Medicago. This genus includes model species like M. truncatula, as well as alfalfa (Medicago sativa L.), an important forage legume. Medicago is particularly suited for these studies, because comparatively extensive information is available. For instance, a detailed dataset was established that characterized different lines in which seven lignin biosynthetic enzymes were independently down-regulated, and the resulting lignin content and monomer composition were determined in several stem segments [11]. In a recent study, we demonstrated that these types of data contain substantial, although hidden, information. In particular, we used these data to show that certain enzymes may co-localize and/or assemble into two independent channels for the synthesis of G and S lignin, and that salicylic acid acts as a potential regulatory molecule for the lignin biosynthetic pathway [12].
Although these earlier results provided significant insights into the mechanisms of regulation in this pathway, several critical questions, especially regarding the biological function as well as the operating mode of the channels, remain unanswered: For instance, are these channels always active in vivo? Are they sufficient to explain all available data in Medicago? Is there crosstalk between them, and if so, how is it organized? Exploring all pertinent scenarios associated with such questions would be experimentally intractable because they are simply too numerous.
Instead, we present here a novel computational approach to investigate exhaustively all regulatory schemes involving the key reactions associated with G and S channels in the lignin biosynthetic pathway (Figure 2). The specific hypothesis is that the formerly postulated and validated channels may have two different modes of operation. Either they are permanent in a sense that the component enzymes are persistently assembled into a complex; such a complex could be realized through membrane co-localization, thereby ensuring that the corresponding alcohol is always synthesized. As an alternative, the channels could be facultative, thereby displaying a functionality that depends on the sub-cellular localization of the component enzymes and the metabolic milieu. This hypothesis, in turn, leads to 19 possible topological configurations (Figure 3A). For each of these topologies, we consider an additional level of regulation, involving individual or combined regulatory mechanisms that may serve as a means of “crosstalk” between the two channels (Figure 3B). The emphasis of this approach is on mechanisms at the metabolic level, but one must not forget that the transcriptional network governing the system could be involved in regulation of the pathway as well [13].
The goal is thus to assess and compare the functionality of all given combinations of topological configurations and crosstalk patterns, each of which we call a design. To obtain insights that are independent of parameter choices, we constructed for each design a library of 100,000 loosely constrained dynamic models and tested each of them against the observed ratios of S to G lignin in four lignin-modified Medicago lines. The resulting analysis of hundreds of designs and millions of models led to the intriguing hypothesis that either a single activation mechanism or a dual-inhibition mechanism lies at the core of all experimentally supported designs. The former mechanism was not supported by an in vitro enzyme assay, while the latter is consistent with several lines of evidence from Medicago and other species. As an added insight, the analysis suggested that functionality of the G lignin channel is more important than that of the S lignin channel. Overall, these findings not only enrich our current understanding of how lignin biosynthesis is regulated, but they also demonstrate the possible application of the proposed approach in entirely different biological scenarios where the task is to identify true regulatory circuit among many theoretically feasible designs that depend on the functionality and localization of interacting molecules.
The base scaffold on which the different topological variants were built is shown in Figure 2. It consists of all relevant steps in the lignin biosynthetic pathway that possibly affect the relative amounts of G and S lignin. Prior work [12] provided evidence that CCR1 and cinnamyl alcohol dehydrogenase (CAD) may organize into a functional complex through which the substrate coniferyl aldehyde is transferred from CCR1 to CAD without much leakage, thereby acting as a channel leading specifically to the synthesis of G lignin. Similarly, caffeic acid O-methyltransferase (COMT) and ferulate 5-hydroxylase (F5H) were suggested to form an analogous complex contributing specifically to the synthesis of S lignin. These two complexes, which we called G and S lignin channels, are represented in Figure 2 as two directed edges, one linking feruloyl CoA and coniferyl alcohol (G channel) and the other linking caffeyl aldehyde and 5-hydroxyconiferyl aldehyde (S channel). The experimentally validated channeling hypothesis permits 19 different topological configurations (Figure 3A) that satisfy the following constraints. First, at least one edge must be leaving caffeyl aldehyde and feruloyl CoA, and at least one edge must be entering coniferyl alcohol and 5-hydroxyconiferyl aldehyde; otherwise mass would unduly accumulate in intermediate pools. Second, if coniferyl aldehyde can be produced by a free CCR1 and/or COMT, it must also be consumed by a free enzyme, thereby decreasing the metabolic burden that would otherwise be imposed on the cell. For reasons that will be explained below, we also consider for each topological configuration various crosstalk patterns between the CCR2/COMT and CCoAOMT/CCR1 pathways. Each pattern is composed of documented or postulated mechanisms of metabolic regulation (activation or inhibition) (Figure 3B). The specific combinations of topological configurations and crosstalk patterns lead to hundreds of different designs, which were analyzed and compared (see Figure S4).
For each design, we first constructed 100,000 Generalized Mass Action (GMA) models by randomly sampling loosely-constrained parameter combinations from a parameter space that was deemed biologically realistic. A notable feature of this approach was that the parameter space was not only constrained at the level of individual parameters (e.g. kinetic orders), but also at the level of steady-state fluxes. For instance, the ratio of fluxes leading to S and G lignin was fixed at a value observed in the wild-type Medicago species (see Materials and Methods and Text S1 for details). Once all parameters for a given GMA model instantiation were specified, we determined steady-state fluxes under conditions that mimic CCoAOMT and COMT down-regulated alfalfa lines as well as ccr1 and ccr2 M. truncatula mutant lines and computed the S/G ratios for which we had experimental data. We declared a model as valid if it yielded quantitatively and qualitatively correct results for both transgenic alfalfa and M. truncatula plants (see Materials and Methods). To assess the robustness of a design to parametric perturbations, we defined Q as the total number of valid model instantiations.
As a reasonable baseline, we first assumed the absence of crosstalk between the CCR2/COMT and CCoAOMT/CCR1 pathways (Figure 4). Of all possible topological configurations lacking crosstalk, only six had at least one parameter combination that yielded quantitatively correct predictions of S/G ratios for CCoAOMT and COMT down-regulated alfalfa plants. Supporting our previous findings [12], all six configurations include either one or both channels, suggesting that the channels are necessary. In other words, the pathway models are consistent with the observed changes in the S/G ratios of CCoAOMT and COMT down-regulated alfalfa plants only if at least one channel is present.
To assess these initially feasible parameter combinations further, we used the models with these parameter values to predict the S/G ratios for ccr1 and ccr2 knockout mutants. The M. truncatula lines harboring transposon insertions in CCR1 and CCR2 show a corresponding reduction in CCR1 and CCR2 activity, and their S/G ratio is decreased or increased, respectively, compared to the wild-type level [7]. Moreover, the activities of CCR1 and CCoAOMT, as well as their mRNA transcripts and proteins, are increased in the ccr2 knockout mutant, indicating that part of their activation might be processed through a hierarchical control of gene expression [14]; see also Figure S4.
Figure 4 shows simulation results for those topological configurations where at least one out of 100,000 randomly parameterized models yielded quantitatively correct predictions of S/G ratios for both CCoAOMT and COMT down-regulated alfalfa plants. In these plots, a model is valid only if its predicted S/G ratios for ccr1 and ccr2 knockout mutants fall within the northwest quadrant.
In the case of no hierarchical regulation, i.e., the ccr2 mutant exhibits only reduced CCR2 activity, some model instantiations from configuration A showed a decreased S/G ratio for the ccr1 knockout mutant, but not a single case exhibited an increased S/G ratio for the ccr2 knockout mutant. This outcome did not improve much when hierarchical regulation was considered: not one of the 1.9 million model instantiations from the 19 possible configurations yielded qualitatively acceptable predictions for both ccr1 and ccr2 knockout mutants. These findings indicate that the S and G channels alone are not sufficient to explain all available transgenic data, and that some type of crosstalk is highly likely to occur between the CCR2/COMT and CCoAOMT/CCR1 pathways.
One potential source of crosstalk between the CCR2/COMT and CCoAOMT/CCR1 pathways is substrate competition. CCR1/2 converts hydroxycinnamoyl CoA esters to their corresponding cinnamyl aldehydes, whereas CCoAOMT and COMT together complete the methylation of the aromatic C3 and C5 positions of the aldehydes and alcohols (Figure 1). All these enzymes are known to be multi-functional, acting upon multiple substrates with distinct catalytic efficiency. Because of their promiscuous nature, different substrates compete with each other if the supply of enzyme is limited. As a consequence, the enzymatic conversion of one substrate is effectively subjected to competitive inhibition by another substrate, and vice versa. This type of cross-inhibition is not necessarily equally strong in both directions because a promiscuous enzyme often displays preference for some substrates over others.
In the case of lignin biosynthesis, two regulatory mechanisms could arise from substrate competition. First, recombinant Medicago CCR2 exhibits similar kcat/KM values for caffeoyl CoA (0.49 µM−1•min−1) and feruloyl CoA (0.40 µM−1•min−1) [7], suggesting that the CCR2-mediated conversion of caffeoyl CoA to caffeyl aldehyde in Medicago might be competitively inhibited by feruloyl CoA (Figure 3B; Mechanism 1). Furthermore, CCR2 is inhibited by feruloyl CoA at a concentration above 20 µM [7]. Conversely, it is highly unlikely that the CCR1-mediated conversion of feruloyl CoA to coniferyl aldehyde is significantly affected by caffeoyl CoA, because CCR1 has a kcat/KM value for caffeoyl CoA (0.019 µM−1•min−1) that is 60 times lower than that for feruloyl CoA (1.14 µM−1•min−1) [7].
Second, the methylation of caffeoyl CoA by the combined activity of COMT and CCoAOMT may be subject to weak competitive inhibition by caffeyl aldehyde (Figure 3B; Mechanism 2). This assumption is based on the following observation. Although the combined O-methyltransferase (OMT) activity against caffeoyl CoA in extracts from internodes 6 to 8 of CCoAOMT-down-regulated alfalfa was reduced by 4.2-fold compared with the empty vector control line, about ∼25% of OMT activity remained [15]. This activity is presumably associated with COMT, for which caffeyl aldehyde is the preferred substrate. Notably, both mechanisms are independent of each other and may work individually or collaboratively to establish crosstalk between the two channels, thereby leading to three different crosstalk patterns and 57 different designs.
In the case where only Mechanism 1 (Figure 3B) was incorporated in the design, we observed a substantial increase in the number of model instantiations showing a decreased S/G ratio for the ccr1 knockout mutant (Figure S1). Yet, even when we accounted for the effect of hierarchical regulation, none of the models was capable of delivering a qualitatively correct change in the S/G ratio for the ccr2 knockout mutant. This finding indicates that the experimentally inferred inhibition evidently exists but is not sufficient. Similarly, we found no valid models when Mechanism 2, either by itself or coupled with Mechanism 1, was employed (Figures S2 and S3). An explanation may be that, with caffeyl aldehyde inhibiting the 3-O-methylation of caffeoyl CoA, knocking down CCR2 activity will consistently lead to a deregulation of CCoAOMT by caffeyl aldehyde, thereby increasing the flux to G lignin and reducing the S/G ratio.
One could surmise that the 3-O-methylation of caffeoyl CoA, for which CCoAOMT is the primary enzyme, is actually activated by caffeyl aldehyde. This conjecture is based on the following argument. When the production of S lignin is compromised due to a knockout of ccr2, the only way of raising the S/G ratio beyond its wild-type level is to further reduce the flux through the CCoAOMT/CCR1 pathway, which can be accomplished if CCoAOMT is activated by caffeyl aldehyde. The simulation results using this type of postulated mechanism, either by itself (Figure 5) or coupled with the documented inhibition of CCR2 by feruloyl CoA (Figure 6), are very intriguing: For each crosstalk pattern where millions of randomly parameterized models were generated, we found thousands of valid instantiations that yielded quantitatively and qualitatively correct predictions for both transgenic alfalfa and M. truncatula plants. Perhaps more surprisingly, only six topological configurations (A, B, E, F, I, O) had at least one valid model (Q>0; see Materials and Methods). To ensure that this result was not due to the use of overly restrictive thresholds, we relaxed the criteria and found more parameter combinations that qualified. Nevertheless, the same six topological configurations always passed the screening test by a wide margin (Table S1). Collectively, these findings suggested that this activation mechanism, acting alone or with the inhibition of CCR2 by feruloyl CoA, is necessary for consistency with the ccr1 and ccr2 knockout data. This conclusion immediately translated into a targeted hypothesis that was independent of specific parameter choices and readily testable by experiment.
To examine whether caffeyl aldehyde indeed activates CCoAOMT, we expressed alfalfa CCoAOMT in Escherichia coli and assayed the purified recombinant enzyme with caffeoyl CoA as substrate and caffeyl aldehyde as the putative activator. As shown in Figure 7, the CCoAOMT activity increased by 16% at 2 µM of caffeyl aldehyde and 20 µM of caffeoyl CoA; at higher substrate concentrations (i.e., 30 and 40 µM of caffeoyl CoA), the increase in mean CCoAOMT activity became less. Assays using lower concentrations of the substrate caffeoyl CoA (2, 4, 5 and 10 µM) and the putative activator caffeyl aldehyde (0.5, 1, 2 and 4 µM) showed no increase in CCoAOMT activity compared to the reaction without caffeyl aldehyde (data no shown). The maximal activation achieved in vitro was only 16%, which may not be biologically significant.
Since a direct activation of CCoAOMT by caffeyl aldehyde was not observed with recombinant enzymes, we tested other regulatory mechanisms by themselves and in combination with known mechanisms. According to one possible mechanism, based again on the concept of substrate competition, caffeoyl CoA could be a competitive inhibitor for the 3-O-methylation of caffeyl aldehyde (Figure 3B; Mechanism 4). This proposal agrees with the fact that CCoAOMT may contribute up to ∼10% of the methylation reaction in alfalfa [15]. In addition, evidence in ryegrass (Lolium perenne) points to the possibility of COMT being inhibited by different substrates, such as caffeyl aldehyde and 5-hydroxyconiferyl aldehyde [16]. Interestingly, substrate inhibition by caffeyl alcohol and 5-hydroxyconiferyl alcohol has also been observed in Selagniella moellendorffii COMT [9]. Thus, we hypothesized that COMT might be inhibited by caffeyl aldehyde (Figure 3B; Mechanism 5) in Medicago as well; direct evidence supporting this hypothesis in Medicago remains to be determined.
In total, there are 24 = 16 different crosstalk patterns that can result from the combination of four independent regulatory mechanisms (Figure 3B; Mechanisms 1, 2, 4 and 5). However, only four of them, when combined with the same six topological configurations (A, B, E, F, I and O) that were identified previously (cf. Figures 5 and 6), gave rise to designs with at least one valid model instantiation (Figure 8). Interestingly, all these crosstalk patterns require that caffeyl aldehyde is an inhibitor of the 3-O-methylation of both caffeoyl CoA and itself (Figure 3B; Mechanisms 2 and 5), providing computational evidence that this synergy between the two seemingly unrelated mechanisms is necessary for consistency with the ccr1 and ccr2 knockout data. Indeed, with respect to the ccr2 knockout, such a combination of two inhibition mechanisms appears to have a similar ultimate effect as a single activation mechanism (see Discussion section).
Inspecting the crosstalk patterns giving rise to at least one design with valid model instantiations (rows colored in red in Figure 8), one might surmise that caffeyl aldehyde would accumulate to an unduly high level, because Mechanism 5, which is employed in all these patterns, reflects substrate inhibition of COMT by caffeyl aldehyde. To examine the validity of this inference, we checked, for all designs with valid model instantiations, the predicted changes in caffeyl aldehyde under conditions that mimic the down-regulation of four lignin biosynthetic enzymes. As shown in Figure 9, it appears that down-regulation of CCoAOMT or COMT is consistently associated with a lower caffeyl aldehyde level compared with wild type, regardless of the crosstalk pattern being considered. Similarly, knocking out ccr2 consistently raises the caffeyl aldehyde level in all crosstalk patterns examined. However, in the case of the ccr1 knockout mutant, the results are mixed in a sense that some crosstalk patterns are associated with significantly higher caffeyl aldehyde levels, whereas others are associated with only modest changes. Interestingly, both crosstalk patterns suffering from an undue accumulation of caffeyl aldehyde contain Mechanism 1. By contrast, this mechanism is absent from other patterns, which maintain a relatively stable caffeyl aldehyde level. This finding suggests that the control pattern in Mechanism 1 may disrupt the metabolic homeostasis via accumulation of caffeyl aldehyde when CCR1 drops below its normal level. As any cellular system is constantly afflicted by a variety of intrinsic and extrinsic noises, this type of fluctuation must be expected to occur frequently and spontaneously, suggesting that Mechanism 1 is disadvantageous.
Investigation of the six robust topological configurations, which contain at least one valid model instantiation, revealed interesting structural features of the pathway. In particular, the G lignin channel is common to all robust designs and thus may be considered critical for the proper functioning of the pathway, at least for the cases studied. The evolutionary conservation of such a feature, one may argue further, is not due to the fact that it cannot possibly be altered, but that this particular design can sustain maximally tolerable changes and variability in other features [17]. These arguments lead to an interesting follow-up question, namely: Are the robust topological configurations related in an evolutionary sense?
To address this question, we constructed a “topology graph” where each node corresponds to a topological configuration. Two nodes are connected if the corresponding topological configurations differ only by one edge. For instance, configurations A and B are directly linked to each other because the only difference between them is whether caffeyl aldehyde can be converted, via free COMT, to coniferyl aldehyde. In other words, moving from a node to its neighbor may be considered a singular evolutionary event where an enzyme's preferred mode of action is changed.
Two outcomes are possible for the structure of such a topology graph. First, the graph may be disconnected, that is, there exist pairs of topological configurations such that no evolutionary path (defined as a series of evolutionary events) connects one to the other. In the most extreme case, the graph would consist exclusively of isolated nodes. Second, the graph is fully connected, so that any pair of topological configurations is connected by at least one evolutionary path. As shown in Figure 10, the actual topology graph of the six robust configurations of lignin biosynthesis is indeed connected, and so is the graph of all configurations, except for design S. This interconnectedness can be interpreted as facilitating the evolvability of the system [17], because the gain or loss of specific features that are needed to produce phenotypically novel traits will be tolerated and survive during evolution if robustness is preserved. Of course, this evolutionary aspect, which was derived purely with computational means, will require additional analysis.
The spatial organization of cooperating enzymes, known as metabolic channeling, has long been recognized as an effective means of regulation in primary and secondary plant metabolism [18], [19], [20]. This channeling phenomenon involves the organization of enzymes into complexes and/or the co-localization of enzymes at the plasma membrane or on the surfaces of organelles, as was demonstrated for the two initial enzymes, L-phenylalanine ammonia-lyase (PAL) and cinnamate 4-hydroxylase (C4H), in the phenylpropanoid pathway [21], [22]. Interestingly, some complexes or interactions are persistent, while others are temporary. In fact, many of the component enzymes such as PAL may be operationally soluble and are therefore only facultatively channeled. Such short-lived or dynamic complexes, while being readily responsive to the metabolic status of the cell, are inherently difficult to study with existing or emerging experimental models.
Using the lignin biosynthetic pathway as a model system, we propose here a novel strategy for studying metabolic channeling in unprecedented detail. Specifically, we consider all possible modes of action for both the G lignin and S lignin channels, and these can be mapped into 19 different topological configurations (Figure 3A). Metabolic channeling is clearly not the only process that affects the functionality of this system, and it is therefore necessary to study control processes affecting a channeled system. In the present case, this control is potentially exerted by individual or combined mechanisms of crosstalk between the CCR2/COMT and CCoAOMT/CCR1 pathways (Figure 3B). Some of these were documented in the literature, while others were hypothesized. Taken together, a topological configuration and a specific crosstalk pattern constitute a design. We evaluated each design with or without consideration of non-allosteric or hierarchical regulation which could involve transcription, as well as a variety of non-transcriptional processes such as phosphorylation, methylation, and targeted degradation of proteins and mRNA.
Ideally, the comparative assessment of design features would be entirely symbolic and independent of specific parameter values. However, systems of a realistic size are rarely analyzable in such fashion. As a reasonable alternative, we analyzed the possible design space comprehensively with widely varying parameter values, which resulted in a computational analysis of millions of models from hundreds of designs. This analysis yielded several interesting findings.
Importantly, it predicted that CCoAOMT is directly or indirectly activated by caffeyl aldehyde. This piece of information by itself is essentially unbiased, but insufficient to explain the exact mechanism of regulation. Nevertheless, it offered a specifically targeted hypothesis and was therefore experimentally testable. However, the hypothesis of a direct activation was refuted by subsequent experiments using the recombinant Medicago CCoAOMT, which failed to provide evidence confirming the putative role of caffeyl aldehyde as an allosteric activator. It might still be possible that activation exists in vivo, but it seems more likely that the activation is indirect rather than direct.
As a possible mechanism, the design analysis suggested that caffeyl aldehyde inhibits the 3-O-methylation of both caffeoyl CoA and itself. Several lines of evidence, although not exclusively from Medicago, support this computational prediction. Most importantly, the same six topological configurations were identified in the indirect design analysis and in the initial analysis of a putative activation mechanism. However, the two most parsimonious mechanisms differ in their proposed control strategies. The original analysis suggested just one activation mechanism, while the second analysis proposed two inhibition mechanisms. To some degree, these two mechanisms have the same ultimate effect. If ccr2 is knocked out, the flux entering the CCR2/COMT pathway and the subsequent synthesis of S lignin decline. The only possibility to increase the S/G ratio is to reduce the flux entering the CCoAOMT/CCR1 pathway. This task can be accomplished either through a diminished activation, as suggested for the single activation mechanism, or through an enhanced inhibition, as suggested for the dual-inhibition mechanism. The latter mechanism seems sufficient to restore consistency with the data, but it is of course possible that more complicated control patterns are present.
The computational analysis suggests that the G lignin channel is necessary for the system to respond correctly and robustly to certain genetic perturbations. By contrast, the S lignin channel appears to be dispensable. This theoretical deduction is indirectly in line with the fact that S lignin has arisen much later in the evolution of higher plants than G lignin [1]. It is also consistent with the observation that its formation, which in many plant species is dictated by F5H expression [23], [24], [25], is directly regulated by a secondary cell wall master switch NST1/SND1 and not by MYB58, a SND1-regulated transcription factor that can activate other lignin biosynthetic genes [26]. It could also be possible that S lignin, which is specifically involved in the pathogen defense of some plants [27], was relatively recently recruited for lignin biosynthesis and thus may not be essential for plant growth. Evidence supporting this postulate includes an Arabidopsis NST1/SND1 double knockout mutant that shows a complete suppression of secondary cell wall thickening in woody tissues, including interfascicular fibers and secondary xylem, but otherwise grows quite well as compared to the wild-type plants [28].
Within an evolutionary context, the multiplicity of robust solutions can be represented with a graph representation that connects any two (robust) topological configurations differing by a single edge. This graph is reminiscent of the “neutral network” concept that was initially proposed in genotype-phenotype models for RNA secondary structures [29] and protein folds [30], but also more recently extended to Boolean models for gene regulatory networks [31]. In the case of proteins, neutral networks are defined as sets of amino acid sequences that are connected by single-mutation neighbors and that map into the same tertiary structure. Such degeneracy of the mapping from genotype to phenotype allows a neutral drift in genotypic space, which is critical for accessing adjacent neutral networks with novel phenotypes that may confer higher fitness to the cells. As of yet, it is unclear whether individual plants within a Medicago population use the same or different designs, or whether the response to selected perturbations is an adequate phenotypic feature. Further investigation of the protein-protein interactions between lignin biosynthetic enzymes is thus necessary to confirm that a G lignin channel is indeed necessary for optimal functioning.
The work in this article describes a novel computational approach that shows promise in deciphering the principles of channel assembly in a biosynthetic pathway when relevant information is limited. It also provides a clear direction in which to proceed with more targeted experiments. Beyond the application described here, the proposed strategy might be beneficial in entirely different biological contexts, such as gene regulatory and signaling networks, where the task is to analyze how information flow is controlled by the spatial organization of molecules in the cell.
Since the two metabolic channels of interest are assumed to affect only the relative amounts of G and S lignin, the analysis is restricted to those critical steps within the lignin biosynthetic pathway system that govern the flow of material either toward G or S (Figure 2). For each possible design, we first formulate the corresponding generalized mass action (GMA) model [32], [33] in a symbolic format, where each intermediate is represented by a dependent variable and each enzymatic process by a product of power-law functions. An important reason for this choice of a modeling format is that it is mathematically sound and minimally biased, because it does not require the a priori specification of a biological mechanism [34]. The model contains either six or seven dependent variables, depending on whether coniferyl aldehyde is explicitly included, and 10 to 16 distinct power-law terms, depending on the topology in a specific design. Also, there are six independent variables, each of them representing the extractable activity of an enzyme. Each power-law term contains two different types of parameters: a non-negative rate constant γi that represents the turnover of reaction i, and kinetic orders fi,j, each of which characterizes the effect of a variable Xj on a reaction i. A kinetic order can take any real value, and the sign of its value has a directly interpretable meaning: a positive value indicates an activating effect, a negative value an inhibiting effect, and zero no effect.
As a typical example, the differential equation for caffeoyl CoA, defined as X1, has two representations (cf. Figure S4B and [33] for further details):(1)The variables Vi in the first equation represent the reaction rates, or fluxes, in a generic format. In the corresponding GMA model, these fluxes are specifically modeled as products of power-law functions of dependent and independent variables Xi. Aside from X1, the fluxes contain two other dependent variables, X2 (caffeyl aldehyde) and X3 (feruloyl CoA), as well as two independent variables, Xn+1 (CCR2) and Xn+2 (CCoAOMT). In GMA representations, n typically denotes the number of dependent variables, so that n+1 and n+2 refer to the first two independent variables. At the steady state, the derivative on the left-hand side becomes zero, thereby turning the differential equation into a linear equation of fluxes (ViS; where S denotes the steady state)(2)with the following constraints(3)
Notably, X3 and X2 are included in the power-low representations of V2 and V3, respectively, because they potentially modulate the two consuming fluxes of X1. Applying the rules for kinetic orders described above, we can immediately impose bounds on the values of f2,3 and f3,2 for different regulatory mechanisms (Figure 3B). For instance, modeling Mechanism 1 requires the following constraints,(4)because X3 is considered an inhibitor, so that f2,3<0, while X2 has no influence on the degradation of X1 through reaction 3 in this design, so that f3,2 = 0, which in effect eliminates the factor from the term on the far right. By convention, all independent variables have a kinetic order of 1.
Determination of all parameters in a GMA model, including kinetic orders and rate constants, is required prior to most simulation tasks. While numerous methods have been developed over the years, parameter estimation is seldom straightforward as each pathway and each dataset has its own challenges [35]. For the lignin pathway in Medicago, very little information is available on exact concentrations of intermediates or fluxes through the pathway; in fact, many metabolites in vivo are below detection level with standard HPLC [36].
To address this issue of insufficient data, we sample parameter values from relatively wide, biologically realistic ranges. The procedure involves the following steps. First, we sample the steady-state fluxes ViS from a set P in an m-dimensional vector space (Rm), where m equals the number of reactions. The sampled steady-state fluxes can be thought of as possible representatives of a wild-type Medicago species, while P defines the boundaries within which the system is able to operate. Biologically, these boundaries are given by many linear equality or inequality constraints with physiological meaning, such as the reaction stoichiometry (e.g. Eq. (2)), the ratio of S to G lignin in a mature stem internode of wild-type alfalfa, and the degree of reversibility of individual reactions (see Text S1 for further information). Mathematically, P is a bounded polyhedron (or polytope) and therefore has a concise parametric description(5)where the m-dimensional vectors ui can be identified using first principles [37]; in a different context, the vectors ui have been called “extreme pathways” [38].
Once a set of steady-state reaction rates is randomly generated, we sample kinetic orders (fi,j) from their respective ranges (Table S2), which are chosen based on the unique role of each kinetic order. Even with this information, the lack of concentration data from a wild-type Medicago species remains an issue that needs to be solved. To this end, we perform two transformations. First, we define a normalization of variables by replacing Xi with Yi≡Xi/XiS, where XiS are the unknown steady-state levels of Xi in wild type. As an example, the differential equation for caffeoyl CoA assumes the form(6)where ViS are the steady-state reaction rates sampled from P. This representation is well suited for the current analysis because the exact values of XiS become irrelevant once all the equations are set to zero, that is, at a wild-type or perturbed steady state (cf. Figure S4C). Second, after all parameters for a given GMA model instantiation are specified, we derive the corresponding S-system equations with straightforward mathematical manipulations that do not require any additional biological information [33; Chapter 3]. At the steady state, GMA and S-system models are equivalent, but they offer different advantages for further analyses. In particular, S-system differential equations, despite being intrinsically nonlinear, become linear at the steady state after a logarithmic transformation, thereby facilitating the computation of secondary steady-state features and bypassing the time-consuming numerical integration that is otherwise required for assessing nonlinear models. Given this convenient feature, we are able to obtain, in a very efficient manner, estimates of steady-state fluxes under conditions that mimic the two transgenic alfalfa lines and two M. truncatula mutant lines; we can also easily compute the S/G ratios for which we had experimental data.
Down-regulation of specific lignin biosynthetic enzymes is simulated by setting the corresponding normalized independent variables Yi to values between 0 and 1 that represent the degree of down-regulation, and solving the steady-state equations. In cases where hierarchical regulation might be effective, such as in ccr2 knockout mutants, all affected Yi are given values that mirror the specific changes in activities (cf. Figure S4D). The Parallel Computing Toolbox™ in MATLAB (version R2009b, The MathWorks, Natick, MA) was used to divide the simulation job among multiple cores for speedup.
Not all models behaved properly during simulation, and some ill-behaved models were excluded from further analysis. These were defined, arbitrarily, as models that showed a more than 1000-fold increase or decrease in any dependent variable during any simulation. Further, a properly behaved parameter set was deemed valid if the following criteria were met:
These criteria for success of a model instantiation were initially determined in an ad hoc fashion: We applied more lenient, qualitative criteria to the predictions of S/G ratios for ccr1 and ccr2 knockout mutants because the experimental data were only available in Medicago truncatula, but not in our model organism alfalfa (the S/G ratio of every model instantiation was set to the experimentally determined value for the sixth internode of a wild-type alfalfa plant). However, when we relaxed the criteria for screening the predictions of S/G ratios for CCoAOMT and COMT down-regulated lines to allow a percentage error as large as 25%, we obtained the exact same set of pathway designs that are consistent with the data, suggesting that the main conclusions are quite robust to the choice of thresholds.
The cloning of the alfalfa CCoAOMT cDNA into the expression vector pET15b was as described previously [15]. E. coli Rosetta strains containing the constructed plasmid were cultured at 37°C until OD600 reached 0.6–0.7, and protein expression was then induced by adding isopropyl 1-thio β-galactopyranoside (IPTG) at a final concentration of 0.5 mM, followed by 3 h incubation at the same temperature. Cell pellets from 25 ml induced medium were harvested and frozen at −80°C for further use. Induced cell pellets were thawed at room temperature, resuspended in 1.2 ml of extraction-washing buffer (10 mM imidazole, 50 mM Tris-HCl pH 8.0, 500 mM NaCl, 10% glycerol and 10 mM β-mercaptoethanol), and sonicated three times for 20 s. Supernatants were recovered after centrifugation (16,000×g), and incubated at 4°C for 30 min with equilibrated Ni-NTA beads (Qiagen, Germantown, MD) under constant inversion to allow the His-tag protein to bind to the beads. The beads were washed three times with 1 ml of extraction-washing buffer, and the target protein was eluted with 250 µl of elution solution (250 mM imidazole, 50 mM Tris-HCl buffer pH 8.0, 500 mM NaCl, 10% glycerol and 10 mM β-mercaptoethanol). The concentration of the eluted target protein was determined using the BioRad protein assay (BioRad, Hercules, CA) and its purity was verified by SDS-PAGE.
Caffeoyl CoA for the enzyme assays, and feruloyl CoA for the calibration curve, were synthesized as described previously [39]. Caffeyl aldehyde was synthesized as described by Chen et al. [40]. Pure recombinant CCoAOMT enzyme (100 ng) was incubated at 30°C for 20 min with 60 mM sodium phosphate buffer pH 7.5, 200 µM S-adenosyl methionine, 600 µM MgCl2 and 2 mM dithiothreitol. The substrate (caffeoyl CoA) concentration was 20, 30 or 40 µM and the putative activator (caffeyl aldehyde) concentration was 0, 2, 5 or 10 µM. Since caffeyl aldehyde was in dimethyl sulfoxide solution, the final concentration of dimethyl sulfoxide in the reaction was 4% and the final volume of the reaction was 50 µl. The reactions were stopped by adding 10 µl of 24% w/v trichloroacetic acid. Reaction products were analyzed by reverse-phase HPLC on a C18 column (Spherisorb 5 µ ODS2, Waters, Milford, MA) in a step gradient using 1% phosphoric acid in water as solvent A and acetonitrile as solvent B. Calibration curves were constructed with authentic standard of the product feruloyl CoA. Activity assays using lower concentrations of the substrate caffeoyl CoA (2, 4, 5 and 10 µM) and the putative activator caffeyl aldehyde (0.5, 1, 2 and 4 µM) were performed using a sensitive radioactive assay method as described previously [15].
|
10.1371/journal.pbio.1000351 | Environmental Change Enhances Cognitive Abilities in Fish | Flexible or innovative behavior is advantageous, especially when animals are exposed to frequent and unpredictable environmental perturbations. Improved cognitive abilities can help animals to respond quickly and adequately to environmental dynamics, and therefore changing environments may select for higher cognitive abilities. Increased cognitive abilities can be attained, for instance, if environmental change during ontogeny triggers plastic adaptive responses improving the learning capacity of exposed individuals. We tested the learning abilities of fishes in response to experimental variation of environmental quality during ontogeny. Individuals of the cichlid fish Simochromis pleurospilus that experienced a change in food ration early in life outperformed fish kept on constant rations in a learning task later in life—irrespective of the direction of the implemented change and the mean rations received. This difference in learning abilities between individuals remained constant between juvenile and adult stages of the same fish tested 1 y apart. Neither environmental enrichment nor training through repeated neural stimulation can explain our findings, as the sensory environment was kept constant and resource availability was changed only once. Instead, our results indicate a pathway by which a single change in resource availability early in life permanently enhances the learning abilities of animals. Early perturbations of environmental quality may signal the developing individual that it lives in a changing world, requiring increased cognitive abilities to construct adequate behavioral responses.
| Animals with higher cognitive abilities should be better capable of producing new, modified, or innovative behaviors as this ability could allow them to cope better with unpredictable environmental changes. Changing environments may hence select for higher cognitive abilities. Similarly, changing conditions during ontogeny can cause plastic responses, helping individuals to adapt to their current environment. In this study, we have used the cichlid fish Simochromis pleurospilus to show experimentally that individuals subjected to a change in food ration early in life (i.e., low to high or vice versa) outperform fish kept on constant rations in a learning task later in life. Remarkably, this result was independent of the direction of the implemented change or the average amount of food each fish received, and the results in the juvenile stage did not change in adulthood. Our results suggest that a single environmental change early in life might enhance cognitive abilities in animals.
| The ability of adapting to changes in the environment is an important driving force of evolution, as recognized already by Darwin in his famous quote: “It is not the strongest of the species that survives…it is the one that is the most adaptable to change” [1]. Animals may adapt by altering their behavior, physiology, or morphology. The construction of behavioral responses is thought to be the fastest and most flexible way of adapting to new situations. Animals often have to deal with new situations for which they must devise novel or flexible solutions [2]. Field observations and laboratory studies showed that the advantages of novel or altered behaviors increase with the complexity of the environment (reviewed in [3],[4]). This suggests that frequent and unpredictable environmental changes may select for increased cognitive abilities allowing animals to meet these challenges by constructing adequate behavioral responses. In mockingbirds, for example, the complexity of songs is assumed to reflect their cognitive abilities, and species inhabiting areas with a low predictability of climatic patterns show more elaborate song displays than species in stable environments [5].
On the level of the individual, environmental instability can be encountered by plastic trajectories of the development of cognitive abilities. Environmental fluctuations early in life are known to enhance the behavioral flexibility of animals with regard to predator avoidance strategies [6],[7], feeding performance [7], and social behavior [6],[8]. A possible explanation for these behavioral effects is that variable environments evoke repeated neural stimulations resulting in faster and better learning [7]. Several studies showed that neural stimulation over longer periods by exposing animals to enriched environments (e.g., [9],[10]) can enhance brain development [5],[11], for example through an increased synaptic density [12], and can lead to improved learning abilities and memory capacity [12]. A food manipulation experiment indicated that a single change of diet can constrain neural development if later cognitive abilities are traded against the benefits of a compensatory growth response [7],[13]. On the contrary, an environmental change early in life should be expected to favor enhanced cognitive development, if this early perturbation signals the developing individual that it lives in a more variable environment. In response to this signal animals should develop increased cognitive abilities, which help them to construct adequate behavioral responses to the environmental challenges. An experimental evaluation of this hypothesis has been hitherto lacking.
Individuals of the African cichlid Simochromis pleurospilus live in a stable environment, but parts of the population experience a habitat shift around maturation [14]. If increased cognitive performance confers a fitness advantage when shifting habitats, we should expect that improved cognitive abilities can be triggered in S. pleurospilus by experimentally varying their juvenile environmental quality. We tested this prediction by investigating how the performance in a learning task was influenced by different juvenile feeding regimes in S. pleurospilus. Fish were fed either on a stable high or a stable low food ration, or rations were switched from low to high or vice versa. We trained the fish to associate a visual cue with food and tested how often they selected the positive stimulus. We tested their cognitive performance twice, at the end of the juvenile period and 1 y later when the fish were adults. We adhere to the broad definition of “cognition” as comprising “all mechanisms that invertebrates and vertebrates have for taking in information through the senses, retaining it, and using it to adjust behavior to local conditions” [15].
The tests of learning abilities yielded similar results in juveniles (J) and adults (A). Neither the amount of food received before the switch (J: p = 0.62, A: p = 0.53) nor after the switch (J: p = 0.21, A: p = 0.12) influenced the number of correct choices significantly. The interaction between early and late food treatment was significant (J: p = 0.029, A: p = 0.005, Table 1) however, demonstrating that fish that had experienced a switch in feeding regime outperformed those fed constant rations. This effect is independent of the direction of the diet change (high-to-low or low-to-high; Figure 1).
Alternatively, learning ability might be affected by the average amount of food during the juvenile period. In that case we should expect the learning performance to increase linearly with food ration or, if an optimal food level exists, to follow a dome-shaped relationship. To test for this possibility, we determined the average amount of food each fish consumed relative to its own body mass. However, the number of correct choices was not related to mean relative food ration, neither with a linear (GLzM: J: p = 0.59, A: p = 0.86) nor with a quadratic predictor (GzLM: J: p = 0.26, A: p = 0.90).
During the test of juvenile cognitive abilities the animals of different treatment groups differed in body size (overall difference: ANOVA, df = 3, F = 103.88, p<0.001; differences between individual groups: Tukey's hsd test, all p<0.001, Figure 2A) and in the latency times to approach the stimulus (ANOVA, df = 3, F = 7.81, p<0.001; differences between individual groups: Tukey's hsd test, all p<0.01, Figure 2B). These differences had disappeared by the time the fish were tested for a second time during adulthood (size: ANOVA: df = 3, F = 0.61, p = 0.61, Figure 2C; latency time: ANOVA: df = 3, F = 0.38, p = 0.77, Figure 2D).
Individual S. pleurospilus that had experienced a change in food ration early in life outperformed those fish kept on constant rations in a learning task, suggesting that changes in environmental quality triggered a better cognitive performance in these fish. Remarkably, this result was independent of the direction of the implemented change and the mean rations received. The difference in learning abilities between treatment groups remained constant between juvenile and adult stages of fish tested 1 y apart, which suggests that a single change in food availability can trigger better cognitive abilities probably for lifetime. Several alternative possibilities of how ration, size, or growth rates can affect cognition can be ruled out as likely explanations for our results.
Juveniles differed in size across the treatment groups as they were subject to different feeding rations during the tests. NHH and SLH fish received near ad libitum food and were presumably satiated, whereas NLL and SHL fish experienced a food shortage most likely resulting in a stronger motivation of these two groups to approach the test apparatus. This is reflected in substantial differences in the times to leave the shelter and to approach the choice apparatus. A differential motivation cannot explain our results however, as in this case the learning performance should differ between NHH/SLH and NLL/SHL fish. Moreover, these differences in latency time had disappeared, when testing the fish the second time during adulthood. Potential motivational differences were now eliminated as the size differences had vanished and all fish received the same food rations.
Poor early nutrition can adversely affect neural development [16]–[18] and it can have a negative impact on song learning in birds [19],[20] and intelligence quotients in humans [21]–[23]. Also this factor cannot explain our findings as NHH fish did not perform better than NLL fish. We assume that the low-food ration was sufficiently high to sustain normal neural development, as in a previous study [14],[24] S. pleurospilus raised on the NLL ration developed and reproduced normally.
Body size is amongst the traits under strongest selection [25]. Juvenile fish should have high incentives to grow fast, since the number of potential gape-size limited predators decreases exponentially with increasing body size [26]. Fast growth can have negative effects, however (reviewed in [27]), including a negative influence on cognitive performance. In zebra finches, birds that had the highest rates of compensatory growth after experiencing a period of a reduced food ration performed worst in a subsequent learning task [13]. This effect might result from a trade-off between investment in growth versus neural development [28],[29] or from prolonged stress due to increased foraging activity leading to chronically elevated levels of corticosterone, which in turn can adversely affect neural development [30]. If compensatory growth had affected the learning performance of S. pleurospilus, NHH fish should have outperformed those groups that had not started on a high-food diet and that exhibited compensatory growth in our experiment (all fish reached similar sizes at the time when adults were tested, but NLL fish took longer to do so than SLH fish). If the brain was especially vulnerable to negative effects of compensatory growth during the juvenile period, then the fish that experienced a switch from a low to a high ration (SLH; highest degree of compensatory growth) should perform worst. The opposite was the case, however, as SLH fish outperformed NLL fish.
To ensure that the learning experience of juveniles did not influence the subsequent learning performance of adults [31], we performed a test series, which confirmed that the fish did not remember the conditioned cue of the juvenile test series before starting the second series. Other fish species have been shown to forget learned foraging techniques already within 2 d [32], whereas the tests of juveniles and adults in our experiment were more than 1 y apart. We are therefore confident that we tested independent learning abilities of the fish in both test series rather than memory effects. Twelve individuals, which replaced fish that had died until the onset of the second test series and which had not been tested as juveniles, slightly outperformed the previously tested fish in the learning tests (Table 1; see Material and Methods for details). Possibly fish tested 1 y before may still have associated the test apparatus with a food reward, as apparently they were less afraid of the test apparatus (shorter latency times until approach; see Material and Methods). They may therefore have been less attentive to the type of stimulus cue during the training phase than previously untested fish. These fish approached the apparatus more cautiously and hence may have had more time to associate the cue with food. The effect of previous learning experience was the same across all treatment groups.
Environmental conditions during development may trigger changes in morphology, physiology, or behavior, which can confer an adaptive advantage later in life if an animal remains in these conditions [33]. The main mechanisms proposed to explain such plastic responses to environmental cues involve repeated stimulation, for example, through physical exercise facilitating muscle development or by early neural stimulation through environmentally enriched raising conditions, which enhances cognitive abilities later in life [7],[12],[34]. But neither environmental enrichment nor training through repeated neural stimulation can explain our findings, as the sensory environment was kept constant during ontogeny and resource availability was changed only once. Our results rather show that already a single event—a change of food ration—during early ontogeny triggers learning ability possibly indicating the existence of a novel pathway of plastic neural development.
It has been hypothesized that changing environments improve learning abilities, which consequently may allow animals to behave more adequately and flexibly [7]. Our results support this hypothesis by showing that environmental change can indeed directly affect learning abilities, independently of motivational differences between individuals. Changing environments experienced early in ontogeny can greatly improve the flexibility of behavior [6]–[8]. If such effects result partially from better learning abilities induced by early environmental change, these studies elucidate the manifold possible consequences of improved learning abilities, which extend to a wide range of behavioral contexts.
The life history and ecology of S. pleurospilus suggests that the improvement of cognitive abilities in response to environmental change is adaptive. S. pleurospilus are algae grazers and hence depend upon the primary production of turf algae, which is influenced mainly by light intensity and a suitable substrate such as rocks and stones [35]. Algae productivity decreases exponentially with depth [35]. While juvenile S. pleurospilus inhabit the shallow regions of the lake with the highest algae intensity and some fish stay there throughout adulthood, other fish start to settle in deeper water around maturation [14]. These fish should benefit from increased cognitive abilities, as they have to cope with entirely different nutritional conditions. Improved cognition may help them to find and remember patches of high-quality turf algae (reviewed in [36]), while those fish remaining stationary in the natal habitat do not necessarily require a better cognitive performance. Hence our findings suggest that habitat shifts can make these animals smarter. More generally, animals forced to cope with environmental changes as caused, for example, by anthropogenic perturbations of their habitats may benefit from improved cognitive abilities induced by these perturbations when forced to adjust to the new conditions.
In conclusion we show for the first time that a single change in food availability early in life can enhance life-long learning abilities. Hence our study provides experimental support for the hypothesis that selection favors higher cognitive abilities in unpredictable or changing environments [5],[9]. It also suggests a mechanism of how animals can acquire better abilities to cope with such environments: an environmental switch early in ontogeny may enhance learning ability persistently.
Simochromis pleurospilus is a maternally mouthbrooding cichlid of the subfamily Tropheini endemic to Lake Tanganyika, East Africa. It lives along the rocky shores of the lake where it feeds on epilithic turf algae. S. pleurospilus reproduces all year round and adult males defend small, adjoining territories of 2–4 m2 where females visit them to spawn. Juveniles and females are non-territorial and use large home ranges. After spawning females leave the male territory immediately and care for the clutch on their own [24]. Approximately 28 d after spawning, the young are independent. Juveniles and adults live sympatrically, but juveniles are confined to the shallow areas between 0.5 and 2 m depth, whereas adults often disperse to feed in greater depth between 1 and 12 m ([14], A. Kotrschal & B. Taborsky submitted).
We raised 130 fishes in separate 20-l Plexiglas tanks, each equipped with a layer of sand, a flower-pot half for shelter, and an internal biological filter (see [24] for details on experimental set-up). The experimental fish were derived from seven clutches of different females, and siblings were proportionally distributed over all treatments. We exposed the fish to two different feeding conditions in early and late adolescence, respectively, using a full-factorial design. Fish either received (1) a high food ration both in early and late life (abbreviated as NHH, where “N” stands for “Not switched”; n = 40); (2) a low food ration both in early and late life (NLL, n = 40); (3) a high food ration in early life, switched to a low food ration in late life (SHL, where “S” stands for “Switched”; n = 22); or (4) a low food ration in early life, switched to a high food ration in late life (SLH, n = 22). Diet switches were performed either at 77 d (i.e., after the first third of the juvenile period; SLH: n77d = 11; SHL: n77d = 11) or at 133 d of age (i.e., after the second third of the juvenile period; SLH: n133d = 11; SHL; n133d = 11). We had switched diets at two different ontogenetic stages to enhance the chances to capture a potential sensitive period when a change in resource availability affects cognitive abilities. As in the learning trials fish switched at day 77 did not perform differently from fish switched at day 133, neither as juveniles (GzLM; SLH: 77 d versus 133 d: χ2 = 2.1, p = 0.15; SHL: 77 d versus 133 d: χ2 = 0.3, p = 0.57) nor as adults [SLH: 77 d versus 133 d: χ2 = 0.09, p = 0.77; SHL: 77 d versus 133 d: χ2 = 2.3, p = 0.13 (details of model see section Statistical Analysis)], we combined the data of early and late switched fish resulting in four treatment groups: NHH, NLL, SLH, and SHL. Fish were fed 6 d a week with standardized agarose cubes containing an amount of Tetramin flake food corresponding to 12% (near ad lib) or 4% of mean body weight plus 5% Spirulina algae. All fish of a treatment group received the same food ration, which was based on the mean body mass of fish within this group. We adjusted the food rations to increasing mean body weights every 14 d. We stopped adjusting the rations to body weight in NHH fish at 189 d, because they no longer depleted the food cubes. We continued to adjust the ration for the NLL, SLH, and SHL fish until day 259 when they reached the same body size as NHH fish. Thereafter all fish were kept on the same food ration. We measured lengths and weights of fish every 3 wk. Standard lengths were read from a measuring board with a 1 mm grid and were estimated to the nearest 0.5 mm by eye. Weights were read to the nearest 1.0 mg from an electronic balance. All measurements were taken before the daily feeding and done by the same person (A.K.).
We first trained the animals to associate a certain visual cue with food and thereafter determined the number of correct decisions made when presenting the cue. We did the first test series in the juvenile phase shortly before maturation (J) at an age of 172 d (±10 d) when the fish still received different food rations and differed in body size between treatments. The second test series was done 1 y later in adults (A) at an age of 585 d (±10 d), when all fish were fed the same rations and were of similar size. Each fish was tested in its individual raising tank.
Twelve juveniles were excluded, as they were used in a pilot study after which we adjusted the testing protocol. Furthermore 12 fish never left their shelter within 12 min, yielding a final sample of 66 fish for the juvenile test series (NHH = 19, NLL = 19, SHL = 14, SLH = 14). One year later some fish that had been tested previously had died in the meantime. Therefore we added 12 previously untested individuals to increase sample sizes. We used all SHL, SLH, and NLL fish still alive and 30 NHH fish for the second test run. Five adults refused to take food from the test apparatus and eight adults never left their shelter within 12 min, resulting in a sample size of 77 fish for the adult test series (NHH = 30, NLL = 25, SHL = 8, SLH = 14). Overall 46 fish participated in all 6 juvenile trials, and 69 fish participated in all 10 adult trials. The mean rate of participation was 5.3 (±1.4 SE) times out of 6 in juveniles and 9.7 (±1.0 SE) times out of 10 in adults. Although juvenile NLL and SHL fish participated more often than NHH and SLH fish (ANOVA: F = 4.63, p = 0.005), this did not bias the results because the statistical model accounts for participation rate (see Statistical Analysis). Adult fish of all treatment groups participated at a similar level (ANOVA: F = 1.04, p = 0.38).
Since not all fish participated in every trial we used binary probit-link generalized linear models (GzLM) to analyze the cognitive performance with the total number of correct choices as the dependent variable and the number of times the fish participated as the independent variable [37]. We included food ration in early adolescence (“early food treatment”) and in late adolescence (“late food treatment”) as fixed factors. Twelve adults that had not been tested as juveniles took on average 70 s longer to enter the choice area (Mann-Whitney U: Z = −2.19, p = 0.028), but they outperformed those fish already tested as juveniles (GzLM, χ2 = 6.11, p = 0.013). As the latter effect occurred across treatment groups (indicated by an absence of a significant interaction between treatment group and previous test experience, GzLM: χ2 = 352, p = 0.84), we included previous test experience (yes or no) in the model of adult learning performance.
To examine whether the amount of food per se influenced the likelihood of correct choices, we tested if a positive, a negative (linear predictor), or a dome-shaped (quadratic predictor) relationship exists between these two variables. We determined the amount of food received by each individual by calculating the percentage of food mass contained in the food pellets relative to the body mass of individuals using data from our tri-weekly body mass measurements. We then took the mean of these values during the entire juvenile period (i.e., until week 30) as a measure of food consumed by individual fish. All analyses were done with SPSS 17.0, SPSS Inc., Chicago.
|
10.1371/journal.pcbi.1004606 | A Model of Drosophila Larva Chemotaxis | Detailed observations of larval Drosophila chemotaxis have characterised the relationship between the odour gradient and the runs, head casts and turns made by the animal. We use a computational model to test whether hypothesised sensorimotor control mechanisms are sufficient to account for larval behaviour. The model combines three mechanisms based on simple transformations of the recent history of odour intensity at the head location. The first is an increased probability of terminating runs in response to gradually decreasing concentration, the second an increased probability of terminating head casts in response to rapidly increasing concentration, and the third a biasing of run directions up concentration gradients through modulation of small head casts. We show that this model can be tuned to produce behavioural statistics comparable to those reported for the larva, and that this tuning results in similar chemotaxis performance to the larva. We demonstrate that each mechanism can enable odour approach but the combination of mechanisms is most effective, and investigate how these low-level control mechanisms relate to behavioural measures such as the preference indices used to investigate larval learning behaviour in group assays.
| The larvae of the fruitfly are attracted to many odours. We use a computational model in which simulated larvae stop, start and redirect their crawling behaviour in response to their experience of changes in odour. We show that three simple rules for switching between behaviours are sufficient to produce larva-like results in a simulated agent.
| It is well established that Drosophila larvae perform chemotaxis towards a wide range of odourants (e.g. [1]). Our aim in this paper is to examine what sensorimotor mechanism(s) account for larval chemotaxis, looking for a minimal model that captures observed phenomena. This will allow us to examine the nature of sensory input and its processing, and identify possible key control outputs that are modulated by conditions or experience. In particular, we are interested in connecting models of odour discrimination and learning to the odour experience of the animal as it moves in a gradient.
Many other organisms also exhibit chemotaxis, using a variety of different strategies [2]. The basic forms of orientation mechanism are reviewed in [3]. Bacteria alternate straight swimming and random tumbling, with the probability of switching modulated by the direction of change in chemical intensity [4]. In C. elegans, a similar modulation of the frequency of large re-orientations (pirouettes) by the odour gradient is accompanied by a more gradual directed bias of runs towards the odour [5]. Insects such as the silkworm moth that navigate in patchy odour plumes make upwind surges in response to odour encounters, interspersed with zig-zag and casting behaviours [6]. Flies approaching odour sources in relatively still air might do so by alteration of their visuomotor control, to increase straight flight and suppress turning if odour concentration is increasing [7, 8]. Note that none of these strategies requires the use of spatially separated olfactory sensors to obtain instantaneous measurement of the direction of an odour gradient, but rather exploit temporal change due to movement of the animal, movement of the chemosensors, movement of the medium carrying the odour, or a combination of all three. However, in many cases a bilateral arrangement of sensors does make instantaneous assessment of the relative concentration across space possible, and this is sometimes exploited: for example, bees [9] and flies [10, 11] exhibit turning towards the antenna experiencing higher concentration.
Drosophila larvae’s olfactory sensors are located at the tip of the head [12, 13]. As larvae have left and right olfactory sensory organs it would seem possible that they could compare between left and right odour concentrations to perform odour taxis. It has been reported that crude unilateral surgical ablation of sensory organs leads to increased turning towards the intact side [14]. However the separation between these sensors is very small and it seems unlikely that the minute instantaneous difference in concentration between left and right could be detected over environmental, sensory and neural noise [15]. Furthermore, using genetic rescue of single olfactory neurons, it has been demonstrated that while bilateral sensory input improves chemotaxis, it is not required [16].
The most salient features of the larva’s movement patterns also seem inconsistent with instantaneous lateral steering. Larval locomotion has two distinct modes [17]. During runs, consistent peristaltic waves cause the larva to move forwards in a relatively straight line (but see below). During turns, unilateral contraction of one side of the body or the other causes the anterior section of the body to sweep from side to side, a behaviour referred to as ‘head casting’. The effective direction of a turn is determined by the casting behaviour ending with the anterior section of the body at an angle to the rest of the body. In this case, when the larva resumes running, it moves off in a new direction with respect to the previous run. As it moves forward, the rear gradually realigns itself with the front. Larvae have been shown to produce run and turn behaviours without the brain [18], suggesting they may have a ‘basic’ locomotion pattern embedded in the ventral nerve cord and motor system, which can be modulated by higher brain areas in response to sensory input.
The most detailed behavioural description of larval Drosophila chemotaxis comes from [19]. By using an arena designed to produce a well-defined odour gradient (described in [16]), and fine-grained tracking of individual larvae exploring this environment, the authors were able to decompose larval behaviours based on orientation with respect to the local odour gradient. This analysis revealed that larvae were 1) more likely to stop runs and start head casting when moving down gradient, and 2) more likely to turn (i.e. finish head casting and return to running) towards the direction of higher odour concentration. Similar results have been reported in a study using linear odour gradients [20].
But how do larvae determine when to turn and which direction to turn? Turn initiation is typically preceded by a period of decreasing sensory experience (defined as a normalised derivative of concentration) corresponding to running down the gradient [19]. Turns to the direction of higher concentration are typically preceded by a large spike in sensory perception, corresponding to a head cast in the direction of higher concentration. Given that the direction of a turn (the alteration in direction between two runs) is determined by the direction of the head cast preceding the turn, a large spike in sensory perception could act as a signal to transition from head casting back to forward movement, resulting in turns generally being towards the direction of high concentration [19].
More recently a third factor contributing to odour-directed paths in larvae has been described [21, 22] which has been termed ‘weathervaning’. During runs, the larva’s path tends to be curved slightly but significantly towards the side of higher odour concentration. This behaviour can still be observed for larvae with single, unilateral olfactory receptors, and has been hypothesised to utilise active sensing of the lateral olfactory gradient through low amplitude head casts during runs [21].
In this paper we use an agent based model to determine if these three control mechanisms—initiating head casting when the odour intensity is decreasing, ending head casting when a sharp increase in odour is experienced, and ‘weathervaning’—can be derived from simple perceptual processing; whether they can replicate fine-grained statistics of larval behaviour; whether they are necessary and/or sufficient to produce chemotaxis; and whether they can be used to provide a low-level account for high-level behavioural descriptions such as preference indices.
We abstract the body of a larva as consisting of two sections of equal length, the head and body, with one articulation between them. The basic larva has two distinct behaviours, runs and head casts (Fig 1a). During a run, the head section moves forward at constant speed vforward. Head orientation remains constant during a run, apart from slight modulation by the weathervane mechanism (see below). The body section is ‘pulled’ behind the head section during runs; when the head body angle is not zero the body section gradually rotates to align with the head section as the larva moves forward. During head casting, the body section remains motionless while the head section rotates from side to side relative to the body section, at speed θ c a s t ′. Upon reaching the limit of rotation, θmax_head_cast, in one direction, head rotation in the opposite direction begins immediately. Head casting may terminate with the head section oriented differently to the body section; this orientation will determine the direction of the following run, and thus the effective size and direction of turns. All our simulations consist of single larva trials, and we therefore do not consider collisions or interactions between larvae.
The intensity of the odour at any location in the simulation is given by a single value; in this paper we consider only single-odour environments (as used in many behavioural experiments). We use both artificial odour gradients (e.g. a Gaussian distribution of concentration around an odour source), and data taken from measurements of real experimental odour landscapes in which larvae have been tested.
For a simulated larva in a given landscape we use the odour concentration C at the tip of the head section as the input to the larva’s ‘perception’. This perceptual value is generally the only information the simulated larva has about the environment, and all behavioural modulation is based on a limited history of this value.
Following [19], we approximate perception with a rule of the form:
ϕ = 1 C · d C d t (1)
This rule gives the larva access to a measure of the relative rate of change of the odour concentration. When moving up gradient the perceptual value will be positive, and when moving down gradient the value will be negative. Note that this perceptual processing is deliberately simple, intending to capture the hypothesis that a relative rate-of-change perceptual signal is sufficient to allow for larva-like chemotaxis. In reality there must be some level of odour which falls below the perceptual limits of the animal, and some level that entirely saturates the response, but these effects are not included in the current model. We address the issue of more realistic sensory processing in the discussion.
Due to the normalisation in our perception rule, the scale of concentration values in our odour landscapes is arbitrary. Thus for simplicity, unless otherwise noted we normalise the values of all odour environments such that the peak concentration is 1.
Our starting point for behavioural control of the simulated larva is based directly on the hypothesis proposed by [19], that directed behaviour emerges out of sensory-driven probabilistic transitions between running and head casting that are controlled solely by the recent history of perception (Fig 1b). Specifically, it is assumed that larvae:
For our purposes, it is simpler to first define rates of transitions, and convert these into probabilities of transitioning between behaviours on each time step by:
p r u n _ t e r m i n a t e ( t ) = r r u n _ t e r m i n a t e ( t ) · d t (2) p c a s t _ t e r m i n a t e ( t ) = r c a s t _ t e r m i n a t e ( t ) · d t (3)
Note that simulations proceed in discrete timesteps of length dt = 0.1s.
We can now define our control problem as converting the larva’s perceptual history (the only information it has available to it) into these transition rates. We do this by defining a kernel for each transition, and obtaining a rate of transitions by convolving the perceptual history with the appropriate kernel (i.e. element-wise multiplying and then summing). As larvae make transitions from runs to turns even in the absence of odour stimuli, we further include a base rate of making a transition regardless of the perceptual history.
r r u n _ t e r m i n a t e ( t ) = r r u n _ t e r m i n a t e_b a s e + ∑ t ′ = 0 t r u n _ k e r n e l ϕ ( t - t ′ ) · k r u n _ t e r m i n a t e ( - t ′ ) (4) r c a s t _ t e r m i n a t e ( t ) = r c a s t _ t e r m i n a t e _ b a s e + ∑ t ′ = 0 t c a s t _ k e r n e l ϕ ( t - t ′ ) · k c a s t _ t e r m i n a t e ( - t ′ ) (5)
To encode the desired behavioural controls into our model, we use simple linear kernels which resemble the average perceptual history preceding behavioural transitions in real larvae, as reported in [19]. Thus the kernel for run termination takes the form of a gradual negative slope, while the kernel for cast termination is a steeper positive slope with a similar duration to a single head cast (see depictions in Fig 1c). Note that both transition rates are continuously calculated regardless of behavioural state, but only affect behaviour when the larva is in the corresponding state. The output of the perception rule ϕ and the control rules rrun_terminate(t) and rcast_terminate(t) for a section of simulated larva paths are shown in Fig 1c.
Initial simulations using the control outlined above raised a number of issues leading to the following modifications to the control scheme.
Twenty parameters need to be set for this model. Some can be taken from available data, but the appropriate values for others were less clear. We discuss here how we chose each of our parameters, with the final values used to generate the results in this paper summarised in table 1.
We take the forward movement speed vforward = 1mm/s from figure 2a in [19]. From analysis of paths of larvae in a no-odour environment we estimate the base rate of transitions from runs to turns at rrun_termination_base = 0.148s−1, that is, turns occur on average every 7 seconds. To determine the corresponding base rate for turn to run transitions, we count the proportion of turns which have a single associated head cast. We assume that this gives the probability of transitioning from head casting to forward crawling behaviour during any given head cast (making the implicit assumption that the distribution of number of head casts before a turn can be described by a geometric series). We then divide this probability by the duration of a head cast in our model to give rcast_termination_base = 2.0s−1, that is, a probability of 0.7 of returning to running after a single head cast.
Inspecting head casts from the no-odour larva data, we found that over 95% of casts did not exceed 120°, and so we set θmax_head_cast = 120°. The speed of head casts needs to be fast enough to allow up to 4 head casts within a 5 second window (as seen in the larval data), so has been set to allow for a head cast (out and in) of maximal size in one second: θ c a s t ′ = 2 * θ m a x _ h e a d _ c a s t. To make turns effective, head casts should only terminate beyond some minimum angle from the centreline. We use the definition of head casts used to define turns in [19], i.e. θmin_head_cast = 37°. The amplitude of weathervane casts was set to θmax_weathervane_cast = 20°, matching the size of small head casts shown in figure 8c in [21]. Weathervane cast speed was set to a moderate value, θ w e a t h e r v a n e _ c a s t ′ = 60 °.
The final parameters to be set are those defining the lengths and shapes of the kernels. As we only use linear kernels, they can be described with three parameters, the duration and the start and end values. We need a relatively long, negatively sloping kernel for the run termination kernel, and a short, positively sloping kernel for the cast termination kernel. On this basis we found approximate values for the kernels by adjusting until the simulated larva displayed navigation towards the odour source.
We then further adjusted kernel parameters by hand until our model matched larval behaviour across a range of behavioural statistics (see Results). This was achieved through gradual adjustment of parameters and visual inspection of resulting behavioural statistics. Automated optimisation of these parameters would have been possible in theory, however, defining a single optimisation criteria when the goal was to match across several distributions would in itself be a subjective process.
The kernel parameters chosen are reported in Table 1; these values are used throughout unless otherwise noted.
To establish a comparison between our model and experimental results from real larvae, we apply a number of metrics from [19] to paths of wild type larvae and simulated larvae. These metrics are constructed from the trajectories of larvae’s head, centroid and tail positions; note that in this analysis no use is made of the internal state of the model. The metrics used are as follows:
We also extract times of turns and head casts from our simulations. These are defined as follows:
Events are categorised as follows:
Our aim when picking kernel parameters was to produce a simulated larva which matches the behavioural statistics of real larvae (n = 42) chemotaxing in an odour gradient of ethyl butyrate, as reported in [19].
We produce behavioural statistics for the simulated larva as follows using the same odour distribution as measured in [19], in a virtual arena of size 65x100mm. Note that the arena size for the larval experiments was 80x120mm, however an estimate of the odour concentration could not be experimentally made at the outer edges.
We run 500 simulated larvae in this arena, for 300s of simulated time each. Each simulated larva begins the run with random starting orientation, at a random position within a 12mm square centred on the odour source. The run of any simulated larva which touches the edge of the arena is truncated at that point, consistent with the acquisition of experimental data.
From these simulated trajectories, we calculate various behavioural metrics (as described in the previous section). These are used to produce the behavioural statistics shown in Fig 2.
Note that this process was repeated multiple times as we tuned kernel parameters; we show here only the behavioural statistics obtained with our final set of parameters as reported in Table 1.
Fig 2 shows two sample paths, and the match between real and simulated behavioural statistics. In the top panel of statistics it can be seen that the probability of initiating a turn (an end-run transition) relative to the odour bearing shows the same form as for the real larva. In the second panel, the probability of making a left turn is altered as expected relative to the bearing of the odour, turning left more often if the odour is on the left. In the third panel, the reorientation rate during runs shows a similar dependence on the bearing angle, and similar amplitude.
These comparisons demonstrate that it is possible to choose kernel parameters which result in our model producing very similar behavioural statistics to the larva, on the three metrics which summarise the run termination, cast termination and weathervaning mechanisms. Given that we tuned kernel parameters to match these statistics it is perhaps not surprising that we achieve a close match; nonetheless, it is not trivially obvious that these statistics would be possible to obtain using only linear kernels convolved with the relative rate of change in odour concentration.
The lower panels present comparisons of additional behavioural statistics, which show a number of emergent effects also match well between the model and the larva without additional tuning. The run termination mechanism produces distributions of turn initiation bearings and run lengths similar to the larva. Similarly, the head cast termination mechanism produces turn direction probabilities and numbers of pre-turn head casts comparable to the larva. Both simulated and real larvae tend to be headed away from the odour (>90 degrees) before a turn, and towards it (<90 degrees) after, but with a general undershoot, that is, only a partial correction in orientation. They also both show a similar distribution of bearings that result in correct (to higher concentration) rather than incorrect (to lower concentration) turns, with wrong turns more likely when heading near to 180 degrees away from the odour, a situation which produces the most ambiguous information during head casting.
In the bottom panel, the relative frequency of patterns of head casts towards the direction of higher (H) or lower (L) concentration is shown. Overall the proportions are similar between the larva and the simulations. The larva and the model both show a bias in the direction of the first head cast, with a majority of head cast groups starting with a cast to high. The mechanism by which the larva creates this bias is not yet understood, however, our model produces a similar bias by simply using the angle of its head at the moment of run termination to determine its initial cast direction.
Having set parameters for our model such that it matches the larva on these low level behavioural statistics, we go on to assess the model’s similarity to the larva by comparing its chemotaxis performance to the larva in three different environments.
Having tuned our model parameters to qualitatively match low level behavioural statistics of the larva when chemotaxing around a point source of odour, we ask whether this leads to our model quantitatively matching the larva on a higher-level metric, namely the distribution of larvae around the the odour source.
Using the data described in the previous section, we computed the distance to the odour source for 42 real and simulated larvae every second for 150s. For comparison, we repeated this process for 19 real and simulated larvae in a ‘no odour’ condition; for the simulated larva this means all behavioural transitions are made at their base rates, with no perceptual modulation. Fig 3 shows sample paths, the temporal evolution of the distance to the source over time, and a snapshot of distances to the source at 120s, for each of these groups.
In the absence of an odour source real larvae gradually disperse; the model performs similarly to the larva in this condition.
With the odour source present, real and simulated larvae both remain located around the source. We use the performance of simulated larvae at 120s as a quantitative measure to determine how closely our model, with parameters tuned to match larvae’s low level behavioural statistics, matches the larva’s chemotaxis performance. The larva and the model’s distances to the source are not significantly different (Mann-Whitney U Test, p>0.05), although this does not provide evidence for the null hypothesis that the medians of the groups are the same. However, using bootstrapping, we can state with a 95% confidence level that the difference in median distance to the source for real and simulated larvae is between -2.4 and 1.9mm, around a single larval body length.
Having confirmed that our simulated larvae show chemotaxis performance at a similar level to real larvae when circling around the odour source, we consider how well they match the larva’s directness of approach to a distant odour source.
For this experiment, we used a second odour gradient of ethyl butyrate (also from [19]), with an odour source centred at one end of a rectangular arena.
Simulations were carried out as above, with a different odour landscape and different starting positions; each simulated larva begins the run at a random position within a 20mm square in line with the odour source on the short axis of the arena and 68mm distant on the long axis. Starting orientations were chosen randomly from a distribution of plus or minus 30 degrees relative to the direction of the odour source. Runs were truncated at the point at which they came within 5mm of the odour peak; runs which did not reach this area were discarded.
For this condition we compare 43 real and simulated larvae. We calculated distances to the odour source every second as above. Following [23], we also use a path tortuosity metric to compare the efficiency of orientation in this landscape; a ‘straightness index’ is assigned to each individual by calculating the ratio of the length of its path to the length of the vector travelled. Paths leading directly to the odour peak will have a straightness index of 1, while paths which follow a less direct route will have a lower straightness index.
Fig 4 shows sample paths, the temporal evolution of the distance to the source over time, and boxplots of straightness indices for real and model larvae. We see a good match between the larva and the model’s approach to the odour peak. The larva and the model’s straightness indices are not significantly different (Mann-Whitney U Test, p>0.05), and using bootstrapping, we can state with 95% confidence level that the difference in median straightness index for real and simulated larvae is between -0.11 and 0.03.
We next consider the behaviour of the simulated larva in linear vs. exponential odour slopes. We compared the model’s paths to paths of wild type larvae in gradients of isoamyl acetate (landscapes from [16]). Runs were truncated at the point at which they came within a 15mm zone at the peak end of the arena; runs which did not reach this area were discarded. Larvae started within a 20mm square in line with the odour source on the short axis of the arena and 80mm distant on the long axis, facing towards the odour peak. We used the same number of simulated larvae as real larvae in each condition: 20 for exponential, 14 for steep linear, and 11 for shallow linear.
Initial results suggested that with parameters set as described above (to match behaviour in a single source environment of ethyl butyrate), simulated larvae performed significantly worse than the real larvae in this condition. We therefore also tested whether we could improve the performance of the model by scaling (i.e. changing the slope of) the model’s kernels, thereby increasing the strength of the simulated larva’s behavioural biases. We show here results for the model with default parameters, and with a scaling factor of 6 on all kernels.
Fig 5 shows sample paths, the temporal evolution of the distance to the source over time, and boxplots of straightness indices for real and model larvae in each environment. These demonstrate that the model (both with normal and scaled kernels) can successfully navigate up both exponential and linear gradients. The ‘straightness index’ shows that the larva more directly approaches the odour peak in an exponential gradient than in a shallow linear gradient; this is also true for the model with either normal or scaled kernels.
However, both the sample paths and the straightness indices highlight the failure of the model with default parameters; simulated larvae in this case follow paths which are clearly more tortuous than the larva. However, by scaling the kernels by a factor of 6, we achieve a much closer match between the model and the larva.
Why do we need to alter our model’s kernel parameters to replicate larval behaviour in this situation? One possibility comes from the difference in the odour used in this condition; our model’s parameters were set to match the behaviour of larvae in an ethyl butyrate gradient, while these gradients were produced using isoamyl acetate. It is possible that the difference in odourants leads to an difference in chemotactic performances for the larva. Alternatively, the differences between model and larval performance in these gradients may be a result of our simplified sensory processing (see discussion). In any case, the fact that the model does chemotax successfully in this environment even with default parameters demonstrates its robustness in a novel odour landscape.
A common measure of larval behaviour in odour experiments is the ‘preference index’ [12, 24]. In typical odour-based experiments, a number of larvae are allowed to freely explore a Petri dish (typically 9cm diameter). One or both sides of the dish contain odour sources. At the end of an allotted time period, the number of larvae on each side of the dish (excluding a 1cm centre region) are counted. A preference index is then calculated as:
P I = # s i d e 1 - # s i d e 2 # t o t a l (6)
Preference indices therefore range from 1 to -1, with a positive preference index indicating a preference for side 1, and a negative preference index indicating a preference for side 2. Typical median preference indices for innate chemotaxis range from 0 to close to 1, depending on the odour and concentration used (Schleyer and Reid, pers. comm.).
Unfortunately, the odour environments used in learning experiments are not as carefully controlled and measured as the data we have used for comparisons so far. Odours are presented in the form of a point source, e.g. in a small cup or on a filter paper. Furthermore, the enclosed nature of the Petri dish is likely to lead to non-uniform distribution of the odour, which may also be changing over time. As there are no detailed recordings of odour gradients in these conditions from which we can draw, we assume a very simple odour distribution; a circular Gaussian (σ = 30mm) distribution centred on the odour source.
For these trials the simulation arena consists of a circular wall with a radius of 45mm, with an odour source 5mm from the left hand side of the dish. For each odour condition, we ran simulations for 400 individual larvae, each of which was allowed to explore the arena for 5 minutes. Each larva began at a random position on the vertical centre-line of the dish, at a random orientation. At the end of 5 minutes, the position of each larva was recorded. The larvae were split into 20 groups of 20, and for each of these groups a preference index was calculated.
Initial results, using the parameter settings described above, showed extreme PIs; all simulated larvae ended the 5 minute run on the odour side of the arena, i.e. PI = 1. We therefore proceeded to investigate how scaling (changing the slope of) the model’s kernels, thereby reducing strength of behavioural biases and the efficiency of chemotaxis, changes the observed PIs. Fig 6 shows the distribution of simulated larvae and the corresponding PIs for different kernel scaling factors. Also shown is the the effect of the kernel scaling on the statistics for run termination bearing, turn direction probability, and run reorientation in the original point source environment (as for Fig 2). A scaling of 0.1 produces a strong PI score, and a scaling of 0.05 a score still comparable to larval experiments. With scaling 0, the simulated larva has no behavioural biases and indeed ends up equally distributed across the dish, with PI around 0. Scaling by a negative value produces a negative PIs, that is, apparent repulsion from the odour.
We might expect that our model’s original parameter settings, which we have demonstrated produce behavioural biases of the same magnitude as the larva’s, should produce larva-like PIs. However, we had to considerably reduce kernel scaling, and therefore reduce behavioural biases, to produce moderate PIs. There are several possible explanations for this requirement. First, note that our model has access to a clear odour gradient across the whole arena, unperturbed by noise or sensory thresholds; real larvae may not encounter a consistent gradient far from the source. Alternatively, moderate preference indices could coexist with high behavioural biases if only a fraction of the larvae were displaying those biases, or if all larvae were displaying those biases only a fraction of the time. In preference tests of relatively long duration, this does not seem unlikely. In either case, averaging some combination of strongly biased and unbiased behaviour would result in seeing lower overall behavioural biases for larvae in PI-type experiments, as our model suggests. Finally, there may be effects of group assays such as random reorientations caused by collisions. Unfortunately, data to resolve larval behaviour at an individual level during group assays is not available, restricting our ability to differentiate between these explanations.
Our model combines three sensorimotor mechanisms that appear to operate in the larva to produce chemotaxis:
Finally, we explored the performance of the model in a selection of distinctly different odour landscapes (linear, Gaussian, and step), with different amounts of noise, and with different combinations of the three behavioural biases. This provides a useful parallel to the analysis in [25] for a model of C. elegans. We analyse each larva’s performance with a Chemotaxis Index (CI); each larva is assigned a CI equal to the proportion of time spent in a region of interest (ROI); note that due to the differences in ROI areas, direct comparisons of performance between conditions cannot be made. The gradients used were all contained within a 9cm diameter circular arena, with parameters as follows. Linear: Concentration varying linearly from 0 at leftmost edge to 1 at rightmost edge. Region of interest is the rightmost 30mm of dish. Gaussian: Gaussian odour distribution, centred at the centre of the dish, with peak concentration 1 and variance 16mm. Region of interest is a 25mm diameter circle around the centre of the dish. Step: Odour concentration of 0 on left side of the dish and 1 on right, with 5mm wide linear transition of concentration between the halves. Region of interest is the right half of the dish.
For each combination of mechanisms (as in Fig 7) we ran 500 simulated larvae in each of the three environments, starting at a random initial position and orientation. This was repeated with multiplicative noise added to the environment by dividing the arena into an 0.08mm square grid and multiplying the concentration in each square by a value picked from a normal distribution with mean 1 and variance 0.04 (low noise) or 0.1 (high noise). CIs for each condition are shown in Fig 8.
The results suggest that some mechanisms may make smaller or greater contributions depending on the conditions. For example, including all three mechanisms seems to make a difference for the Gaussian gradient but not the other conditions, where weathervaning makes little contribution. CIs in the linear gradient are more affected by noise than the other conditions. The run termination mechanism seems more effective than the cast termination in the step gradient, and in general the contribution of cast termination is more affected by noise.
Our aim in this paper was to implement a minimally complex model that captures the observed odour taxis behaviour of Drosophila larvae. We show that larva-like behaviour can be achieved using relative change of odour concentration at the tip of the head combined with simple linear kernels to trigger transitions between behavioural states. Tuning the parameters of this model to match the detailed observations of larvae reported in [19, 21] produces behaviour that also matches the larva on other measures (such as proportions of head casts, under-correction of the heading angle, behaviour on different odour gradients). However, to reproduce typical preference indices reported for en masse assays we needed to alter the parameters to substantial weaken the effects of odour concentration on state transitions. This is discussed further below.
Our model combined three mechanisms, acting to modulate the probability of state transitions between running and head casting. In the absence of odour, the simulated larvae exhibit exploratory behaviour, making runs with regular small ‘weathervane’ head casts that are paused on random occasions leading to shallow curves, and also randomly stopping the run and making larger head casts, with random transition back to running, which can produce sharp turns. If this behaviour occurs in an odour gradient, the probability of stopping a run is enhanced for decreases and suppressed for increases in the change in odour concentration. The probability of restarting a run is enhanced by sharp increases in odour concentration during head casting. Small head casts during runs are also paused more often when the experienced odour concentration increase is greater than that already occurring in the run, resulting in a ‘weathervane’ curve towards the odour. From our simulations, it appears that either of the first two mechanisms would be sufficient to get the larva to an odour in a smooth gradient. Combining both with weathervaning produces the best performance; the improvement is most apparent in a Gaussian distribution, which is perhaps closest to the expected gradient for a point source of odour. Run termination is more robust to noise, which might be expected as (at least in our implementation) it averages input over a longer time scale than the other mechanisms.
The first mechanism (alteration in the rate of transitioning from running to head casting depending on the change in concentration) is equivalent to bacterial klinokinesis or the modulation of pirouette frequency observed in C. elegans [26] and is well-known to be sufficient to ascend a gradient. The second mechanism (ending head casting and resuming running when the head cast produces a sharp change in concentration) is klinotaxis, and we have shown it is also sufficient for chemotaxis on its own, producing similar performance to pure klinokinesis (when the latter is tuned to match larval behaviour). The combination of these two mechanisms substantially improves the efficiency of odour localisation over either alone. However, pure klinokinesis can potentially produce better chemotaxis if the parameters are tuned to optimise its performance. The final mechanism, weathervaning, is seen to make marginal improvements to chemotaxis performance, although it does not lead to robust chemotaxis without additional biases in the other mechanisms. It has been hypothesised that biasing of larvae’s run curvature is facilitated by low amplitude head casts made while running [21]. Our model concretely implements this hypothesis by including continuous low-amplitude head casting, which is temporarily paused by increases in the perceptual signal. We demonstrate that this mechanism can produce biases in run curvature comparable to that of the real larva, but other mechanisms, such alteration in the size of these casts, are also possible.
[19] show that larvae’s first head casts after terminating runs tend to be in the direction of higher odour concentration. We suggest that this is possible due to the larva having information about the lateral gradient from weathervaning during its run, and show that a similar level of first headcast bias can be produced if the simulated larva simply casts in the direction of its current head angle when terminating a run. This interaction of weathervaning and cast direction accords with the observation that the direction of run curvatures and subsequent turns are correlated [21].
Although treated here as a distinct mechanisms, klinotaxis and weathervaning could be interpreted as the same underlying orientation algorithm, i.e., exploiting the lateral sweep of the head through the gradient to obtain information about the odour direction, and altering the timing or extent of the sweep to orient the animal up the gradient. In the case of C. elegans, their oscillatory forward locomotion naturally produces a substantial sweep (relative to body length) and can be altered to produce relatively tight curves. In larvae, the peristaltic propulsion during runs appears to be inconsistent with large head casts, so while biasing the production of small head casts can steer the animal up the gradient, direct approach is only possible by stopping to make larger casts (or indeed, it may be that making a large cast forces a stop). It remains to be discovered how independent are the neural mechanisms underlying these behaviours in the larva.
The majority of behavioural experiments on larvae report only preference indices (PI), i.e., a binary classification of larvae as either within or without a designated region defined in relation to the stimulus of interest. It is important to understand how such global measures relate to the underlying behavioural control if the neural circuits involved in innate and learned sensorimotor control are to be explained. An issue revealed by our analysis is the difficulty of interpreting behavioural statistics that are derived by summing over many individuals, and over relatively long time durations. It was necessary to make each of the behavioural biases around 20 times weaker in a simulated larva to obtain ‘typical’ PIs. The discrepancy between the level of behavioural bias required to match PI data and the low-level behavioural statistics reported in [19, 21] could have multiple sources: different larvae may have different innate capabilities or preferences for particular odour sources; the attraction of an individual larva to an odour source may change over time due to habituation or changing motivational state or competition from other cues; or the odour gradient itself may vary substantially in the reliability with which it corresponds to the actual odour direction, both over time and space, in a typical Petri-dish experiment. It is clear that this issue can only be resolved by studies that track individuals over time in well-controlled or measurable stimulus conditions.
It is important to note that both previous biological experiments [16] and our simulations indicate that the larva can locate odours with a single point sensor on its head, and does not need spatially separated sensors, even for weathervaning. Rather, gradient information is gained through stereotypical movements over time. Nevertheless it is clear that the perceptual response to the odour gradient used in the simulation, which performs perfect differentiation and normalisation, is not realistic. In the majority of the behaviour analysed here the larva spends a large proportion of its time close to the odour peak, and thus in a relatively limited range of concentrations. As such, normalisation should not have a large impact on the model’s global behavioural statistics. However, it is likely to play a significant part in the model’s ability to localise the peak from a distance, by making the simulated larvae unrealistically sensitive to small differences in areas of low odour concentration. In future work, it will be interesting to incorporate more detailed olfactory receptor responses (such as described in [27]) into the model, and see how these interact with both different odour gradients and the dynamics of the motor actions to shape the overall behaviour, particularly in relation to approaching an odour from a distance.
We have also used a highly simplified model of the larva’s motor system. Although ‘runs’ and ‘head casts’ are reasonable approximations to the main observable actions by the larva, further analysis may reveal important subtleties. For example, the peristaltic pattern that produces the run also imposes a pattern on the sensory input, as the head moves forward and pauses on each cycle, and also changes its orientation with respect to the substrate. Similarly, the rather arbitrary distinction between ‘small’ and ‘large’ head casts used in the simulation may need more detailed representation of the form, size and location of body bends of which the larva is capable.
Finally it may be interesting to ask whether a simpler control scheme than the state transitions illustrated in Fig 1b might give rise to qualitatively similar behaviour. It is interesting to note that the mechanisms used to produce the different behavioural transitions in our model are all fundamentally the same, involving differentiation, integration and a non-linear switch, and differ only in their timescales and their weighting of the perceptual signal. Should we assume the current characterisation will map onto distinct ‘decision’ circuits in the animal for changing between runs and head casts? Or is it possible that these are emergent properties of lower level control that integrate the muscle contractions producing both peristalsis and body bends and modulates them in response to sensory input?
|
10.1371/journal.pgen.1007227 | Dek overexpression in murine epithelia increases overt esophageal squamous cell carcinoma incidence | Esophageal cancer occurs as either squamous cell carcinoma (ESCC) or adenocarcinoma. ESCCs comprise almost 90% of cases worldwide, and recur with a less than 15% five-year survival rate despite available treatments. The identification of new ESCC drivers and therapeutic targets is critical for improving outcomes. Here we report that expression of the human DEK oncogene is strongly upregulated in esophageal SCC based on data in the cancer genome atlas (TCGA). DEK is a chromatin-associated protein with important roles in several nuclear processes including gene transcription, epigenetics, and DNA repair. Our previous data have utilized a murine knockout model to demonstrate that Dek expression is required for oral and esophageal SCC growth. Also, DEK overexpression in human keratinocytes, the cell of origin for SCC, was sufficient to cause hyperplasia in 3D organotypic raft cultures that mimic human skin, thus linking high DEK expression in keratinocytes to oncogenic phenotypes. However, the role of DEK over-expression in ESCC development remains unknown in human cells or genetic mouse models. To define the consequences of Dek overexpression in vivo, we generated and validated a tetracycline responsive Dek transgenic mouse model referred to as Bi-L-Dek. Dek overexpression was induced in the basal keratinocytes of stratified squamous epithelium by crossing Bi-L-Dek mice to keratin 5 tetracycline transactivator (K5-tTA) mice. Conditional transgene expression was validated in the resulting Bi-L-Dek_K5-tTA mice and was suppressed with doxycycline treatment in the tetracycline-off system. The mice were subjected to an established HNSCC and esophageal carcinogenesis protocol using the chemical carcinogen 4-nitroquinoline 1-oxide (4NQO). Dek overexpression stimulated gross esophageal tumor development, when compared to doxycycline treated control mice. Furthermore, high Dek expression caused a trend toward esophageal hyperplasia in 4NQO treated mice. Taken together, these data demonstrate that Dek overexpression in the cell of origin for SCC is sufficient to promote esophageal SCC development in vivo.
| The DEK oncogene is overexpressed in nearly all human cancers and portends a poor prognosis for many cancer types. High DEK expression causes cancer related phenotypes such as increased cellular proliferation, migration, and invasion in vitro. Despite the well documented link between high DEK expression and cancer, the consequences of Dek overexpression in vivo are poorly understood. To determine if Dek contributes to carcinogenesis in vivo, we generated a Dek inducible transgenic mouse model wherein Dek can be overexpressed in specific tissues and inhibited with the drug doxycycline. We targeted Dek overexpression to keratinocytes, the cell of origin for squamous cell carcinoma, and exposed the mice to the chemical carcinogen 4NQO to induce oral cavity and esophageal carcinogenesis. We found that DEK overexpression was sufficient to increase gross esophageal squamous cell carcinoma incidence and caused a trend toward increased cellular proliferation in adjacent non-tumor tissue. Importantly, we demonstrate for the first time that Dek overexpression specifically targeted to basal keratinocytes is sufficient to promote cellular and squamous cell carcinoma growth in vivo.
| The human DEK oncoprotein is a predominantly chromatin-bound factor that regulates nuclear processes such as chromatin architecture, epigenetics, transcription and DNA repair [1–18]. DEK was originally identified as a fusion protein with the CAN/NUP214 nucleoporin in a patient with acute myeloid leukemia harboring the chromosomal translocation (t6;9)(p23;q34) [19]. Since its discovery, DEK was also shown to be increased in acute myeloid leukemia types that do not harbor the DEK-NUP214 fusion protein [20–22] and to be frequently overexpressed in solid tumors including colon, breast, gastric adenocarcinoma, ovarian carcinomas, bladder cancer, retinoblastoma, lung, pancreatic, neuroendocrine prostate cancer, hepatocellular, skin cancer, head and neck cancer squamous cell carcinoma (HNSCC), and esophageal squamous cell carcinoma (ESCC; S5 Fig) [23–40]. Additionally, high DEK expression is associated with poor prognosis in melanoma, gastric, ovarian, breast, prostate, bladder, lung, pancreatic, skin cancer, and head and neck SCC [25, 26, 30, 31, 33–35, 40–43].
Esophageal carcinomas are the sixth most common cause of cancer related death worldwide, and eighth in incidence worldwide [44–46]. Esophageal carcinoma occurs as either SCC or adenocarcinoma [47]. Esophageal SCC accounts for one third of esophageal cancer cases in the United States but represents more than 90% cases of esophageal cancer worldwide [47, 48]. The most common risk factors for ESCCs, similar to HNSCC, include tobacco smoke, heavy alcohol consumption, and infection with human papillomavirus [49, 50]. Several studies have additionally revealed that ESCC and HNSCC harbor similar genetic and molecular alterations [44, 51–54] and are treated with similar regimen of surgery and chemoradiation [50]. However, the 5-year survival rate for patients with HNSCC is over 50%, while for patients with ESCC it remains at a dismal 5–15% [45, 46, 48]. Current treatment regimens frequently result in irreparable tissue damage and disfiguration that additionally highlight the need for continued identification of oncogenic drivers and targeted therapies [55]. SCC arises from keratinocytes in squamous epithelium, and the overexpression of DEK has been shown to promote cell survival, proliferation, and transformation in combination with classical oncogenes while inhibiting apoptosis, cellular differentiation and senescence [16, 56–59]. DEK overexpression occurs through various mechanisms including gene amplification, increased transcription, and mutations in microRNAs and ubiquitin ligases responsible for DEK mRNA and protein degradation, respectively [60–70]. Several in vivo studies demonstrate the critical role of human and murine Dek in driving benign and malignant tumor growth. For example, Dek knockout (Dek-/-) mice are partially resistant to the formation of benign skin papillomas when treated with DMBA and TPA, a tumor initiator and promoter, respectively [16]. In a breast cancer mouse model, Dek-/- mice bred to Ron receptor tyrosine kinase transgenic mice, displayed a delayed onset of mammary tumors compared to Dek+/+ mice [71]. In another study, Dek knockout (Dek-/-) HPV E7 oncogene transgenic mice were protected from 4-nitroquinoline 1-oxide (4NQO)-induced HNSCC and ESCC tumor growth, but not initiation, when compared to their Dek+/+ counterparts [39]. Taken together, these studies support the possible importance of Dek overexpression as a key driver of uncontrolled cellular growth and tumor development.
Historically, most of the data that links DEK overexpression to oncogenic phenotypes were obtained from knockdown and knockout model systems. Only recently has DEK overexpression been investigated in vivo. In a 2017 report, Nakashima et. al. generated tetracycline inducible, whole body, Dek over-expressing mice [72]. The mice were treated with 4NQO in the drinking water for 28 weeks to induce oral lesions, then induced to overexpress Dek for 4 weeks before sacrifice. 4NQO is a chemical carcinogen that mimics the effects of tobacco smoke by forming DNA adducts and mutations similar to those seen in human HNSCC and ESCC [73, 74]. When administered in drinking water, 4NQO stimulates susceptibility to squamous cell carcinomas in the tongue, oral cavity, and esophagus [73, 75]. In the study, the mice over-expressing Dek for 4 weeks, post 4NQO treatment, harbored significantly increased hyperplasia in the tongue with a trend toward increased tongue tumor incidence. Interestingly, Dek overexpression significantly decreased tongue tumor diameter [72]. This suggests that a short term induction of Dek overexpression after long term carcinogen exposure has pro- and anti-tumorigenic effects. Importantly, this study demonstrated that Dek overexpression promotes cellular proliferation in tissues exposed to carcinogens. However, whether these effects are due to high Dek expression in keratinocytes as the cell of origin, and/or other cell types, remains unknown. Therefore, we targeted long term induction of the Dek transgene to the stratified squamous epithelium, and monitored resulting tumor phenotypes.
To this end, a tetracycline responsive Dek and luciferase transgenic Bi-L-Dek mouse model was newly generated. Bi-L-Dek transgenic mice harbor a tetracycline response element (TRE) that controls the bi-directional expression of Dek and firefly luciferase. The TRE allows for temporal and tissue specific control of Dek overexpression, thus making it a versatile mouse model wherein the Dek transgene expression is controlled by tetracycline or its more stable derivative, doxycycline (dox). Bi-L-Dek mice were crossed to keratin 5 tetracycline transactivator (K5-tTA) transgenic mice to target Dek and luciferase expression to basal keratinocytes that serve as progenitor cells for stratified squamous epithelium and are the cell of origin for squamous cell carcinoma (SCC) of the tongue and esophagus. The tTA protein produces a tet-off system where expression of the Dek transgene is repressed by dox. Dek overexpression and transgene repression by dox was verified in the skin, tongue and esophagus. Once validated, Bi-L-Dek_K5-tTA mice were subjected to 4NQO treatment in the presence or absence of dox. Dek caused a trend toward increased proliferation in tongue and esophageal epithelium after 4NQO treatment. Furthermore, Dek overexpression was sufficient to increase the incidence of gross esophageal, but not oral, SCC tumor formation in this system. This data suggests that Dek contributes to ESCC tumorigenesis at least partially through keratinocyte intrinsic pathways which promote cellular and tumor growth.
In order to overexpress Dek conditionally in a tissue specific manner, we utilized a construct Bi-L-Dek wherein Dek and luciferase gene expression were driven by a tetracycline response element (TRE). To generate the Bi-L-Dek transgene, Dek cDNA was cloned into the Bi-L-Tet plasmid expression vector (Clontech, Mountain View, CA, USA) as published by others [76], described in the Materials and Methods, and illustrated in S1A Fig. Following validation of TRE-dependent Dek and luciferase expression in vector transfected cells (S1B–S1E Fig), the transgene was excised and injected into the pronucleus of fertilized mouse eggs for generation of Bi-L-Dek transgenic founders (S1F Fig). Bi-L-Dek mice harbor a TRE that controls two mini cytomegalovirus (CMV) promoters driving bi-directional transcription of Dek and luciferase (Fig 1A). Four Bi-L-Dek transgenic founder lines were assessed for transgene stability over four generations before screening for doxycycline responsive expression of Dek (Fig 1B; the data for founder line #317 used in subsequent experiments is shown).
To determine which lines expressed Dek and luciferase under control of the TRE, Bi-L-Dek mice were crossed to K5-tTA mice (Fig 1C and 1D). The keratin 5 promoter targets tTA protein expression to the basal layer of stratified squamous epithelium including that of the esophagus, tongue, and skin (Fig 1E). In this system, administration of doxycycline (dox) represses tTA binding to the TRE to inhibit Bi-L-Dek transgene expression (Fig 1E). Founder #317 was chosen for subsequent experiments, harbors approximately three copies of the transgene (Fig 1B), and is referred to as Bi-L-Dek from here on.
Bi-L-Dek transgene expression in Bi-L-Dek_K5-tTA mice was validated with multiple methodologies. These included an in vivo imaging system (IVIS), and the detection of Dek mRNA and protein expression by quantitative, real time polymerase chain reaction (RT-qPCR), western blot analysis, in situ immunohistochemistry (IHC) and immunofluorescence (IF) (Fig 2). IVIS and ex vivo imaging confirmed luciferase expression in the skin of Bi-L-Dek_K5-tTA bi-transgenic mice and in the esophagus (Fig 2A and 2B). Dek mRNA levels were induced 3.5 fold over endogenous levels in the skin of Bi-L-Dek_K5-tTA mice, and repression to endogenous levels was achieved by feeding with dox chow for seven days (Fig 2C). Dek protein expression in the skin also increased approximately 3 fold over the levels of endogenous Dek in the Bi-L-Dek_K5-tTA mice (Fig 2D). Dek overexpression in the tongue was detected by IHC along with the expected decrease in the corresponding mice on dox chow (Fig 2E). Finally, we isolated keratinocytes from Bi-L-Dek_K5-tTA skin for cell culture and performed IF with antibodies against Dek and K5, then stained with DAPI to detect DNA. As expected, Dek expression was higher in the Bi-L-Dek_K5-tTA derived keratinocytes compared to those treated with dox, or compared to keratinocytes from single transgenic control mice (Fig 2F and 2G). Exogenous Dek localized to the nucleus as expected, and co-localized with endogenous Dek and DAPI (Fig 2F). Altogether, Bi-L-Dek_K5-tTA mice overexpressed Dek in the squamous epithelium of the skin, tongue, and esophagus, and Dek expression was repressed by doxycycline.
To assess the extent of Dek overexpression in the Bi-L-Dek_K5-tTA mice, we quantified transgene expression in the context of Dek knockout mice. Bi-L-Dek and K5-tTA transgenic mice were interbred with Dek-/- mice to generate Dek-/-_ Bi-L-Dek_K5-tTA offspring (Fig 3A). IVIS confirmed luciferase expression (Fig 3B), and RT-qPCR and western blot analysis confirmed Dek mRNA and protein expression, respectively, in the epidermis (Fig 3C and S2 Fig). Dek mRNA levels were induced by approximately four fold in the Bi-L-Dek_K5-tTA compared to control mice (S2 Fig) and Dek protein levels were induced by approximately 2–3 fold in the Bi-L-Dek_K5-tTA over control mice. These levels are similar to the levels of DEK expression that can be routinely achieved in normal epithelial cells transduced with retroviral or lentiviral DEK expression vectors and are within the range of DEK levels observed in cancer cells [16, 42, 56, 77–81].
To explore the broader utility of this model system, conditional Bi-L-Dek mice were bred to Dlx5/6-tTA mice to target Dek expression to neurons originating from the ventral forebrain (S3 Fig). As expected, robust Dek protein overexpression was detectable in cortical interneurons and striatal projection neurons from the ventral forebrain, as previously demonstrated for other Dlx5/6-driven transgenes [82]. Additionally, we crossed the Bi-L-Dek mice with Rosa-tTA mice to produce global Dek overexpressing mice. IVIS demonstrated luciferase expression from the Bi-L-Dek transgene throughout the body, which was repressed with dox chow (S4 Fig). No overt phenotypes were observed in the mice similar to results in previously published Tet-O-Dek_ Rosa26-M2rtTA mice. In all, these results further validate Bi-L-Dek-mediated transgene expression in murine epithelia, and demonstrate broad utility of this genetic mouse model for studies of Dek overexpression in other organ systems.
Based on data in the cancer genome atlas (TCGA), DEK is more highly expressed in ESCC compared to normal tissue and in ESCC compared to esophageal adenoma (S5 Fig). However, the contribution of Dek overexpression to ESCC development is unknown. To determine if DEK contributes to ESCC development or progression, we utilized the Bi-L-Dek_K5-tTA mice with Dek overexpression targeted to basal keratinocytes that form the epithelium. Bi-L-Dek_K5-tTA mice overexpressed Dek in stratified squamous epithelium of the tongue and esophagus, and Bi-L-Dek_K5-tTA mice on dox expressed only endogenous levels of Dek. Exposure of mice to drinking water containing the soluble quinoline derivative 4NQO promotes the development of oral and/or esophageal cancer. Therefore, we exposed two groups of mice +/- Dox to 4NQO in order to determine whether Dek overexpression in basal keratinocytes is sufficient to promote SCC and if early onset of Dek overexpression increases oral and/or esophageal tumor incidence or tumor burden. The experimental design is illustrated in Fig 4A. At six weeks of age, Bi-L-Dek_K5-tTA mice in the absence of dox (n = 7) or in the presence of dox (n = 5) were exposed to 10ug/mL of 4NQO in their drinking water to promote SCC susceptibility. After 16 weeks, mice were given normal water, sacrificed at 45 weeks of age or when moribund, and analyzed after experimentally induced death or sacrifice. One hour prior to sacrifice, mice were injected with the thymidine analog Bromodeoxyuridine (BrdU) to quantify proliferation. In the absence of 4NQO treatment, Dek overexpression did not significantly increase cellular proliferation in the tongue or the esophagus when compared to control mice on dox (Fig 4B). However, following 4NQO treatment there was a trend toward increased proliferation in the epithelia of Dek overexpressing tongue (p = 0.07) and esophagus (p = 0.15) (Fig 4C and 4D). These results are in line with recently published data wherein global Dek overexpressing mice did not exhibit hyperplasia or other phenotypes under normal conditions, but displayed tongue hyperplasia after 4NQO treatment [72]. Bi-L-Dek_K5-tTA mice exposed to 4NQO or not were then analyzed for the presence of tumors in the tongue and esophagus. Detailed results are shown for each mouse in Fig 5A. After 4NQO treatment, Bi-L-Dek_K5-tTA mice continued to express higher levels of Dek protein in esophageal epithelium compared to the control group on dox (Fig 5B). In addition, Dek overexpressing mice had a significantly higher incidence of gross esophageal tumors (Fig 5C). Specifically, all of the Bi-L-Dek_K5-tTA mice developed at least one visible esophageal tumor (100%), in contrast to only one of five Bi-L-Dek_K5-tTA mice on dox (20%). Furthermore, one Dek overexpressing mouse harbored an excessively large tumor, while two others harbored two separate grossly apparent tumors (Fig 5A, 5C and 5D). From published studies, the 4NQO protocol utilized was expected to result in a 10% incidence of gross tumors in Dek wild type mice [75, 83–85]. This compares roughly to the 20% cancer incidence in mice exposed to dox. Overall, the number of invasive tumors and mice with multifocal tumors was not significantly different between the two groups (Fig 5C); however, survival of Dek overexpressing mice was less than 60% while all dox-treated mice survived until sacrifice at week 45 (p = 0.11; Fig 5E). Taken together, these data provide evidence that Dek overexpression promotes esophageal squamous cell carcinoma growth.
Esophagi and tongues from all mice were microscopically examined to define tumor phenotypes and quantify microscopic lesions. Histological analysis confirmed that gross tumors in Dek overexpressing mice were squamous cell carcinomas with stromal invasion confirmed histologically in 67% of the mice (Fig 5A, 5C and 5H). Additional multifocal microscopic squamous cell lesions were detected in 50% of Dek overexpressing mice with all mice developing 1–3 squamous cell lesions including at least one grossly apparent tumor. In contrast, microscopic lesions predominated in dox treated mice, with one mouse harboring no lesions and the other mice harboring 1–2 squamous cell lesions including a single grossly apparent tumor (Fig 5A, 5G and 5I). The single tumor apparent at necropsy in this group was a well differentiated squamous cell carcinoma with abundant keratin production which differed from the moderate to poorly differentiated squamous cell carcinomas that predominated in Dek overexpressing mice (Fig 5I and 5H). Microscopic tumors in dox treated mice consisted primarily of papillary squamous cell lesions with a single focus of very superficial invasion in one lesion. This differed from the more extensive invasion and necrosis in tumors that arose in Dek overexpressing mice (Fig 5I and 5H). Dek levels appeared to be high in all tumors regardless of whether these originated in the Dek overexpressing group or the dox control group, thus suggesting strong selection for the upregulation of endogenous Dek during tumorigenesis (Fig 5F and 5G, bottom row). With regards to the one tumor that arose in the dox control group, endogenous upregulation or leaky transgenic expression of Dek could be responsible. No tongue tumors were identified in either group of mice. Taken together, we demonstrate for the first time that Dek overexpression promotes the growth of esophageal SCC in vivo.
A number of studies have linked DEK overexpression in various malignancies to cellular growth, motility/invasion and chemoresistance [16, 36, 37, 43, 58, 71, 77]. Relevant mechanisms have not been fully elucidated in each case. However, DEK loss has been shown to attenuate proliferation and survival, while inducing senescence or apoptosis, depending upon the cell type and model system studied. Required signaling pathways included those controlled by p53 and ΔNp63 to inhibit apoptosis and promote proliferation, respectively [17, 39, 58], Wnt/beta-catenin to drive invasion and cellular proliferation [32], VEGF to foster angiogenesis [7], Rho/ROCK/MLC to support migration [86], and NFkB to regulate cellular survival and growth [6, 8, 59]. One caveat regarding cancer-related interpretation of these results is that many of the experiments are based upon DEK loss of function, and thus only address the requirement for DEK in tumor cell growth and not the contribution of DEK overexpression to tumor growth. For instance, Dek knockout mice are viable and resistant to chemically induced papillomas and HPV E7 driven HNSCC. In vitro, DEK overexpression in primary keratinocytes extends life span, stimulates transforming activities of classical oncogenes, and de-regulates cellular metabolism [16, 17, 78]. These data are in line with, but do not prove, oncogenic activities that promote cancer development at the organismal level.
Here we demonstrate that Dek overexpression targeted to the epithelium stimulates proliferation specifically in the presence of 4NQO in the tongue and also in the esophagus. Furthermore, concurrent Dek overexpression and 4NQO exposure increased the incidence of gross esophageal tumors demonstrating for the first time that Dek overexpression contributes to ESCC tumor growth in vivo. The observed increase in hyperplasia in the tongue is similar to that seen with sequential 4NQO exposure followed by ubiquitous Dek overexpression reported by Nakashima et. al.(72). Interestingly, and in contrast to our data, Nakashima et. al. reported that Dek overexpression decreased the volume of resulting tongue tumors [72]. Key differences in the experimental designs between the two models likely account for the observed differences in tumor location and size in. Specifically, in the current study: 1) Dek overexpression was targeted to the basal epithelium as opposed to ubiquitous Dek overexpression including immune and stromal cells that modulate cancer cell growth, 2) 4NQO and Dek overexpression were concurrently administered rather than sequential exposure to 4NQO followed by Dek overexpression, 3) 4NQO exposure duration and dosage was 16 weeks at 10 μg/ml compared to 28 weeks at 20 μg/ml, 4) Dek overexpression duration was 52 weeks compared to four weeks, 5) exogenous Dek was unmodified and localized to the nucleusas compared to FLAG-tagged exogenous Dek protein localized predominantly to the cytoplasm, and 6) FVB/N mice were used compared to C57BL/6 mice. Interestingly, C57BL/6 and FVB/N harbor variations in immune phenotype, raising the intriguing possibility that immune surveillance and/or evasion account at least in part for the differing tumor phenotypes in mice with ubiquitous versus epithelial cell targeted Dek overexpression. Preliminary studies in the esophageal tumors in the current model did not reveal a significant CD3 positive T cell infiltrate by immunohistochemistry. Additional studies are needed to definitely determine the role of inflammatory cells in Dek dependent tumorigenesis, however, the lack of a prominent T-cell infiltrate suggests that differences in tumor growth in the two models cannot be simply explained by tumor infiltrating T cells acquiring an exhausted T-cell phenotype. The availability of these distinct complementary mouse models now provide a valuable system to identify cell specific functions that drive Dek induced carcinogenesis.
Distinct effects of global versus tissue-specific Dek expression might reflect interesting cell-type specific functions of Dek in the tumor microenvironment including immune cells, or systemic effects on epidermal proliferation and tumor growth. The complexity of DEK functions in vivo is exemplified in studies of non-vertebrate organisms For instance, in Arabidopsis, DEK3 overexpression decreases germination efficiency under high salinity conditions, and conversely, plants deficient in DEK3 germinated significantly better compared to wild-type plants suggesting DEK3 levels are crucial for stress tolerance [87]. The overexpression of human DEK in the Drosophila eye caused a rough-eye phenotype due to caspase-9 and 3-mediated apoptosis suggesting that DEK overexpression caused (rather than diminished) apoptosis [88]. These non-vertebrate eukaryote model systems highlight the need for balanced DEK expression and its versatile functions in vivo.
In the Bi-L-Dek_K5-tTA mouse model, Dek overexpression at the message and protein level was approximately 2–4 fold over that of endogenous Dek. This relatively modest level is in agreement with other published studies suggesting DEK expression levels are tightly regulated [1, 16, 58, 77–79]. Achieving strong overexpression of DEK in vitro in our hands has been notoriously difficult, potentially due to toxicity and cell death, e. g. in the above Drosophila study [88]. Importantly, a modest level of DEK overexpression in epithelial cells has been linked to oncogenic phenotypes in vitro. These DEK dependent oncogenic activities include enhanced cancer stem cell growth, colony formation, cellular invasion, mitotic abnormalities, and metabolic de-regulation, providing evidence that subtle increases in DEK protein expression are sufficient to elicit significant cellular consequences [16, 77–79]. In human ESCC, HNSCC, breast, bladder, colorectal, hepatocellular, and non-small cell lung carcinoma, DEK protein levels were increased in tumor versus adjacent normal tissue, and the extent of overexpression was variable. Per cell DEK protein detection in various tumor types can range from intense to weak staining by IHC, and overexpression by western blot analysis can range from 2–30 fold [23, 25, 26, 31, 34, 39, 42, 77, 80, 81]. Overall, this patient data suggests that high levels of DEK can be tolerated by some human tumor cells, and that even modest DEK expression is associated with cancer growth and/or maintenance.
In conclusion, Bi-L-Dek_K5-tTA mice subjected to 4NQO harbor trends toward increased cellular proliferation in the tongue and esophagus (Fig 4B–4D) and a significantly increased incidence of gross esophageal tumors (Fig 5C). Tongue tumors were not detected in these same mice. Importantly, control Bi-L-Dek_K5-tTA mice on dox nonetheless developed microscopic ESCC tumors, thus suggesting that Dek overexpression does not stimulate tumor initiation, but promotes tumor growth in the esophagus. This is in alignment with previously published Dek loss of function data from HNSCC-prone K14E7 transgenic mice wherein keratinocyte proliferation and tumor growth, but not the presence of microtumors, were diminished in the absence of Dek [39]. While an abundance of data has suggested that DEK promotes tumor growth in the presence of oncogenic stimuli, the above experiments do not unequivocally rule out a role for Dek in tumor initiation. Overall larger tumors in the Dek overexpressing mice may be due to increased growth of tumors once initiated, or due to premature initiation and thus extended time for growth. In either case, Dek overexpression significantly increased the incidence of gross tumors and over 40% of Bi-L-Dek_K5-tTA mice died prior to the 45 week end point, while all mice in the dox treated control group survived. A plethora of Dek knockdown experiments have shown the importance of DEK expression for cancer cell growth and survival [10, 16, 39, 58, 77]. These data, in conjunction with evidence that transformed keratinocytes are more sensitive to DEK loss when compared to their normal or differentiated counterparts [16], make DEK an attractive therapeutic target. Furthermore, Dek knockout mice are healthy and fertile, suggesting potential feasibility and relative safety for the targeting of DEK in cancer. However, no DEK inhibitors exist commercially nor have been published. Thus, the inducible targeting of Dek in Bi-L-Dek mice harboring ESCC tumors should now be an attractive model to interrogate the requirement of continued Dek expression for cancer maintenance and progression. Taken together, we have generated and validated a new mouse model of esophageal transformation using an inducible Bi-L-Dek transgene which is now available for broader studies of Dek in health and disease of the intact organism.
Murine Dek (mDek) DNA sequences were excised from the previously published R780 retroviral vector, using the restriction enzymes Sal I and Not I [16, 71], and cloned into the pBi-L plasmid (Clontech, Mountainview, CA Catalog No. 631005; GenBank Accession No.: U89934.) cleaved with the same restriction enzymes. The resulting pBi-L-Dek construct harbors the bi-directional Pbi-1 promoter which is responsive to the tTA regulatory protein in this Tet-Off system. The Tet-responsive element (TRE) consists of seven copies of the 42-bp tet operator sequence (tetO), and is located between two minimal CMV promoters that lack the CMV enhancer. Gene expression is silent in the absence of the tTA bound to tetO sequences and is silenced with the addition of doxycycline. The pBi-L-Dek transgene sequences were liberated using the restriction enzymes AatII and AselI. A 5247bp (Bi-L-Dek) DNA sequence was purified and microinjected into the pronucleus of a fertilized egg and inserted into a pseudo-pregnant mouse to produce Bi-L-Dek founders. Transgene transmission was validated, and pups from the F1 generation were mated with K5-tTA mice. Resulting F2 Bi-L-Dek_K5-tTA mice were further characterized. Four Bi-L-Dek founders were generated. One founder line never produced offspring. Another founder died before producing a pup that harbored the transgene. Of the two remaining lines, both overexpressed Dek but founder #317 was a better breeder. The murine Dek sequence that was cloned into the pBi-L-Tet vector is:
5’-ATGTCGGCGGCGGCGGCCCCCGCTGCGGAGGGAGAGGACGCCCCCGTGCCGCCC TCATCCGAGAAGGAACCCGAGATGCCGGGTCCCAGGGAAGAGAGTGAGGAGGAGGAGGAGGATGACGAAGACGATGATGAAGAGGACGAGGAGGAAGAAAAAGAAAAGAGTCTTATCGTGGAAGGCAAGAGAGAGAAGAAGAAAGTAGAGAGACTGACGATGCAAGTGTCTTCCTTACAGAGAGAGCCATTTACAGTGACACAAGGGAAGGGTCAGAAACTTTGTGAAATTGAAAGGATACATTTCTTTCTGAGTAAGAAAAAACCAGATGAACTTAGAAATCTACACAAACTGCTTTACAACAGGCCGGGCACAGTGTCCTCGTTGAAGAAGAACGTGGGTCAGTTCAGTGGCTTTCCATTCGAAAAAGGCAGTACCCAGTATAAAAAGAAGGAAGAAATGTTGAAAAAGTTTCGAAATGCCATGTTAAAGAGCATCTGTGAGGTTCTTGATTTAGAGAGGTCAGGCGTGAACAGCGAACTCGTGAAGAGGATCTTGAACTTCTTAATGCATCCAAAGCCTTCTGGCAAACCATTACCAAAGTCCAAAAAATCTTCCAGCAAAGGTAGTAAAAAGGAACGGAACAGTTCTGGAACAACAAGGAAGTCAAAGCAAACTAAATGCCCTGAAATTCTGTCAGATGAGTCTAGTAGTGATGAAGATGAGAAGAAAAATAAGGAAGAGTCTTCGGAAGATGAAGAGAAAGAAAGTGAAGAGGAGCAACCACCAAAAAAGACATCTAAAAAAGAAAAAGCAAAACAGAAAGCTACTGCTAAAAGTAAAAAATCTGTGAAGAGTGCTAATGTTAAGAAGGCAGACAGCAGTACCACCAAGAAGAATCAAAAAAGTTCCAAAAAAGAGTCTGAATCCGAAGACAGTTCTGATGATGAACCCTTAATTAAAAAATTGAAAAAGCCACCTACAGATGAAGAGCTAAAGGAAACAGTGAAGAAATTACTGGCTGATGCTAACTTGGAAGAAGTCACAATGAAGCAGATTTGCAAAGAGGTATATGAAAATTATCCTGCTTATGATTTGACTGAGAGGAAAGATTTCATTAAAACAACTGTAAAAGAGCTAATTTCTTGA-3’
K5-tTA mice were obtained internally at CCHMC and have previously been published [90]. Dek knockout mice (Dek-/-) have previously been published [16]. Dlx5/6-tTA mice were obtained internally at CCHMC and were generated in Dr. Kenneth Campbell’s lab by Lisa Ehrman. Dlx5/6-tTA mice have been analyzed for tTA expression, and will be fully described and characterized in a separate publication. Dlx5/6 tTA expression is similar to Cre expression in the reported Dlx5/6-Cre-IRES-EGFP (CIE) transgenic mouse model [82]. E2A-Cre mice were obtained from Jackson Laboratory and are strain number 003724. The Cre transgene is under the control of the adenovirus EIIa promoter, which targets expression of Cre recombinase to the early mouse embryo. This model is useful for deletions, in the germ line, of loxP-flanked genes. E2A-Cre mice were bred to Rosa-LNL-tTA transgenic mice. These mice were also obtained from Jackson laboratories and are strain number 008600. Rosa-LNL-tTA mice contain a loxP-flanked nonsense sequence inhibiting expression of tTA that is removed once exposed to E2A controlled Cre.
Ear clips were digested with 25mM NaOH in 0.2mM EDTA at a pH of 12 and incubated at 95°C for 20 minutes. The reaction was neutralized with 40mM Tris-HCl. For PCR analysis, one ul of the digest with DNA was added to JumpStart Taq Ready Mix from Invitrogen (Carlsbad, CA, product # P2893) using the manufacturer’s specifications. Transgenes were detected with the following primers:
Bi-L-Dek:
Forward: GAAATGTCCGTTCGGTTGGCAGAAGC;
Reverse: CCAAAACCGTGATGGAATGGAACAACA.
K5-tTA:
Forward: GCTGCTTAATGAGGTCGG
Reverse: CTCTGCACCTTGGTGATC.
Bi-L-Dek primers that do not detect endogenous Dek (exogenous Dek cDNA primers)
Forward: CAGTGACACAAGGGAAGGGTCAGA
Reverse: AGCCACTGAACTGACCCACGT.
Genomic DNA was isolated from the tails of mice from successive generations of offspring from founder 317. A minimum of two mice were used per generation and analyzed as replicates. The DNA concentration was adjusted to 20ng/ul in each case, and 60ng of DNA was used for qPCR per sample and performed in duplicate. Primers were used to quantify the beta actin gene and a region in exon 6 of the Dek gene. This region is present in Dek+/+ mice but absent in Dek-/- mice thus allowing for a negative control. The following sequences were used:
Beta actin forward: GATATCGCTGCGCTGGTCGTC
Beta actin reverse: ACCATCACACCCTGGTGCCTAG
Dek Exon 6 forward: AGGTCAGGCGTGAACAGCGA
Dek Exon 6 reverse: TGCCAGAAGGCTTTGGATGCATTA
The critical threshold (CT) values for Dek exon 6 primers were normalized to actin, and quantified relative to Dek wild type mice using the delta delta CT method. Values were multiplied by two to account for the two endogenous Dek alleles in WT mice and the number of Bi-L-Dek transgene insertions was determined. Error bars represent multiple mice from the same generation.
Mice were injected with 15ng of luciferin per gram in body weight, and allowed to metabolize the luciferin for five minutes prior to sedation with isoflurane. Mice were imaged in the Perkin Elmer IVIS Spectrum CT, Waltham, Massaschusetts, USA. For ex vivo IVIS, mice were allowed to metabolize luciferin for eight minutes following luciferin injection, and then sacrificed with CO2. The mice were then dissected and tissues placed in PBS containing 300ug/mL of luciferin, kept on ice, and protected from light before immediate analysis by IVIS.
For validation of pBi-L-Dek expression, the plasmid was transfected into previously isolated and cultured K5-tTA expressing murine keratinocytes [91]. Cells were collected for Dek protein expression by western blot analysis. K5-tTA keratinocytes were grown in E-media supplemented with 0.05 mM Ca2 and 15% serum as previously published [92].
Keratinocytes were isolated from Bi-L-Dek_K5-tTA mice and single transgenic littermate controls using a previously published protocol with modifications [93]. Briefly, pups were euthanized within 48 hours of birth, rinsed in 70% ethanol, and placed in PBS. Flank skin was removed, and placed dermis side down in 1 mL of dispase (Dispase Gibco/Invitrogen, Calsbad, CA, USA, product# 17105–041) and 1 mL of DMEM (1:1 mixture) in a 35mm plate, and incubated overnight at 4° Celsius. The epidermis was removed and placed in 1 mL of accutase (Sigma, St. Louis, MO, USA, product # A6964) for 20 minutes with agitation to release the keratinocytes. Cells were collected and centrifuged, then plated on irradiated MEFs and overlaid with CnT07 media (CellnTec, Bern, Switzerland). Cells were used for experiments in passage 0 or 1.
Keratinocytes were plated onto 100 mg/ml poly-D-lysine coated coverslips, and fixed with 2% paraformaldehyde for 30 minutes. Coverslips were incubated in 0.1% Triton X-100 for three minutes, blocked with 5% normal goat serum, and incubated with primary antibody for one hour at 37°C. Antibody dilutions were as follows: DEK-antibody (Cusabio, Baltimore, MD, USA) 1:300 dilution; keratin 5 antibody (Acris, San Diego, CA, USA) 1:500; and sealed with a coverslip using Vectashield with DAPI (Vector Laboratories, Burlingame, CA). ImageJ (National Institutes of Health, Bethesda, Maryland, USA) [89] was used to quantify Dek staining. Dek immunofluorescences (IF) images were converted to 8-bit images, followed by the identification of the location of cells with the nucleus counter ImageJ plugin. Each cell was visually validated and added to the regions of interest (ROI). The mean grayscale intensity was measured in these ROIs. Quantification was from successive images to encompass the entire coverslip of keratinocytes isolated from each genotype with or without dox treatment.
Tissues were lysed using mortar and pestle, resuspended in RIPA buffer (1% Triton, 1% deoxycholate, 0.1% SDS, 0.16M NaCl, 10 mmol/L Tris pH 7.4, and 5 mmol/L EDTA), supplemented with a protease inhibitor cocktail (Pharmingen, San Diego, CA, USA), and analyzed as described previously (48). Primary antibodies used for DEK were as follows: DEK (1:1000; BD Biosciences, San Diego, CA, USA), pan-actin (1:20,000; a gift from James Lessard). Membranes were exposed to enhanced chemiluminescence reagents (Perkin Elmer, Boston, MA, USA) and imaged using the BioRad Chemidoc (Hercules, CA, USA).
All mice were maintained in a hemizygous state for the Bi-L-Dek and K5-tTA transgenes. All Bi-L-Dek mice were F3 and F4 generations from founder 317. Bi-L-Dek mice were bred to K5-tTA mice and bi-transgenic offspring were given 4NQO water for 16 weeks at a dose of 10mg/ml starting at six weeks of age. Mice on doxycycline were continuously fed dox chow from the start of 4NQO treatment until sacrifice. After 16 weeks on 4NQO, mice were returned to normal water until sacrifice at week 45 or when determined excessively morbid by veterinary services thus warranting sacrifice. At the time of sacrifice, tumors were resected and counted, localization was noted, and tumors were measured by calipers. Tumor volume was measured by (length x width x depth). All statistical analyses were performed in GraphPad Prism. The survival curve was analyzed using the log-rank (Mantel-Cox) test. Tumor incidence was determined significant/non-significant using the Chi Square (and Fisher’s exact) test.
Mouse tumors and tissues were fixed in 4% paraformaldehyde, embedded in paraffin, sectioned at 5 μm thickness, and fixed onto slides. Routine H&E stained sections were analyzed for histopathology.13 The area of microscopic tumors was determined by multiplying the widest part of the tumor by the longest part that was observed in the sections. Paraffin sections were deparaffinized in xylene and rehydrated for antigen retrieval in sodium citrate. Sections were then treated with the Mouse on Mouse peroxidase immunostaining kit (Vector Labs, Burlingame, CA, USA). Sections were stained with diaminobenzidine (DAB) and counterstained with Nuclear Fast Red (Poly Scientific, Bay Shore, NY, USA) and mounted with Permount (Fisher Scientific, Pittsburgh, PA, USA). Images were captured at the indicated magnifications and antibodies used are noted in each case. Antibody dilutions were used as follows: BrdU (1:100, Invitrogen, Calsbad, CA, USA), and DEK (1:200, BD Biosciences, San Jose, CA, USA; or 1:300, Proteintech Group, Chicago, IL, USA; or 1:50, Cusabio, Baltimore, MD, USA).
10x or 20x magnified images of BrdU stained tongue or esophagus were analyzed for BrdU positive cells using ImageJ (National Institutes of Health, Bethesda, Maryland, USA). In ImageJ, the bottom portion of the basal cell layer of the stratified squamous epithelium was traced using the freehand tool and measured in the indicated tissue. The distance was converted into millimeters using scale bars based on magnification to determine BrdU positive cells per millimeter of epithelium. Statistical analysis was performed using GraphPad Prism with t-tests and the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli.
Luciferase assays were performed using the Dual-Luciferase Reporter Assay System from Promega and following manufacturer specifications.
All animal work was conducted according to Cincinnati Children's Hospital Medical Center Institutional Animal Care and Use Committee guidelines under protocol number #2017–0004.
To ameliorate animal suffering mice were euthanized with carbon dioxide when moribund as determined by veterinary services.
|
10.1371/journal.ppat.1002755 | Structural and Functional Insights into the Malaria Parasite Moving Junction Complex | Members of the phylum Apicomplexa, which include the malaria parasite Plasmodium, share many features in their invasion mechanism in spite of their diverse host cell specificities and life cycle characteristics. The formation of a moving junction (MJ) between the membranes of the invading apicomplexan parasite and the host cell is common to these intracellular pathogens. The MJ contains two key parasite components: the surface protein Apical Membrane Antigen 1 (AMA1) and its receptor, the Rhoptry Neck Protein (RON) complex, which is targeted to the host cell membrane during invasion. In particular, RON2, a transmembrane component of the RON complex, interacts directly with AMA1. Here, we report the crystal structure of AMA1 from Plasmodium falciparum in complex with a peptide derived from the extracellular region of PfRON2, highlighting clear specificities of the P. falciparum RON2-AMA1 interaction. The receptor-binding site of PfAMA1 comprises the hydrophobic groove and a region that becomes exposed by displacement of the flexible Domain II loop. Mutations of key contact residues of PfRON2 and PfAMA1 abrogate binding between the recombinant proteins. Although PfRON2 contacts some polymorphic residues, binding studies with PfAMA1 from different strains show that these have little effect on affinity. Moreover, we demonstrate that the PfRON2 peptide inhibits erythrocyte invasion by P. falciparum merozoites and that this strong inhibitory potency is not affected by AMA1 polymorphisms. In parallel, we have determined the crystal structure of PfAMA1 in complex with the invasion-inhibitory peptide R1 derived by phage display, revealing an unexpected structural mimicry of the PfRON2 peptide. These results identify the key residues governing the interactions between AMA1 and RON2 in P. falciparum and suggest novel approaches to antimalarial therapeutics.
| Malaria arises from infection of erythrocytes by single-cell parasites belonging to the genus Plasmodium, the species P. falciparum causing the most severe forms of the disease. The formation of a moving junction (MJ) between the membranes of the parasite and its host cell is essential for invasion. Two important components of the MJ are Apical Membrane Antigen 1 (AMA1) on the parasite surface and the Plasmodium rhoptry neck (RON) protein complex that is translocated to the erythrocyte membrane during invasion. The extra-cellular region of RON2, a component of this complex, interacts with AMA1, providing a bridge between the parasite and its host cell that is crucial for successful invasion. The parasite thus provides its own receptor for AMA1 and accordingly this critical interaction is not subject to evasive adaptations by the host. We present atomic details of the interaction of PfAMA1 with the carboxy-terminal region of RON2 and shed light on structural adaptations by each apicomplexan parasite to maintain an interaction so crucial for invasion. The structure of the RON2 ligand bound to AMA1 thus provides an ideal basis for drug design as such molecules may be refractory to the development of drug resistance in P. falciparum.
| Plasmodium spp., and P. falciparum in particular, are devastating global pathogens that place nearly half the human population at risk to malaria, leading to more than 250 million cases yearly and over one million deaths [1]. The success of the malaria parasite can be attributed to its intracellular lifestyle, invading host cells both in liver and blood stages. Invasion of red blood cells is an active process involving a moving junction (MJ), which is formed by intimate contact between erythrocyte and parasite membranes and is thought to be coupled to the parasite's actin-myosin motor [2], [3]. A number of merozoite antigens, either exposed on the surface or stored in secretory organelles, play a role in the invasion process [4]. One of these is Apical Membrane Antigen 1 (AMA1), a type-one transmembrane protein secreted from the micronemes to the merozoite surface and present at the MJ [5], [6]. AMA1 is highly conserved in the Plasmodium genus [6] and, moreover, in the Apicomplexa phylum to which Plasmodium belongs [7], [8], suggesting a common functional role in diverse host cell invasion scenarios. In the apicomplexan organism Toxoplasma gondii, the receptor for AMA1 was shown to be Rhoptry Neck Protein 2 (RON2), a component of the parasite-derived RON protein complex that is secreted into the host cell during invasion and integrated into the host cell membrane [9], [10]. This interaction was subsequently confirmed in P. falciparum as well [11], [12]. Apicomplexans thus provide both receptor and ligand to drive active invasion.
In many malaria-endemic regions, P. falciparum has become resistant to classic drugs, such as chloroquine, and is rapidly developing resistance to recently introduced drugs. Since both AMA1 and RON2 are specific to Apicomplexa and essential for invasion, interruption of the AMA1-RON2 interaction presents an ideal new target for the design and development of inhibitors. This is supported by the recent observation that the invasion-inhibitory peptide R1 [13], [14] blocks interaction between AMA1 and the RON complex in P. falciparum [15], but due to the polymorphism of AMA1, the effectiveness of this peptide inhibitor is limited to a subset of parasite isolates. Interestingly, R1 does not prevent apical contact but no formation of a functional MJ ensues from this event [15].
Crystal structures of PfAMA1 in complex with invasion-inhibitory antibodies [16], [17] have implicated a hydrophobic groove on Domain I (DI) of PfAMA1 as being critical for function. The topological nature of the PfAMA1 groove [18] is conserved in P. vivax AMA1 [19] and T. gondii AMA1 [20], and contains a number of residues that are conserved or semi-conserved across Plasmodium species, as well as other members of Apicomplexa [21], suggesting that it contributes to the receptor-binding site of AMA1. This was recently confirmed by the crystal structure of TgAMA1 in complex with a synthetic peptide, TgRON2sp, which inserts in the groove of TgAMA1 [22].
Here, we report the crystal structure of the complex formed between PfAMA1 and peptide segments of PfRON2, which, together with our previous structural results on the TgAMA1-TgRON2 co-structure [22], highlights a conserved, crucial interaction in apicomplexan host cell invasion. Functional characterization of hot-spot residues driving AMA1-RON2 complex formation leads to a deeper understanding of key interactions occurring at the MJ of P. falciparum and reveals the molecular basis of cross-strain reactivity while preserving specificity for the species. We also describe the crystal structure of PfAMA1 in complex with the invasion-inhibitory peptide R1 [14], and show that this peptide presents an intriguing structural mimicry of PfRON2. Collectively, our results provide an important structural basis for designing cross-strain reactive molecules that inhibit invasion by P. falciparum.
From the 67-residue construct, PfRON2-5, that we previously showed to have affinity for PfAMA1 [11], and guided by the TgAMA1-TgRON2sp structure [22], we synthesized two analogous PfRON2 peptides: PfRON2sp1 (residues 2021–2059; numbering from the initiation methionine in PF14_0495), and PfRON2sp2 (residues 2027–2055). Significantly, there is no polymorphism in this sequence among P. falciparum isolates. Both constructs incorporate a disulfide-bound β-hairpin loop proposed to be critical in complex formation [22] while PfRON2sp2 is truncated at both the N- and C-termini (Fig. 1A). Since the extracellular region of PfRON2 is non-polymorphic, we determined the affinity of both peptides for PfAMA1 by Surface Plasmon Resonance (SPR) measurements using the 3D7, CAMP, FVO and HB3 proteins to explore the possible effects of AMA1 polymorphisms. The affinity of PfAMA1 from 3D7 for PfRON2sp1 is 25-fold higher than for PfRON2sp2 (Fig. 1B to E, Table 1), highlighting a moderate, yet influential, role for the N- and C-terminal tails. Interestingly, KD values for the PfRON2sp peptides showed no significant variation in binding to PfAMA1 from the four strains.
PfRON2sp1 and PfRON2sp2 were co-crystallized with the first two ectoplasmic domains (DI, DII) of recombinant PfAMA1 3D7 or CAMP strains, respectively. The co-structure of PfAMA1 3D7 PfRON2sp1 (PDB entry code 3ZWZ) was refined to 2.2 Å resolution, while PfAMA1 CAMP PfRON2sp2 (PDB entry code 3SRI) was refined to 1.6 Å resolution (Tables 2, 3). The two co-structures overlay with a root mean square deviation (rmsd) of 0.81 Å in 304 Cα positions, and the two peptides alone overlay with a rmsd of 0.34 Å over the complete length of the modeled PfRON2sp2 (25 Cα) (Fig. 2A). These data confirm that the reduced affinity of PfRON2sp2 is due to the truncated N- and C-termini. Since PfRON2sp1 is more biologically relevant than its truncated counterpart, it is used for the following analyses unless otherwise noted.
PfRON2sp1, traced from Thr2023 to Leu2058, includes a disulfide bridge between Cys2037 and Cys2049 and makes several direct contacts with PfAMA1 (Fig. S1), resulting in a total buried surface area of 3154 Å2 (1441 Å2 for PfAMA1 and 1713 Å2 for PfRON2sp1). Overall, the binding paradigm established by TgAMA1-TgRON2sp [22] is maintained, with an N-terminal helix seated at one end of the AMA1 receptor-binding groove and extended through an ordered coil to a disulfide-closed β-hairpin loop, generating a U-shaped conformation (Fig. 2A). Similarly, exposing a functional receptor-binding groove on AMA1 requires displacement of the extended non-polymorphic DII loop, which adopts a disordered state (not modeled between Lys351 to Ala387); this region is stabilized by DI in apo PfAMA1 (Fig. 2B). Intriguingly, the backbone of the N-terminal helix and additional coil of PfRON2sp1 (2024-QQAKDIGAG-2032) overlays remarkably well with a section of the apo PfAMA1 DII loop (360-YEKIKEGFK-368) (rmsd<0.4 Å), which also includes a helical region (Fig. 2B - box 1). Three water molecules buried by the DII loop in the apo form are retained in the receptor-bound state and facilitate a network of hydrogen bonds that bridge PfAMA1 DI to either the DII loop or PfRON2sp in apo PfAMA1 or the receptor complex, respectively (Fig. 2C). The majority of intermolecular contacts are formed by the segment Lys2027-Met2042 of PfRON2sp1. An influential residue on PfRON2 appears to be Arg2041, a residue specific to the P. falciparum species, located at the tip of the β-hairpin with its guanidyl group fitting snugly into a preformed pocket of PfAMA1 (Fig. 2D).
The invasion-inhibitory peptide R1, comprising 20 residues (VFAEFLPLFSKFGSRMHILK) [14], has been shown by nuclear magnetic resonance (NMR) to bind to the PfAMA1 hydrophobic groove, but this study gave little structural detail of the interaction [15]. We therefore crystallized PfAMA1 3D7 (DI and II) with R1 to compare with the PfRON2 complex. Surprisingly, two molecules of R1 are bound to PfAMA1, which we denote respectively as the major peptide (R1-major), lying deeply in the binding groove, and the minor peptide (R1-minor), lying above R1-major and making fewer contacts with PfAMA1 (Fig. 3 and Table S1). Several solvent molecules bridge directly between PfAMA1 and R1-major. As in the PfAMA1-PfRON2sp complexes, the N-terminus of R1-major binds to a region of PfAMA1 that becomes exposed after displacement of the DII loop.
R1-major makes several direct contacts with PfAMA1 (113 interatomic distances<3.8 Å), including 19 hydrogen bonds and a salt bridge between the amino group of Lys-P11 (R1 peptide residues numbers are prefixed by P) and the Asp227 carboxylate group of PfAMA1 (Table S1A). Contacts made by R1-minor to PfAMA1 are fewer (26 contacts<3.8 Å) and include only five hydrogen bonds (Table S1B). Interactions between R1-major and R1-minor are maintained by a total of 24 interatomic contacts, including three hydrogen bonds (Table S1C). In total, 3025 Å2 of molecular surface is buried between PfAMA1 and the two peptides, with R1-major contributing about 75% to this area. The buried surface between R1-major and R1-minor is 563 Å2, reflecting the smaller number of close interatomic contacts between these two components.
Since the structure of the PfAMA1 3D7-R1 complex revealed two bound peptide molecules, binding measurements of R1 to PfAMA1 3D7 were made by isothermal titration calorimetry (ITC) to examine the stoichiometry (Fig. S2). The measured KD of 145 nM is comparable with previous measurements by SPR [13] and the deduced stoichiometry was 1∶1 over the peptide concentrations used. This implies that the second binding site in the crystal structure (R-minor) has an affinity that could not be determined under the experimental conditions used for ITC but can be estimated to be at least 10-fold weaker than the major site.
While R1-major follows the general contour of the receptor-binding groove, it does so in a linear rather than the U-shaped conformation adopted by PfRON2sp1 (Fig. 4A). R1-minor occupies a similar region in space as the second strand of the PfRON2sp β-hairpin, contacting the same DI loop of PfAMA1 but running in the opposite direction to form a parallel two-stranded β-sheet with the major peptide (Fig. 4A). Portions of R1-major exhibit structural similarity to PfRON2, displaying a 1.2 Å rmsd in the twelve Cα positions (PfRON2sp1, Ala2031 to Met2042; R1-major, Phe-P5 to Met-P16) (Fig. 4A). Moreover, sequence alignment based on the structural superposition reveals a remarkable similarity between the central regions of the two ligands; the segments Ala2031-Met2042 of PfRON2 and Phe-P5–Met-P16 of R1 have five identical amino acids and two conservative differences (Fig. 4B). R1-major residue Arg-P15 contributes the most contacts to PfAMA1 and is positioned within the same pocket of PfAMA1 as PfRON2 Arg2041 (Fig. 4A - box 3) where it maintains six of the seven hydrogen bonds observed for PfAMA1-PfRON2sp. Interestingly, while PfRON2 mimicry is observed in the cystine loop-binding region (Phe2038/Phe-P12 to Arg2041/Arg-P15), R1-major establishes clear anchor points in the hydrophobic groove different from PfRON2; Phe-P2 and Phe-P5 brace the peptide N-terminus in the region exposed by displacement of the DII loop, with Phe-P5 occupying the pocket left vacant by Phe367 of PfAMA1 (Fig. 4A - box 1).
R1 is strain specific, binding to PfAMA1 from the 3D7 (cognate antigen) and D10 strains, but with much reduced affinity to the HB3 or W2mef proteins, as determined by ELISA [14] or SPR [13] measurements (recapitulated in Table S2). In contrast, PfRON2sp1 bound to all the PfAMA1 proteins tested (Table 1) with a higher affinity than for R1 peptide. Consistent with these values, PfRON2sp1 displayed a higher capacity to inhibit red cell invasion by P. falciparum 3D7 than the R1 peptide (Fig. 5). Moreover, PfRON2sp1 shows cross-strain inhibition of invasion as expected from its biological function (Table 1), contrasting with the more restricted strain specificity of R1 (Fig. 5, Table S2) [14].
The PfAMA1 3D7-R1 crystal structure shows that three polymorphic residues (175, 224 and 225) contact R1-major (Table S2). The 224 polymorphism, Met/Leu, is conservative and since contacts are formed by the main chain only, this should not affect R1 specificity. The 3D7 and D10 antigens both carry Tyr175 and Ile225; for the W2mef and HB3 antigens, residue 175 is Tyr and Asp, respectively, and residue 225 is Asn in both. Thus, polymorphisms at positions 225 and possibly 175 appear to be determinant for the 3D7 specificity of R1 at the major peptide-binding site (Table S2A). R1-minor contacts polymorphic residue 230, which is Lys in all strains studied (Table S2B). As our data suggest a weak affinity for this binding site, however, it is unlikely that this polymorphism has a significant effect on the specificity for R1. We examined these polymorphisms further using the mutant PfAMA1 Dico3 [23], which differs only at residue 175 for the 3D7-contacting residues (Table S2A), and a 3D7 mutant with the substitution Ile225Asp, which we call 3D7mut. The equilibrium KD, determined from the SPR steady-state responses to R1 binding, was 15.2±1.9 µM for 3D7mut and 22.3±3.3 µM for Dico3, showing a reduction in affinity of over 200-fold with respect to the native 3D7 antigen (Fig. 6, Table S2C). This affinity is comparable to that observed for HB3 and W2mef [13] (recapitulated in Table S2), and confirms that both Tyr175 and Ile225 are important for the strain-specific recognition of R1. Tyr175, located at the tip of a flexible DI loop that is solvent-exposed in the apo antigen [18], becomes buried by R1-major and forms a hydrogen bond to this ligand via the phenol group. Ile225 is also buried by R1-major, forming a pair of hydrogen bond via its main chain to the R1-major main chain.
Guided by the similarities between the PfRON2sp and R1 co-structures, and the conservation of key contact residues (Fig. 7A), we probed the functional importance of a subset of PfRON2 residues by testing the binding to BHK-21 cells expressing PfAMA1 of GST-PfRON2-5 fusion proteins carrying single alanine mutations at: Pro2033 (aligns structurally with Pro7 of peptide R1, which was shown to be critical for binding [24]), Phe2038 (interacts with invariant residue Phe183 in the hydrophobic groove and aligns structurally with Phe12 of R1), Arg2041 (extensive contacts with PfAMA1 and structurally equivalent to Arg-P15 of R1) and Pro2044 (the peptide bond Ser2043-Pro2044 is cis and is thus important for the β-hairpin conformation). Consistent with the structure, mutation of Arg2041 to Ala abrogated binding to PfAMA1 (Fig. 7B). Similar effects were observed with Pro2044, Phe2038 and Pro2033 mutations, the latter also shown to be a key residue in the TgAMA1-TgRON2 interaction [22].
Similarly, a subset of key PfAMA1 residues was also chosen for mutation: Phe183 (an invariant residue that contributes to the hydrophobic groove and that interacts with Phe2038 of PfRON2 via aromatic interactions), Asn223 (which makes important polar interactions with PfRON2), residue 225 (a polymorphic residue that contributes many contacts to PfRON2 in the structure both the CAMP (Asn225) and 3D7 (Ile225) complexes), Tyr234 (which makes polar contacts to Arg2041 of PfRON2) and Tyr251 (which has been suggested by previous studies to be important [12], [25]). A clear role for Phe183 in the PfAMA1-PfRON2 complex formation was evident when expressed on the surface of BHK-21 cells and tested for their ability to bind GST-PfRON2-5 fusion protein (Fig. 7C). A less pronounced role of Tyr234 was observed and none for the remaining residues, including Tyr251. Although these conclusions differ from those of others [12], [25], these results are consistent with the limited contacts shown by this residue in the structures and with our earlier findings on the TgAMA1-TgRON2 interaction, where the equivalent TgAMA1 residue, Tyr230, had a minimal effect on the binding.
The structure of PfAMA1 in complex with the extracellular region of its receptor PfRON2 and the accompanying functional analysis reveal atomic details of the interaction between two key partners at the MJ. The binding site on PfAMA1 includes the hydrophobic groove and a region that becomes exposed by displacement of the flexible DII loop from its apo conformation. Comparison of residues from both components at the PfAMA1-PfRON2 interface with those of other apicomplexan homologs underscores the separate co-evolution of the receptor-ligand pair in members of the phylum.
The DII loop displays a strong propensity for mobility in P. falciparum [16], [18] and P. vivax AMA1 structures [19], particularly at its N- and C-terminal extremities (weak or absent electron density); the central region of the DII loop is more structured and stabilized by contacts with DI, and is better defined in some of these AMA1 structures. Here, we show that the DII loop is displaced by PfRON2sp, as well as by the R1 peptide. In T. gondii, the DII loop is 14 residues shorter than in the Plasmodium orthologs and appears less mobile [20] but nonetheless is readily displaced by TgRON2sp [22]. Flexibility may therefore have an important functional role: it protects a significant portion of the binding site in apo AMA1 against the host's immune response but can be readily displaced to extend the hydrophobic groove for effective binding to RON2. The anti-PfAMA1 invasion-inhibitory monoclonal antibody 4G2, which binds to the N- and C-termini of the DII loop [19], probably prevents its displacement for effective binding to PfRON2. The absence of polymorphisms in the DII loop in spite of immune targeting of this region underlines its important functional role [21].
We have previously demonstrated an evolutionary constraint on the AMA1–RON2 interaction within apicomplexan parasites [11]. Our functional analysis of the TgAMA1-TgRON2sp co-structure suggested that the cystine loop initially anchors the receptor to the hydrophobic groove, causing expulsion of the DII loop to promote interaction throughout the entire binding site [22]. Comparison of the TgAMA1-TgRON2sp and PfAMA1-PfRON2sp co-structures reveals that the cystine loop, while conserved across the two genera, is the most divergent region within the RON2 (Fig. 8). The separate co-evolution of the AMA1-RON2 pair in Apicomplexa is clearly illustrated by the difference between the cystine loop conformations of PfRON2sp and TgRON2sp. In particular, this allows Arg2041 to access the specific PfAMA1 pocket (Fig. 8), where it participates in an intricate network of polar interactions. From mutagenesis, we have demonstrated a crucial role of Arg2041 in complex formation (Fig. 7B). Moreover, this region of the cystine loop also appears to play an influential role in species selectivity as superposition of PvAMA1 structure [19] onto PfAMA1-PfRON2sp shows that Arg2041 would be sterically hindered at the interface but Thr, the equivalent residue in PvRON2 from P. vivax, can be accommodated (Fig. 9A). This accounts for our prior observation that the original 67-residue segment of PfRON2 does not bind to PvAMA1 [11].
An additional feature of the PfRON2sp cystine loop region is the presence of a cis peptide bond between Ser2043 and Pro2044; the Ser-Pro-Pro segment contributes negligible buried surface area but is important for maintaining the β-hairpin conformation for efficient complex formation. Sequence alignment reveals that the Pro duo (Pro2044–Pro2045) is preserved in all analyzed Plasmodium species (Fig. 8A) and is thus likely important for specific recognition of AMA1. We propose that it provides necessary internal structure at the tip of the cystine loop and places the disulfide bond in the proper orientation to brace the AMA1-RON2 interaction. The influential role of Pro2044 is confirmed by mutagenesis where substitution with Ala, which would disfavor the cis peptide bond, abrogates PfAMA1-PfRON2 binding (Fig. 7B). While T. gondii does not share the conserved proline pair, its cystine loop is two residues shorter (Fig. 8A), which mirrors the narrower groove of TgAMA1. Altogether, the overall U-shape architecture of RON2 in complex with AMA1 appears to be remarkably well maintained within apicomplexan parasites but specific features are clearly visible in the cystine loop of PfRON2 and TgRON2, highlighting how a receptor-ligand complex has evolved to maintain a common and crucial event in the biology of these parasites.
Although the PfAMA1-PfRON2 interface is highly conserved, five polymorphic residues of PfAMA1 contact the non-polymorphic PfRON2sp [26]. Of these, however, only residue 225 (Asn/Ile) varies significantly. The remaining polymorphisms should not affect binding as they involve main chain contacts only (residues 172, 174, 187 and 224). Our study allows a detailed structural assessment of polymorphism at residue 225 since complexes with PfAMA1 from the 3D7 (Ile225) and CAMP (Asn225) strains were determined. The 3D7 and CAMP orthologs both maintain two hydrogen bonds between the main chain of residue 225 and PfRON2 Thr2039. However, Ile225 presents a deep pocket to Arg2041 with apolar contacts formed between the aliphatic regions of these two side chains, while Asn225 presents a shallower pocket to Arg2041 with the Asn225 amide group stacking against the guanidyl group. Nonetheless, our binding studies by SPR show no significant difference in the affinity of these two PfAMA1 homologs for PfRON2sp2. Sequence variations at PfRON2-interacting positions, 172(Glu/Gly), 187(Glu/Asn) and 225 (Ile/Asn) are represented by the strains 3D7, CAMP, FVO and HB3 that we have analyzed by SPR; the very similar KD constants, ranging from approximately 10 to 20 nM, confirm that these exert little effect in the strength of the interaction.
Peptide R1 shows a more restricted specificity as it binds strongly to the cognate 3D7 and closely related D10 antigens but only weakly to orthologs that do not carry the same polymorphic amino acids at position 175 or 225 (Table S2). Tyr175 in PfAMA1 3D7 makes a hydrogen bond to the main chain of R1-major but, as this residue is located in a flexible loop with some freedom to adapt to the PfAMA1-R1 interface, it is unclear why the Asp175 polymorphism leads to reduced affinity. In the case of Ile225 of PfAMA1 3D7, the main chain forms two hydrogen bonds to the main chain of R1-major but the preference of R1 for the Ile225 polymorphism remains unexplained as it contrasts with PfRON2sp where main chain hydrogen bonds are also formed by both Ile225 (3D7) and Asn225 (CAMP) to the main chain of PfRON2. This emphasizes that specificity differences may present subtleties that are difficult to decipher. Here, the crystal structure of R1 in complex with the 3D7mut (Ile225Asn) and Dico3 (Tyr175Asp) mutants of PfAMA1 would provide invaluable insights into this question. Taken together, these results highlight that unlike the natural ligand PfRON2, R1, which was selected by phage display, is highly susceptible to polymorphisms.
R1 exhibits a close structural similarity to PfRON2, with the major/minor peptide pair displaying a similar boomerang form as PfRON2, binding to the same region of PfAMA1 and following the same general contour of the binding-site groove. Our structural data show that binding of R1-minor is dependent upon prior binding of R1-major as it lies above the latter in the binding groove and makes fewer contacts to PfAMA1. This, indeed, is consistent with the ITC measurements that show a stoichiometry of 1∶1, indicating a weaker affinity for the minor peptide-binding site. R1-major is thus favored as the principle inhibitor of the interaction with PfRON2, but this does not preclude a contribution by the minor peptide-binding site at high peptide concentrations.
Therapeutic strategies aimed at inhibiting the interaction between PfAMA1 and PfRON2 should be very effective in treating malaria as they address a critical phase in the life cycle of the parasite and, importantly, should not be compromised by polymorphism since the PfAMA1-PfRON2 interface is highly conserved. Our results provide a structural basis for designing inhibitors against the most virulent malaria parasite. The PfRON2sp1 peptide used in this study has a very high affinity to PfAMA1 and is very efficient at inhibiting invasion. Moreover, in contrast to the less strongly binding peptide R1, PfRON2sp1 is not strain specific. Structural details of the PfAMA1-PfRON2 interaction offer the possibility to design molecules with the desired specific inhibitory properties by in silico screening and structural validation. The binding of PfRON2 Arg2041 to a specific pocket on PfAMA1 could be a critical target region. Indeed, the important role played by Arg-P15 at the PfAMA1-R1 interface closely mirrors the equivalent interaction in the PfAMA1-PfRON2sp complexes and, interestingly, the same pocket is occupied by Arg and Lys in PfAMA1 complexes with the invasion inhibitory antibodies IgNAR [17] and 1F9 [16], respectively (Fig. 9B). Phe2038 (corresponding to Phe-P12 in R1) is also a key residue, as its substitution by Ala affected binding. The importance of this sub-site is further highlighted by the concomitant loss in affinity when Phe183 (with which it interacts) was mutated in PfAMA1. Collectively, these data provide a firm basis for designing molecules with optimal inhibitory properties to treat malarial infection.
(i) Baculovirus insect cell expression: A synthetic codon-optimized gene encoding DI and DII of PfAMA1 3D7 [27] (residues 104–438; numbering based on the initiation methionine, PF11_0344) (GenScript) was subcloned into a modified pAcGP67B vector (Pharmingen) for expression in insect cells using established protocols [20]. Final yield of recombinant protein was approximately 3 mg per L of culture.
(ii) P. pastoris expression: Synthetic genes were optimized for PfAMA1 coding of residues 97–442, from strains 3D7 (Genbank accession number U33274), CAMP (accession number M34552) and HB3 (accession number U33277). Potential N-glycosylation sites were mutated and genes were cloned EcoRI-KpnI in the pPicZalpha A vector (Invitrogen), resulting in an 11-residues sequence extension followed by myc-epitope and hexa-His tags at the C-terminus), expressed in P. pastoris, and purified as described [28]. Yield after purification was approximately 20 mg per L of culture. PfAMA1 FVO (residues 25–545, no tags, accession number AJ277646) was produced as described before [29]. The DiCo3 protein was modified compared to the published protein [23]; it includes the PfAMA1 FVO prodomain (amino acids 25–96) and one additional mutation to minimize proteolytic cleavage Lys376–>Arg (B. Faber, unpublished results). The PfAMA1 3D7mut (Ile225–>Asn, residues 25–545, no tags) mutant was generated by site-directed mutagenesis (Genscript) and produced in P. pastoris in a similar fashion to the native protein [29].
A 39-residue peptide corresponding to residues 2021 to 2059 of PfRON2 (PfRON2sp1) was synthesized by Kinexus (Vancouver, Canada) and disulfide cyclized. Lyophilized PfRON2sp1 was solubilized in 100% DMSO and subsequently diluted in HBS (20 mM HEPES pH 7.5, 150 mM NaCl) for use in co-crystallization and functional studies. Peptides PfRON2sp2 (residues 2027 to 2054) and R1 were synthesized by PolyPeptide (Strasbourg, France) and solubilized in 3.5% DMSO for subsequent use.
Crystals of PfAMA1 3D7 PfRON2sp1 were grown in 30% PEG400, 100 mM Tris-HCl pH 8.5, 200 mM tri-sodium citrate dihydrate and the protein (5 mg/mL final concentration) incubated with PfRON2sp1 (1∶2 molar excess). A crystal in cryoprotectant buffer was flash cooled at 100 K and diffraction data were collected on beamline 9-2 at SSRL (Stanford Synchrotron Radiation Laboratory, Stanford, US). Crystals of PfAMA1 CAMP PfRON2sp2 were obtained in 20% PEG 4000, 0.1 M Tris/HCl pH 8.6, 0.1 M sodium acetate and 20% isopropanol and the protein (6.4 mg/mL final concentration) incubated with PfRON2sp2 (1∶5 molar excess). Diffraction data were collected from a crystal in cryoprotectant buffer at 100 K on beamline ID29 at European Synchrotron Radiation Facility (Grenoble, France). Crystals of PfAMA1 3D7 R1 were obtained in 15% PEG 4000, 0.1 M Tris/HCl pH 8.5, 0.1 M sodium acetate and 10% isopropanol and the protein (5.4 mg/mL final concentration) incubated with R1 (1∶6 molar excess). Diffraction data were collected at 100 K on beamline PROXIMA 1 at SOLEIL (St. Aubin, France).
Diffraction data were processed using Imosflm [30] or XDS [31] and Scala [32] in the CCP4 suite of programs [33]. Crystallographic parameters and data collection statistics are given in Table 2. Initial phases were obtained by molecular replacement using PHASER [34] or AMoRe [35] with the unliganded PfAMA1 structure (PDB 1Z40). Tracing of the PfRON2 and R1 peptides, and addition of solvent molecules, was performed manually in COOT [36] and refinement was performed with Refmac5 [37] or autoBUSTER (Global Phasing Ltd, Cambridge, UK). A summary of refinement statistics is given in Table 3. All molecular representation figures were generated in the PyMOL Molecular Graphics System, version 1.2r3pre, Schrödinger, LLC. Coordinates and structure factors have been deposited in the Protein Data Bank with the following entry codes: PfAMA1-PfRON2sp1, 3ZWZ; PfAMA1-PfRON2sp2, 3SRI; PfAMA1-R1, 3SRJ.
SPR measurements were made with a Biacore 2000 instrument (Biacore AB). AMA1 proteins diluted in 10 mM sodium acetate pH 4.5 for 3D7, CAMP, HB3 and FVO strains, or pH 4.0 for 3D7mut and Dico3, were covalently immobilized by an amine-coupling procedure on CM5 sensor chips (GE Healthcare). The reference flow cell was prepared by the same procedure in absence of protein. Binding assays were performed at 25°C in PBS and 0.005% Tween 20 by injecting a series of peptide (PfRON2sp1 and PfRON2sp2 on 3D7, CAMP, HB3 and FVO, and R1 on 3D7mut and Dico3) concentrations at a constant flow rate of 5 µL/min. A heterologous peptide was used to verify the absence of non-specific binding. Peptide dissociation was realized by injecting the running buffer, and the surface was regenerated by injecting glycine/HCl pH 1.5 followed by SDS 0.05%. Control flow cell sensorgrams were subtracted from the ligand flow cell sensorgrams and averaged buffer injections were subtracted from analyte sensorgrams. For peptide R1, steady-state signals (Req) were obtained directly from the plateau region of the sensorgrams, while for PfRON2sp peptides, estimated values of Req were obtained by extrapolation from the experimental curves since the association phase did not reach a final equilibrium state. All calculations were made using the BIAevaluation 4.2 software (BIAcore AB). The saturation curves obtained by plotting Req versus the peptide concentration were fitted with a steady-state model to obtain the Rmax and the apparent equilibrium dissociation constants, KD. To normalize the response for the different ligands, these curves were reported as the percentage of bound sites (ratio Req/Rmax) versus the analyte concentration..
The P. falciparum cell cultures and the invasion assays were performed as described previously [11]. Briefly, highly synchronized P. falciparum 3D7 and HB3 schizonts (1.5% hematocrit, 1.5% parasitemia) were incubated with R1 or PfRON2sp1 peptides. Blood smears were collected 16 hours post-invasion and used for ring-stage parasites counting. The results presented are representative of three independent experiments, each performed in triplicate.
Cell binding assays using PfAMA1-expressing BHK-21 cells and recombinant GST-PfRON2-5 fusion proteins were performed as previously described [11]. Although not quantitative, this cell-binding assay truly reflects the interaction between AMA1 and RON2 as we carefully checked all the experimental steps as well as the image recording as described below. Transfections were carried out using Lipofectamine Reagent (Invitrogen) as instructed by the manufacturer with 3×105 BHK-21 cells grown on coverslips for 24 h in 6 well plates. Cells were grown for an additional 24 h post-transfection before subsequent analysis. Expression and correct folding of PfAMA1 (and the mutants) at the host cell surface was verified by IFA performed with or without permeabilisation, using antibodies either specific to the cytoplasmic tail (anti-myc tag) or specific to the extracellular ectodomain of PfAMA1 (mouse mAb F8.12.19 [38]). For binding assays, coverslips from a same transfection experiment were washed in HBSS (Invitrogen) before addition of recombinant PfRON2-5 wild type or mutants diluted in HBSS at 10, 1 or 0.1 µg/ml. Coverslips incubated with GST were systematically used as a control. After five washes in PBS to remove unbound protein, cells were fixed in 4% PAF and further processed for IFA as described above [11]. The binding characteristics of RON2 (anti-GST labelling) on the PfAMA1 mutant were only considered valid when its signal was identical to that of wild type PfAMA1. All other micrographs were obtained with a Zeiss Axiophot microscope equipped for epifluorescence. Adobe photoshop (Adobe Systems, Mountain View, CA) was used for image processing. Matching pairs of images were recorded with the same exposure time and processed identically.
The PfAMA1 and GST-PfRON2 mutated constructs were generated by site directed mutagenesis using Quickchange II XL protocol (Stratagene).
|
10.1371/journal.pgen.1000753 | Accelerated Evolution of the Prdm9 Speciation Gene across Diverse Metazoan Taxa | The onset of prezygotic and postzygotic barriers to gene flow between populations is a hallmark of speciation. One of the earliest postzygotic isolating barriers to arise between incipient species is the sterility of the heterogametic sex in interspecies' hybrids. Four genes that underlie hybrid sterility have been identified in animals: Odysseus, JYalpha, and Overdrive in Drosophila and Prdm9 (Meisetz) in mice. Mouse Prdm9 encodes a protein with a KRAB motif, a histone methyltransferase domain and several zinc fingers. The difference of a single zinc finger distinguishes Prdm9 alleles that cause hybrid sterility from those that do not. We find that concerted evolution and positive selection have rapidly altered the number and sequence of Prdm9 zinc fingers across 13 rodent genomes. The patterns of positive selection in Prdm9 zinc fingers imply that rapid evolution has acted on the interface between the Prdm9 protein and the DNA sequences to which it binds. Similar patterns are apparent for Prdm9 zinc fingers for diverse metazoans, including primates. Indeed, allelic variation at the DNA–binding positions of human PRDM9 zinc fingers show significant association with decreased risk of infertility. Prdm9 thus plays a role in determining male sterility both between species (mouse) and within species (human). The recurrent episodes of positive selection acting on Prdm9 suggest that the DNA sequences to which it binds must also be evolving rapidly. Our findings do not identify the nature of the underlying DNA sequences, but argue against the proposed role of Prdm9 as an essential transcription factor in mouse meiosis. We propose a hypothetical model in which incompatibilities between Prdm9-binding specificity and satellite DNAs provide the molecular basis for Prdm9-mediated hybrid sterility. We suggest that Prdm9 should be investigated as a candidate gene in other instances of hybrid sterility in metazoans.
| Speciation, the process by which one species splits into two, involves reproductive barriers between previously interbreeding populations. The question of how speciation occurs has rightly occupied the attention of biologists since before Darwin's “On the Origin of Species.” Studies of recently diverged species have revealed the presence of hybrid sterility genes (colloquially referred to as “speciation genes”), alleles of which are associated with sterility of interspecies hybrids. Mouse Prdm9 is the only known such gene in vertebrate animals. Here we report that the Prdm9 protein has evolved extremely rapidly in its DNA-binding domain, comprising an array of “zinc fingers.” This suggests that hybrid sterility may arise from a mismatch between the DNA-binding specificity of Prdm9 and rapidly evolving DNA. We propose that Prdm9 binds to satellite-DNA repeats evolving rapidly within and between different species. Prdm9 evolution is unusual because other hybrid sterility genes appear only to evolve rapidly in isolated bursts, whereas Prdm9 has evolved rapidly over 700 million years, in many rodent species, diverse primates and other metazoans. This leads to the tantalizing possibility that Prdm9 may have served as a “speciation gene” on other occasions in metazoan evolution, a possibility that will now need to be investigated.
| The question of how two species originate from one has fascinated biologists since before Darwin's iconic treatise on the subject [1]. Postzygotic reproductive barriers between species are thought to result from the acquisition of genetic incompatibilities as an incidental by-product of divergence between two populations. In its simplest form, this Dobzhansky-Muller model involves genetic interactions between two loci (e.g. a and b) [2]. In isolated populations, new alleles can arise and go to fixation in two isolated populations (A in one and B in the other) since they remain compatible with ancestral alleles. However, a negative epistatic interaction between the two new alleles (A with B) in hybrids might result in sterility or inviability, a hallmark of postzygotic isolation in hybrids between two species [3]. Theory predicts that additional incompatibilities will accumulate rapidly following an initial genetic incompatibility [4]. One of the earliest postzygotic isolating barriers in interspecies hybrids is the sterility of the heterogametic sex (XY males or ZW females), a pattern referred to as Haldane's rule that holds almost universally across animal taxa [3],[5]. Examination of early events in speciation that lead to hybrid sterility (for example [6],[7]) is thus vital to gain insight into this mysterious process.
The first hybrid sterility gene to be discovered was the Drosophila Odysseus-site homeobox (OdsH) gene. The D. mauritiana allele of OdsH causes hybrid male sterility when introgressed into D. simulans together with adjacent loci [8],[9]. OdsH encodes a presumptive DNA-binding protein which is exclusively expressed in male reproductive tissues [9]. OdsH function within Drosophila species remained unclear until recently (ablation of the gene in D. melanogaster has a very modest effect on male fertility [10]) as did the molecular basis for why it causes hybrid sterility. However, the manifestation of hybrid sterility appears to be correlated with rapid evolution of OdsH specifically in its DNA-binding homeobox domain, in the species clade that includes D. mauritiana and D. simulans [11].
A second hybrid sterility gene was discovered not as a Dobzhansky-Muller incompatibility but as a result of gene transposition. Hybrids between D. melanogaster and D. simulans, which carry two 4th chromosomes from D. simulans in an otherwise D. melanogaster genetic background, are sterile. This sterility is caused by the transposition of the JYAlpha gene away from the 4th chromosome in D. simulans [12]. Since JYAlpha is required for male fertility, D. melanogaster male flies that only possess D. simulans 4th chromosomes lack JYAlpha and are therefore sterile. In contrast to OdsH, the biological cause of hybrid sterility is well understood but involves no sequence divergence of the underlying sterility gene and only affects a fraction of F2 hybrids.
A third hybrid sterility gene was recently discovered in crosses between the Bogota and USA subpopulations of D. pseudoobscura. F1 males resulting from crosses between Bogota females and USA males are almost completely sterile when young. When aged, however, these F1 males recover partial fertility but produce all female progeny. Intriguingly, a single gene Overdrive (Ovd) was found to be causal for both the segregation distortion and hybrid male sterility [13]. Like OdsH, Ovd encodes a putative DNA-binding protein whose biological function is unclear. Like OdsH, rapid evolution of Ovd in the Bogota lineage appears to be associated with hybrid sterility. Genetic results with Ovd strongly suggest that hybrid sterility is a by-product of intraspecies genomic conflict, manifest as segregation distortion [13].
Prdm9 (Meisetz) is the fourth hybrid sterility gene, the first to be described in vertebrates. It was discovered in crosses between the mouse subspecies Mus musculus musculus and Mus musculus domesticus. Allelic differences at Prdm9 provide the genetic basis for the Hybrid sterility 1 (Hst1) locus, which together with other genetic loci [6],[7],[14], is responsible for spermatogenic failure in sterile hybrids between Mus m. musculus and Mus m. domesticus [15]. Polymorphism linked to Hst1 is associated with sterility traits not only for Mus m. domesticus strains but also, separately, for Mus m. musculus strains [16]. In natural Mus m. musculus populations these polymorphisms appear to have arisen very recently [16]. Prdm9 is a meiosis-specific gene that is only expressed in germ cells entering meiotic prophase in both female and male mice [17]. Loss of Prdm9 causes sterility in both sexes due to impaired meiotic progression at the pachytene stage. Furthermore, nonsynonymous SNPs in human PRDM9 are associated with infertility and azoospermia via meiotic arrest [18],[19]. Prdm9 encodes 3 protein isoforms, of which the largest isoform contains an N-terminal KRAB motif, a central histone H3 Lysine-4-methyltransferase (SET) domain, and several zinc fingers in its carboxy-terminal region (Figure 1). Similar zinc fingers in other proteins have been shown to mediate sequence-specific binding to DNA. The number of zinc fingers encoded in mouse Prdm9 appears to directly affect hybrid sterility. Whereas an allele of Prdm9 encoding 13 zinc fingers causes postzygotic hybrid sterility, an allele containing 14 zinc fingers does not (Figure 1) [15]. The finding that changes in a single DNA-binding determinant appears to be causal for hybrid sterility motivated our analysis to study the evolutionary constraints that shape the sequence and copy number of zinc finger motifs in Prdm9 across a broad taxonomic panel of metazoans, starting with rodents.
We sequenced the terminal zinc fingers from the final exons of Prdm9 from 11 rodent species to which we added the genomic sequences of mouse (C57BL/6J) and rat Prdm9 (Figure 2A), thereby sampling a ∼25 million year period of rodent phylogeny [20]. The C57BL/6J strain of mice is a mosaic of M.m. musculus, M.m. domesticus and M.m. castaneus [21]. The C57BL/6J mouse genome assembly harbours the M.m. domesticus Prdm9 allele [22]. We found that rodents vary greatly in their numbers of zinc fingers present in the C-terminal array: from 7 in Peromyscus polionotus to 12 in Mus musculus (Figure 2A). Even closely-related species pairs, such as field and water voles (Microtus agrestis and Arvicola terrestris), and M. macedonicus and M. spicilegus, differ in their numbers of zinc fingers (Figure 2A).
Rodent Prdm9 zinc finger sequences have been subject to concerted evolution. Many changes in numbers of zinc fingers have resulted from very recent lineage-specific duplications (Figure 2A). Twelve of the 13 rodent species we examined possess at least one pair of Prdm9 zinc fingers that were so recently duplicated that they have identical nucleotide sequences. In one case (Peromyscus leucopus, Figure 2A), Prdm9 encodes a cluster of five zinc fingers that are identical at the nucleotide level, together with another pair of identical zinc fingers. Consistent with concerted evolution, Prdm9 zinc fingers from the same species often form monophyletic clades, even in comparisons of closely related rodents (Figure S1). Such concerted sequence evolution may result from multiple rounds of zinc finger duplication and deletion (‘birth-and-death’ model [23]) to change zinc finger numbers. However, we favor non-allelic gene conversion as a dominant mechanism [24] since it more easily accounts for the many interdigitated and non-adjacent zinc finger duplications, as well as the complexity of the inferred zinc finger phylogeny. Although more occasional gain and loss of zinc finger sequences have been observed previously for other genes [25], the extreme degree of sequence similarity between different zinc finger pairs is far greater for Prdm9 than for any other zinc finger gene present in the C57BL/6J mouse genome sequence (Figure 2B).
In addition to concerted evolution, our analyses reveal evidence for positive selection at particular codons responsible for DNA binding specificities within Prdm9 zinc fingers in rodents. Due to the high degree of concerted evolution, it is not formally correct to carry out a pairwise analysis of the non-synonymous to synonymous rate ratio (dN/dS) when comparing Prdm9 sequences from two different species. Instead, by comparing all Prdm9 zinc fingers within a species, we find that all but one of these 13 rodent species have acquired more amino acid substitutions than would be expected under neutral evolution within their Prdm9 zinc fingers (Figure 3). For instance, in the Prdm9 encoded zinc fingers from Mus musculus strain C57BL/6J (Figure 3A), two codons are predicted to have evolved under positive selection (positions labelled −1 and 3 in Figure 3A). Intriguingly, positive selection is restricted to only a small number of positions within these zinc finger sequences. Sites labelled −1, 3, and 6 were identified as having evolved by positive selection in the majority of the 13 rodent species we examined when comparing all zinc fingers from a particular species (tabulated in Figure 3B). Codons at these sites are turned over rapidly. For instance, two recently diverged vole species, Microtus agrestis and Arvicola terrestris exhibit species-specific codons at positions −1, 3 and 6 (Figure S2) despite their independent evolution only over the last 0.5 million years [26]. In each case, we use the Sitewise Likelihood-ratio method (SLR) [27] with p-value thresholds of 0.05 after multiple testing correction. Since these methods can be strongly affected by tree topology, we tested both the most likely and other competing topologies to conservatively estimate non-synonymous substitutions; this will reduce the chance of false-positives in our analysis (see Materials and Methods). These unusually elevated values may reflect the sustained action of positive selection, consistent with the elevated rates observed for many rodent species (Figure 3). Rapid evolution and addition/deletion of zinc fingers (that provide the basis for hybrid sterility among M. musculus strains [15]) are thus recurrent across rodent evolution.
We also inferred evolutionary rates for each codon from an alignment of every Prdm9 zinc finger from all of these 13 rodent species. Rates for three sites (sites −1, 3 and 6), together with a fourth (site −2), greatly surpass the neutral rate with values of dN/dS up to 8 (Figure 3C). These ratios greatly exceed those found for corresponding positions in other mammalian zinc finger genes [28]–[30]. These three positions (namely −1, 3 and 6) correspond exactly to the positions known to be involved in sequence-specific DNA-binding [31],[32]; structural studies have shown that amino acids within the zinc finger α-helix at positions −1, 3 and 6 make contacts with bases 3, 2 and 1 in the primary DNA strand respectively, whilst the amino acid at position 2 interacts with the complement of base 4 [33]. Thus the finding that positive selection on residues −1, 3 and 6 indicates that it has specifically acted to alter DNA-binding preferences encoded by Prdm9.
Based on our findings in rodents, we next undertook a survey of PRDM9 divergence in the primate lineage to ask whether the extraordinary evolution of Prdm9 was limited to rodents alone. In humans, there appear to be two genes that are orthologous to a single mouse Prdm9, suggesting a recent gene duplication [34],[35]. These two genes, PRDM7 and PRDM9, are found at chromosomal locations 16q24.3 and 5p14, respectively. It is clear that since the gene duplication PRDM7 has acquired distinct tissue-specific patterns of expression and has undergone major structural rearrangements, dramatically altering the number of encoded zinc fingers (2 in macaques, 5 in orangutans) while diverging from ancestral patterns of transcript splicing [34]. Furthermore, there is evidence for a frame-disruption affecting PRDM7 in some humans. Consequently, we do not investigate PRDM7 further in this report.
Primate PRDM9 appears to show a large variation in numbers of zinc fingers in its C-terminal array similar to what we found in rodents (Figure 2A). Chimpanzee, orangutan, rhesus macaque and marmoset PRDM9 genes encode 15, 10, 9, and 9 C-terminal zinc fingers as opposed to 13 in human PRDM9 (Figure 4A). As in rodents, primate zinc fingers also show evidence for concerted evolution. For example, there are three identical pairs out of the C-terminal array of 13 zinc fingers encoded by human PRDM9.
When we compared the PRDM9 gene sequence between humans and chimpanzees, we found the nucleotide divergence to be 7.1%, over 5-fold higher than the divergence observed genome-wide (1.23% [36]) although the high degree of concerted evolution complicates this human-chimpanzee ortholog comparison. However, it does appear that much of the divergence has resulted from a combination of positive selection and concerted evolution. Estimated dN/dS values for positions −1, 3 and 6 of human PRDM9 zinc fingers are 12.6, 9.9 and 13.9 respectively, substantially greater than 1. Indeed, either by a species-specific zinc finger analysis (Figure 4B) or by a pooled analysis of all primate PRDM9 encoded zinc fingers (Figure 4C), we find strong evidence for positive selection at these positions.
Our findings suggest that positive selection and concerted evolution have directly and dramatically altered DNA-binding specificity of the encoded PRDM9 protein in primates as was observed in rodents. For instance, for 12 of the 15 C-terminal array of chimpanzee PRDM9 zinc fingers, codons at position −1 are not found in any human PRDM9 zinc finger at the same position; similarly, 6 human zinc fingers have codons at this position that are not present in the chimpanzee ortholog (Figure 5A and 5B). Like in rodents (Figure 2 and Figure 3), the PRDM9 genes of closely related primate species are differentiated not only by the numbers of zinc fingers they encode, but also by species-specific codons, particularly at key positions that dictate DNA-binding specificity (Figure 4 and Figure 5).
We next investigated whether positive selection on PRDM9 had left population genetic signatures of selection that still remained evident among modern humans. Each of the two methods we employed exploits SNP data and accounts for issues concerning population structure and growth (see Materials and Methods). Particularly recent selective sweeps are characterized by long extents of linkage disequilibrium (LD) that ensue when the haplotype carrying the advantageous allele rises in frequency more rapidly than a neutral allele. Conversely, tests based on this characteristic are particularly sensitive for detecting recent episodes of positive selection [37]. Looking at patterns of LD, we did not find evidence for very recent selective sweeps at PRDM9. In our test we computed the maximum correlation coefficient (r2) between SNP pairs spanning the PRDM9 locus, and compared these to the empirical distribution of this statistic across the genome. These maximum r2-statistics were not significantly different from the background (p values of 0.24, 0.23 and 0.24 for the African, European and Japanese/Chinese population panels).
Since tests based on long extents of LD or haplotypes are sensitive for very recent sweeps [37] only, while tests based on Tajima's D maintain power until some time after fixation of the advantageous allele [38], we also used a Tajima's D estimate to investigate whether polymorphisms linked to PRDM9 exhibit an unusual population frequency spectrum. When an advantageous allele has risen to fixation, the extended haplotype associated with it will, for a considerable time thereafter, carry young and low-frequency polymorphisms, which may be observed as a reduction of Tajima's D, defined as the scaled difference of two estimators of heterozygosity which are identical under the standard neutral model [39]. There are significant caveats to the calculation of Tajima's D from genotyping data which bias against the recovery of low frequency SNPs. The Perlegen genotyping data have been shown to provide useful Tajima's D statistics after empirically accounting for this ascertainment bias [40],[41]. Using these methods, we calculated Tajima's D at the PRDM9 locus [41] in African Americans (D = −0.130; p = 0.038), European Americans (D = −0.259, p = 0.068), and Asian Americans (D = 1.7). With the caveat that there might be uneven distribution of ascertainment biases across the genome, there appears to be weak evidence for a recent selective sweep in African Americans. In contrast to PRDM9, Tajima's D provides no evidence for recent sweeps in any of the three populations at the PRDM7 locus.
We were interested in using intraspecies human polymorphisms to gain further insight into the evolutionary forces that drive the concerted evolution of PRDM9. To this end, we sequenced the terminal PRDM9 zinc finger sequences from 50 Han Chinese individuals, seeking sequence polymorphisms that might have arisen by gene conversion. Under gene conversion, we would expect to observe a nucleotide polymorphism in one zinc finger that is identical to its fixed paralogous base in another. We observed 7 codons containing single nucleotide polymorphisms (SNPs; blue rectangles in Figure 5A and 5C). Of these, 4 (numbered 1, 2, 5 and 7 in Figure 5A and 5C) represent changes to codons that are not represented among any of the remaining zinc finger sequences and thus are unlikely to have arisen by gene conversion. The remaining 3 changes are to codons that are also present in at least one paralogous position within the other zinc fingers. A separate study identified 17 non-synonymous SNPs within human PRDM9 zinc fingers, of which 13 showed evidence for having arisen by gene conversion from paralogous sequences [18]. We infer, therefore, that non-allelic gene conversion has contributed to the rapid evolution of primate PRDM9, and this provides a likely mutational mechanism for many other PRDM9 orthologues.
What are the functional consequences of these non-synonymous SNPs in PRDM9? Two recent genetic association studies have investigated PRDM9 SNPs and their association with azoospermia. The first study [19] did not find correlated SNPs in the C-terminal zinc fingers. However, a second study found that individual nonsynonymous SNPs in the zinc finger domain are associated with a significantly decreased risk of infertility [18]. For instance, human non-synonymous SNPs (labelled 3, 6, 8 and 9 in Figure 5A and 5C) are associated with decreased risk of sterility in a cohort of Japanese men [18], of which two (numbers 3 and 6) were found among the 50 Han Chinese individuals we sequenced. In addition, 3 out of 4 non-synonymous SNPs associated with fertility are found at zinc finger position 6, a site predicted to determine DNA-binding specificity and which we show has evolved under positive selection in human PRDM9 (Figure 4 and Figure 5A). Surprisingly, in each instance, the ‘minor’ allele at each position is associated with protection against sterility in Japanese men [18]. Intriguingly, in both studies, the effect on ameliorating azoospermia or oligospermia was manifest even in the heterozygous condition [18],[19], suggesting that PRDM9's effect is semi-dominant (consistent with results of hybrid sterility seen in mouse Prdm9). In a situation where a minor allele provides a protective benefit against sterility, we might expect that high frequency retention of these alleles would be favored by balancing selection in this population. Consistent with this expectation, we point out that Asian American individuals had a striking Tajima's D of +1.7 in contrast to the negative Tajima's D in the other two populations in the Perlegen dataset, although this statistic by itself is not strong evidence of balancing selection given the ascertainment bias.
The two evolutionary themes (concerted evolution and positive selection) that typify PRDM9 evolution in primates and in rodents also have occurred recurrently across metazoan evolution (summarized in Figure 6). For instance, we found evidence of concerted evolution among Prdm9-encoded zinc fingers in the sea anemone Nematostella vectensis, the gastropod snail Lottia gigantea, and the polychaete worm Capitella sp. I (Figure S3, S4, S5), organisms that last shared a common ancestor with mammals approximately 700 million years ago [42]. In addition, we find strong evidence of positive selection in zinc fingers of N. vectensis Prdm9 for the same 3 positions (namely, −1, 3 and 6) also identified from analyses of rodent and primate lineages (summarized in Figure 6). Estimated dN/dS values for these positions were exceptionally high, ranging between 25 and 32. A single codon of the Capitella worm Prdm9 zinc fingers also shows evidence of positive selection (Figure 6). Thus, even early branching metazoans show strong evidence of both concerted evolution and positive selection within Prdm9-encoded zinc fingers.
Concerted evolution is also apparent in Prdm9 zinc fingers for many mammals including elephants (Loxodanta africana), cats (Felis catus), common shrews (Sorex araneus), cattle (Bos taurus), muntjak deer (Muntiacus reevesi and Muntiacus muntjak vaginalis), bats (Myotis lucifugus) and rabbits (Oryctolagus cuniculus) (data not shown). It is also evident among the zinc fingers of Prdm9 from the Atlantic salmon (Salmo salar) and the rainbow trout (Oncorhynchus mykiss). Of the four complete zinc fingers in rainbow trout Prdm9, two are identical in nucleotide sequence, and the remaining pair are more closely-related to each other than they are to those of Prdm9 for the Atlantic salmon (Figure S6), with which it last shared a common ancestor approximately 20 million years ago [43]. Evidence for positive selection is, however, less compelling outside of these fish, the sea anemone, rodents and primates. This is perhaps owing to the stringent multiple testing correction we employed, especially in cases where there are insufficient zinc fingers to obtain significant power for this kind of analysis (see Materials and Methods).
Despite strong evidence of concerted evolution and/or positive selection in many metazoan Prdm9 sequences, this pattern is not universal across all metazoans. In comparisons of Prdm9 in other ray-finned fishes (including Danio rerio) and in tunicates (including Ciona intestinalis), we found no evidence for either concerted evolution or positive selection within their zinc fingers. Among mammals, we found two homologs of Prdm9 in the platypus Ornithorhynchus anatinus, but evidence for neither concerted evolution nor positive selection. When we investigated the Prdm9 ortholog in the marsupial Monodelphis domestica and the nematode Caenorhabditis elegans, we were surprised to find a complete loss of all zinc fingers. Despite Prdm9 being essential for fertility in mice, Prdm9 appears lacking in chicken (Gallus gallus), frog (Xenopus tropicalis) and fly (Drosophila melanogaster) genomes, while the dog (Canis familiaris) genome has acquired multiple disruptive mutations (“pseudogenization”) within its Prdm9 ortholog [44]. This either implies that Prdm9 function in meiosis is carried out by another gene in these lineages, or that Prdm9's essential function in meiosis is itself lineage- or species-specific.
Our finding of recurrent and dramatic episodes of rapid evolution of Prdm9 in different lineages indicates that the protein-DNA interface at which Prdm9 acts, has frequently altered between, and within, species. These evolutionary observations allow us to revisit some key models of Prdm9 function and how its divergence might give rise to hybrid sterility. The currently prevailing model is that Prdm9 encodes a transcription factor for euchromatic genes during meiosis. Mouse Prdm9 (Meisetz) was first discovered for its essential role in meiotic prophase of both male and female meiosis [17]. Its SET domain was later found to catalyse the specific transition from di- to tri-methylation of the Lysine-4 residue on histone H3 (H3-K4), an activity that is characteristically associated with transcriptional activation [45]. Indeed, by tethering experiments, Prdm9 was shown to be able to activate transcription. Furthermore, in Meisetz−/− testes, the transcriptional regulation of close to 125 genes was disturbed. Thus, Prdm9 (Meisetz) was proposed be a master transcriptional regulator of entry into meiosis in mammals, and all data including the intriguing association with human azoospermia [18],[19] are consistent with this view [17].
However, the accelerated evolution of the Prdm9-DNA interface challenges whether Prdm9's only, or even primary, role is a transcription factor for euchromatic genes. Such a function would leave unexplained why cis-acting (promoter) sequences to which Prdm9 binds, would be subject to repeated positive selection over the long time course of metazoan evolution. Rapid evolution at the protein-DNA interface would be especially disfavoured if it was required for fertility. We cannot formally rule out the unprecedented possibility that a transcription factor may evolve rapidly in concert with all of its (at least 125 [17]) cis-acting binding sites if indeed Prdm9 directly mediates the transcription activation of meiotic promoters. However, in general, the larger the number of cis-acting sequences that Prdm9 has to bind, the more its DNA-binding would be expected to be evolutionarily constrained which, we suggest, argues against its primary role as a transcription factor.
We considered the possibility that the rapid evolution of Prdm9 was actually required for, rather than an impediment to, its function. One of the strongest observations in favor of the transcription model was the fact that the SET domain catalyzed transition from di- to tri-methyl H3-K4, a chromatin mark most often associated with transcriptional activation. And yet, this chromatin mark is not unique to transcriptional activation. Indeed, the same transition from di- to tri-methyl H3-K4, distinguishes canonical H3-nucleosomes at centromeric versus pericentric heterochromatic regions at mitotic centromeres of organisms as diverse as flies and humans [46]. Inactivation of a centromere on a human artificial chromosome directly results in loss of H3-K4 dimethylation and accumulation of H3-trimethylation [47].
We hypothesize that Prdm9's essential role in meiosis is directly related to its ability to bind rapidly-evolving DNA elements. While we do not know the identity of these DNA elements, we speculate that Prdm9 may function by binding directly to repetitive DNA sequences that are found at pericentric and centromeric regions (Figure 7A). Such repetitive DNA sequences (or ‘satellite repeats’) evolve exceedingly rapidly across multiple lineages [48]–[52]. It has been previously proposed that this rapid evolution results from centromere-drive [53],[54], a process in which meiotic products compete during female meiosis for retention in the egg versus exclusion as polar bodies. The genetic opportunity to ‘cheat’ during female meiosis is the evolutionary thread common among many repetitive DNA elements [55]–[58]. Further, DNA-binding proteins are thought to rapidly evolve their DNA-binding specificity to suppress this ‘meiotic drive’ [59]–[63]. Under this model, rapid changes in satellite-DNA sequences potentially ensuing from centromere-drive are followed by positive selection of non-synonymous substitutions within Prdm9 DNA-binding determinants to counter the deleterious effects of the meiotic (centromere) drive process. This would explain not only the rapid evolution and retention of Prdm9 in most metazoans but also the loss of Prdm9 genes in some lineages, when a second satellite-DNA binding protein may have taken over this suppressor function.
A recent study on the Drosophila OdsH hybrid sterility gene provides interesting parallels to the Prdm9 study [64]. Due to its evolutionary descent from the unc-4 transcription factor [11], OdsH was also believed to be a transcription factor. Since the DNA-binding homeobox domain had undergone rapid evolution, hybrid sterility was proposed to result from altered gene expression in Drosophila testis [65], much the same as it has been suggested for Prdm9. However, functional analyses of OdsH revealed it to function as a heterochromatin-binding protein, with altered DNA binding resulting in altered heterochromatic localization and chromosome decondensation [64]. A transcription factor function of Prdm9 (like in OdsH) may be directly tied to a chromosome decondensation function. Indeed, work from a number of model systems especially the fission yeast Schizosaccharomyces pombe has revealed that transcription of heterochromatic repeats is a prequel and often a pre-requisite for the deposition of heterochromatin-specific histone modifications and proteins required for transcriptional silencing and condensation [66]–[69]. Prdm9 binding to satellite-DNA may facilitate its heterochromatinization by virtue of its transcriptional activity (Figure 7A), and alterations of Prdm9's binding specificity could allow it to act on a wider array of satellite-DNAs, consistent with its semi-dominant effect in hybrid sterility and human azoospermia. The chromosome decondensation and synapsis defects in male meiosis observed in sterile hybrids between M. m. musculus and M. m. domesticus species [15] would be explained by an inability to correctly bind and package satellite DNA (Figure 7B). Indirect consequences of such decondensation could be the transcriptional misregulation of some genes, as observed in Prdm9−/− mice [17]. Alternatively, ‘mismatched’ binding of Prdm9 to centromeric satellite-DNA repeats would result in their inappropriate heterochromatinization, again leading to chromosome condensation defects and male sterility. Under this model, mismatched Prdm9-satellite DNA configurations would be predicted to result in sterility only in hybrid males, but not in hybrid females [53]. We would like to emphasize the current absence of functional data to support such a hypothesis. However, the precedence provided by the OdsH study [64] and the consistent rapid evolution seen at Prdm9's DNA-binding interface provides a simple, testable explanation for the onset of highly context-specific hybrid sterility. Variation in Hst1 (Prdm9) occasions a genetic incompatibility between the Prdm9 DNA-binding protein encoded by this locus and the satellite DNAs to which Prdm9 binds (or fails to bind). The finding that human azoospermia is rescued by heterozygous PRDM9 alleles [18],[19], including some that alter DNA-binding preferences, further suggests that a reduced repertoire of satellite-DNA binding ability may be responsible for the meiotic arrest at pachytene seen not only in the hybrid mice species [15], but also in the Prdm9−/− mice [17], a possibility that directly lends itself to genetic and cytological scrutiny.
The proposal that episodes of meiotic drive and suppression drive hybrid incompatibilities is not new [70],[71]. Indeed, cryptic meiotic-drive suppressor systems have been uncovered by introgression analyses between different Drosophila species [72]. Moreover, recent studies of hybrid inviability amongst Drosophila species have revealed the very likely role that pericentric heterochromatin plays in the manifestation of genetic incompatibility [73],[74]. While the molecular function of Prdm9 remains to be fully elucidated, our findings directly implicate the Dobzhansky-Muller incompatibility underlying Prdm9-mediated sterility as residing at a rapidly evolving protein-DNA interface.
The onset of interspecies hybrid incompatibilities is widely believed to ensue as the by-product of acquired genetic differences in geographically isolated populations. This process can be imagined to take place in the absence of any selective pressure, purely by genetic drift [75]. However, the accumulation of genetic incompatibilities is more likely with accelerated evolutionary change, especially if recurrent genetic conflicts were driving the divergence. Consistent with this, many hybrid incompatibility genes for both sterility and inviability are associated with dramatic episodes of positive selection [11],[73],[76].
Here, we have shown that the Prdm9 gene, which was identified as a hybrid sterility gene in mice [15], has evolved rapidly due to the dual forces of concerted evolution and positive selection. This rapid evolution is seen not just across the rodent lineage, but also in primates and especially humans, whereby some alleles at positively selected sites are associated with male sterility via azoospermia due to meiotic arrest. Strikingly, rapid evolution of Prdm9 is observed in some fish, in the sea anemone and a polychaete worm and thus, parsimoniously, is an ancestral feature of metazoan evolution, an evolutionary period spanning 700 million years. This recurrent evolution of Prdm9 is in stark contrast to both the Ovd and OdsH hybrid sterility gene in Drosophila, which appear to have evolved rapidly only in isolated lineages in which its role in hybrid sterility is manifest [11],[13] whereas the gene transposition of JYalpha is also highly lineage-specific [12]. From sequenced transcripts, Prdm9 is known to be expressed in male and female germ-line tissues across diverse metazoans such as trout, cattle, pig, sea urchin, and gastropod snail (accessions: CR372724, EF432552, EW634943, AM222434 and CAXX2975) in line with its previously described expression profile for mouse [17].
Hybrid sterility has been shown to arise from the simple deletion or insertion of a zinc finger domain in Prdm9 in mice [15]. The loss or gain of a single zinc finger is among the least perturbing of all changes in zinc finger number and sequence we have observed. For example, even closely related species, such as humans and chimpanzees, or bank and field voles, or rainbow trout and Atlantic salmon, differ much more dramatically at DNA-binding positions of their Prdm9 zinc fingers (Figure 3, Figure 4, Figure 5, Figure S2, Figure S6). Moreover, findings from human genetic association studies demonstrate that even individual amino acid changes in PRDM9 can affect male fertility even within species [18]. Finally, recent studies clearly demonstrate that Hst1 (Prdm9) associated genetic incompatibilities have evolved independently and are polymorphic in both M. m. musculus and M. m. domesticus mouse subspecies [16]. Our study has found even more radical alterations within Prdm9 zinc fingers than are observed in the M. m. musculus x M. m. domesticus cross. These changes, by themselves, may not be sufficient to result in reproductive isolation, as incompatibilities with a (as yet unknown) rapidly evolving DNA component would be required for hybrid sterility. In addition, hybrid sterility is clearly affected by multiple other loci [6],[7] whose discovery will lend further insight into the biological forces behind hybrid sterility. Nevertheless, our findings of recurrent rapid evolution of Prdm9 suggest its candidacy as a postzygotic hybrid sterility gene in other metazoan taxa.
Prdm9 genes, and their 3′ (carboxy-terminal) arrays of zinc fingers (Figure 1, Figure 2, Figure 3, Figure 4, Figure 5, Figure 6), were predicted from genome sequences available from UCSC, Ensembl and JGI genome browsers. Additional Prdm9 sequences were identified from the interrogation of nucleotide sequence databases using TBLASTn. Prediction of 3′ Prdm9 zinc finger sequences is greatly facilitated by their presence in the single 3′ terminal coding exon in all species. Orthology of Prdm9 sequences was confirmed using phylogenetic analysis [44], by consideration of the KRAB-SET-zinc finger domain architecture that is conserved among many but not all (including some fish, C. elegans and Monodelphis) Prdm9 proteins (see text), and by reciprocal best BLAST hits. Details of Prdm9 gene predictions from all species investigated are provided in Dataset S1.
In addition to genomic data obtained for Mus musculus and Rattus norvegicus, sequencing of the final exon of Prdm9 was performed from genomic DNA purified from reproductive tract tissue from a total of 11 additional (sub-)species: Mus musculus castaneus, Mus macedonicus, Mus spicilegus, Coelomys pahari, Apodemus sylvaticus, Meriones unguiculateus, Peromyscus leucopus, Peromyscus maniculatus, Peromyscus polionotus, Microtus agrestis and Arvicola terrestris. PCR products were amplified using primers designed from the most highly conserved regions from mouse and rat genomic sequence flanking the last exon; either: Mus-Prdm9-F1 5′ CAAAGAACAAATGAGATCTGAG or Mus-Prdm9-F2 5′ AGAACAGGCCAGACAACAAAT with Mus-Prdm9-R1 5′ GTCTT(C/T)CTGTAATTGTTGAGATG or Mus-Prdm9-R2 5′ GCT(G/A)TTGGCTTTCTCATTC. Products were amplified using the proof-reading Pfx DNA polymerase (Invitrogen), purified from agarose gels using the Qiaquick gel purification kit (Qiagen) and sequenced in both directions from 2 or more independent amplification reactions. Sequence traces were initially curated and assembled using Chromas 2.0 (http://www.technelysium.com.au/chromas.html) and Bioedit (http://www.mbio.ncsu.edu/BioEdit/bioedit.html). Genbank accessions are provided in Dataset S2.
Sequencing of the zinc finger repeat domain of PRDM9 was performed from the genomic DNA of 50 Chinese normal control samples. PCR amplification, purification and sequencing was carried out as above using the primers Hs-PRDM9-F 5′-GGCCAGAAAGTGAATCCAGG-3′ and Hs-PRDM9-R 5′-TGAAGCCACCTCACACAGCTG-3′. Products were gel purified and A-tailed prior to sub-cloning into the pCR4-TOPO vector (Invitrogen). T7 and T3 vector primers were used to sequence mini-prep DNA from positive clones. Genbank accessions are provided in Dataset S2.
Chimpanzee (Pan troglodytes) PRDM9 C-terminal zinc fingers were sequenced by PCR using the primers Pt-PRDM9-F 5′-GCCTGACCAAAACATCTACCCTGACC-3′ and Pt-PRDM9-R 5′-GTCATGAAAGTGGCGGATTTG-3′. PCR products were both directly sequenced as well as cloned into the pCR4-TOPO vector (Invitrogen) and six independent clones sequenced using vector-specific primers. The genomic DNA sample was obtained from Coriell (ID#NG03448). The Genbank accession can be found in Dataset S2.
For the prediction of positively selected sites, we included all zinc finger sequences from the 3′ terminal array only if they were complete (28-codon) and retained, at conserved positions, two cysteine and two histidine residues expected to coordinate a single Zn2+ ion. This excludes, for example, the first two zinc finger motifs in primates and rodents. Phylogenetic trees for each multiple alignment were constructed by applying the Fitch-Margoliash criterion to distance matrices of synonymous substitutions per synonymous site (dS) as calculated by the codeml programme [77],[78]. Tree topologies were accepted if they were corroborated by phyml [79] and treebest (http://treesoft.sourceforge.net/treebest.shtml) programs.
Amino acid sites under positive selection were inferred using “site likelihood method” (SLR) [27] with p-value thresholds of 0.05 after multiple testing correction. We observed that inferences of positive selection among sequence similar zinc fingers from the same species were sensitive to tree topology. Nevertheless, the use of alternative less-well supported topologies tended only to increase evidence for positive selection. As a result, we have, conservatively, used inferences from the most strongly supported tree.
SLR, and other maximum-likelihood approaches that take account of codon evolution, have proved reliable provided that assumptions in evolutionary models are not greatly violated. One such assumption is vertical inheritance without gene conversion, which is demonstrably violated for Prdm9. However, gene conversion is more likely to affect analyses of sequence-similar zinc fingers from the same species and is less of a factor in analyzing zinc fingers from all the rodent or primate clades due to the greater sequence divergences involved (for instance, all identical zinc fingers are essentially treated as one representative sequence in analyses). Our inferences of positive selection among all zinc fingers in rodent or primate clades (Figure 3C and Figure 4C) are accordingly the most robust to phylogeny variations and show high dN/dS values, and low and significant p-values.
Rapid fixation of an advantageous allele changes the pattern of polymorphisms around the locus under selection, and various methods have been developed to formally test whether such patterns are compatible with evolution under a neutral model. Other effects, such as geographical structure, population admixture, non-random mating, and varying population sizes, can also give rise to a departure from the neutral model, thereby confounding this analysis. To address this problem, here we use data from recent large-scale surveys of population variation that allow us to compare our observations to empirical, genomic distributions rather than to model-based predictions. This approach accounts for non-local genomic effects such as population structure and growth, at the expense of some loss of power.
Tajima's D values were acquired from the UCSC genome browser for American individuals of African, European and Asian ancestry populations [40]. These were computed at 10 kb intervals, each using 100 kb of data. Since both PRDM7 and PRDM9 span about 20 kb, we took the average of two neighbouring values. For the background distribution, averages were similarly computed for all neighbours.
To assess the existence of long haplotype blocks, we used HapMap data (public release 26). We computed derived allele frequencies (DAF) by polarizing using the chimpanzee genome. To avoid miscalls, we removed all potential CpG SNPs. Finally, we used r-squared values computed for SNPs at a minimum distance of 50 kb, as including more proximal SNPs which are often in strong LD would further reduce power. For any locus, we identified all pairs of SNPs spanning the locus that satisfied these filters; the maximum r-squared value among these pairs was taken as the observable for that locus. We computed this value for all genomic loci to create the empirical distribution. The entire procedure was done separately for each of the HapMap populations.
Clusters of zinc finger repeats (Figure 2B) were identified in each of six possible reading frames of the mouse genome using the hmmsearch programme [80] and a hidden Markov model derived from the SMART domain resource [81]. We discarded all zinc finger clusters which show frameshift or stop codon disruptions, giving 473 putative open reading frames (ORFs). Within each ORF, zinc fingers which do not possess the canonical zinc finger Cys2His2 structure were excluded from subsequent comparisons. A multiple alignment of conceptual cDNA zinc finger sequences was constructed from peptide alignments using the MUSCLE programme [82]. Pairwise cDNA sequence alignments were calculated and the proportions of pairs which were higher than a given threshold calculated. Mouse Prdm9 was an extreme outlier for zinc finger pairwise sequence identities greater than 90%, and also for other thresholds (data not shown).
|
10.1371/journal.pntd.0004910 | Economic Analysis of the Impact of Overseas and Domestic Treatment and Screening Options for Intestinal Helminth Infection among US-Bound Refugees from Asia | Many U.S.-bound refugees travel from countries where intestinal parasites (hookworm, Trichuris trichuria, Ascaris lumbricoides, and Strongyloides stercoralis) are endemic. These infections are rare in the United States and may be underdiagnosed or misdiagnosed, leading to potentially serious consequences. This evaluation examined the costs and benefits of combinations of overseas presumptive treatment of parasitic diseases vs. domestic screening/treating vs. no program.
An economic decision tree model terminating in Markov processes was developed to estimate the cost and health impacts of four interventions on an annual cohort of 27,700 U.S.-bound Asian refugees: 1) “No Program,” 2) U.S. “Domestic Screening and Treatment,” 3) “Overseas Albendazole and Ivermectin” presumptive treatment, and 4) “Overseas Albendazole and Domestic Screening for Strongyloides”. Markov transition state models were used to estimate long-term effects of parasitic infections. Health outcome measures (four parasites) included outpatient cases, hospitalizations, deaths, life years, and quality-adjusted life years (QALYs).
The “No Program” option is the least expensive ($165,923 per cohort) and least effective option (145 outpatient cases, 4.0 hospitalizations, and 0.67 deaths discounted over a 60-year period for a one-year cohort). The “Overseas Albendazole and Ivermectin” option ($418,824) is less expensive than “Domestic Screening and Treatment” ($3,832,572) or “Overseas Albendazole and Domestic Screening for Strongyloides” ($2,182,483). According to the model outcomes, the most effective treatment option is “Overseas Albendazole and Ivermectin,” which reduces outpatient cases, deaths and hospitalization by around 80% at an estimated net cost of $458,718 per death averted, or $2,219/$24,036 per QALY/life year gained relative to “No Program”.
Overseas presumptive treatment for U.S.-bound refugees is a cost-effective intervention that is less expensive and at least as effective as domestic screening and treatment programs. The addition of ivermectin to albendazole reduces the prevalence of chronic strongyloidiasis and the probability of rare, but potentially fatal, disseminated strongyloidiasis.
| Intestinal parasites including hookworm, Strongyloides stercoralis, Ascaris lumbricoides and Trichuris trichuria have been found to be prevalent in refugee populations. More than 50,000 refugees resettle in the United States annually. Since parasitic disease associated with these infections are rare in the United States, there may be delays in diagnosis or improper treatment of refugees, especially for strongyloidiasis. The Centers for Disease Control and Prevention began overseas presumptive treatment programs in some refugee populations in 1999. We examined the cost and economic impact of an overseas presumptive treatment program with albendazole and ivermectin for Asian refugees. We found that the program costs about $15.12 per refugee and that the burden of intestinal parasites would be reduced by about 80% considering cases, hospitalizations, life years, and QALYs. Overall, the cost per QALY gained was $2,219 (95% CI: 600–29,500). Compared to an alternative program in which refugees would be screened and treated for these infections after arrival in the United States, the overseas presumptive treatment program is less expensive and at least as effective.
| More than 50,000 refugees resettle in the United States annually [1] and often arrive with much higher prevalence rates of parasitic infections than is seen in the general U.S. population. [2–5] Since parasitic diseases are rare in the United States, there have been delays in diagnosis or inappropriate treatment [6–10] due in part to inadequate screening strategies. [11] The combination of high prevalence and inadequate domestic management can lead to significant morbidity or even mortality. [12] To better remediate intestinal parasitoses in arriving refugees, the Centers for Disease Control and Prevention (CDC) began overseas presumptive treatment programs in some refugee populations in 1999. Presumptive treatment is administered shortly before departure to the United States to minimize the risk of re-infection.
The initial programs included a single dose of albendazole for intestinal nematodes to all refugees departing from Asia and Africa and sulfadoxine-pyrimethamine for anti-malarial treatment among refugees departing from sub-Saharan Africa. In 2005, the recommendations were expanded to include either ivermectin or a 7-day course of albendazole for Strongyloides stercoralis infections. However, because of funding and logistic issues for procuring and delivering ivermectin, it has only been used in a pilot program. The full CDC guidelines are available at: www.cdc.gov/immigrantrefugeehealth/guidelines/refugee-guidelines.html.
The prevalence of helminth infection was estimated to be 18.6% among Asian refugees according to stool ova and parasite testing prior to the implementation of CDC’s presumptive treatment program [5]. Serologic testing of Asian refugees in the United States have also identified high rates (>20%) of Strongyloides infections. [2, 3] Although Ascaris, Trichuris, and hookworm infections are likely to self-resolve within 1 to 6 years after arrival [13], Strongyloides infections are likely to persist for many years after arrival.
Previously published work suggested that presumptive treatment of immigrants from high-burden countries with albendazole and ivermectin is cost-effective. [14, 15] However, these previous economic analyses had limited data on burden of disease in these populations. Newly available data [2, 3, 5, 6, 16, 17] allow for more accurate estimates of the burden and consequences of these infections in refugee populations. Additionally, earlier studies of the cost-effectiveness are out-of-date because of the changing cost of interventions and testing (e.g., the price for albendazole in the United States has increased from $2.64 per 400 mg in 2000 [14] to almost $120 in 2013). [18] In addition to lower overseas drug costs (frequently <$1.00 per dose), the logistics of overseas presumptive treatment allow for simultaneous interventions in large groups, decreasing the labor resources needed and allowing for better monitoring. We also have detailed cost estimates for overseas programs from ongoing or pilot programs. These new data can be used to quantify the economic impact of CDC’s presumptive treatment recommendations for Asian refugees, both the proposed, but largely unimplemented, use of ivermectin and the already-implemented albendazole. We selected the Asian refugee population for this study to focus on albendazole and ivermectin. The African refugee population was excluded from the analysis because they also receive praziquantel for schistosomiasis and coartem for malaria. However, an analysis of presumptive treatment of African refugees with albendazole and ivermectin is included in the appendix (section 9).
Our analytic objective was to quantify and compare the benefits and costs of overseas presumptive treatment with domestic screening and treatment programs and no intervention for intestinal parasitoses. The conditions included in the analysis were infections with Ascaris lumbricoides, Trichuris trichiura, hookworm, and Strongyloides stercoralis. Per CDC guidelines, domestic screening (stool samples + serology for Strongyloides) is unnecessary for refugees that receive presumptive treatment overseas unless they present with clinical symptoms or persistent eosinophilia.
An economic decision tree model was developed to assess the costs and health impacts of four interventions:
Two other alternatives are included in the appendix (Section 8), which evaluate the replacement of overseas presumptive treatment regimens prior to departure with domestic presumptive treatment programs after arrival. Domestic albendazole and ivermectin presumptive treatment has similar health outcomes as “Overseas Albendazole and Ivermectin”, but costs significantly more. Domestic presumptive albendazole and screening for Strongyloides has similar health outcomes as “Overseas Albendazole and Domestic Screening for Strongyloides”, but again costs significantly more.
We used Markov processes to estimate the probability of outpatient and inpatient treatment and to estimate the number of years with a parasitic infection. To estimate the number of QALYs, we assumed that refugee QALY weights would be similar to that of the average U.S. population by age. [19] Then, we subtracted a small QALY decrement (0.001) for each year spent with a parasitic infection from age-specific QALY weights for U.S. adults. [19] There is considerable uncertainty in this decrement, and it is evaluated in a sensitivity analysis. We discounted costs and health outcomes at 3% annually over a period of 60 years after arrival. In the model, we assumed that refugees died of causes unrelated to parasitic infections at a rate equivalent to that for all Asian Americans [20] because refugee-specific mortality rates were unavailable. We assumed the same drug efficacy for overseas and domestic treatment, but considered an adjustment to reduce overseas efficacy in the sensitivity analyses.
The average age and average number of Asian refugees were estimated using U.S. entry data from 2002–11. [1] Domestic screening and outpatient treatment costs were estimated based on expected clinical procedures valued at Medicare [21, 22] and private insurance [23] reimbursement rates. Presumptive treatment costs were estimated by the International Organization for Migration in 2013; the primary contracted health service provider for U.S.-bound refugees. Key assumptions used to estimate epidemiologic and economic parameters are summarized in Table 1, with more detailed information of all input parameters in S1 Table.
The incremental societal costs per case, hospitalization, and death averted or per life year or quality adjusted life year (QALY) gained were: C_program2 + C_illness2 − [C_program1 + C_illness1] divided by the difference in cases, life years, QALYs, hospitalizations, or deaths (e.g., [Cases1 − Cases2] or [QALYs2 − QALYs1) for these programs. The QALY calculation excluded illness opportunity costs (e.g., time lost from work or other activities due to illness) from the net cost calculation in the numerator.
Although it was assumed that 100% of Asian refugees would have access to presumptive treatment for this analysis, at present, the overseas presumptive treatment program is not available in all countries from which refugees travel. Those who do not receive overseas presumptive treatment (e.g., no presumptive treatment program in country) would receive the same battery of tests and treatment as described in “Domestic Screening and Treatment”. A range of 75%-100% of refugees traveling from countries with overseas presumptive treatment programs was considered in the sensitivity analysis with the remaining refugees that travel from countries without presumptive treatment still requiring domestic screening and treatment. This limitation is examined in a sensitivity analysis. In addition, one-way sensitivity analyses were conducted by varying individual parameters across their uncertainty ranges while holding all other parameters at base case values. Multivariate probabilistic sensitivity analysis was based on Monte Carlo Simulation using simultaneous random draws for each uncertain variable using probability distributions summarized in S2 Table.
We used Treeage Pro 2012 (Williamstown, MA) to build decision tree and Markov models and to conduct one-way and multivariate sensitivity analyses. Treeage output data were exported to Microsoft Excel (Redmond, WA) to create summary tables and figures. The Treeage model is available as a supplemental file.
This activity uses previously collected aggregate data and does not involve contact with human subjects. This analysis was ruled exempt from human subjects review by CDC.
The “No Program” option is the least expensive, costing $165,369 per cohort of 27,700 refugees, $5.99 per refugee (Table 2). Program costs are zero and only outpatient (all four parasites) or inpatient strongyloidiasis cases incur costs. The “Overseas Albendazole and Ivermectin” option ($418,824 or $15.12 per refugee) is less expensive than “Domestic Screening and Treatment” ($3,832,572 or $138.36 per refugee). The “Overseas Albendazole and Domestic Screening for Strongyloides” option is intermediate in cost ($2,182,483 or $78.79 per refugee). The cost of “Overseas Albendazole and Ivermectin” includes overseas presumptive treatment program costs ($269,244) and illness costs ($31,855). In addition, “Overseas Albendazole and Ivermectin” includes domestic costs associated with the need to perform stool ova and parasite testing for 5% of refugees treated overseas with albendazole and ivermectin ($108,030).
For “No Program”, our discounted estimates were 145 outpatient cases, 4.0 hospitalizations, and 0.67 deaths over a 60-year post-arrival period of a one-year cohort of 27,700 Asian refugees. The parasite that has the greatest morbidity and mortality potential is Strongyloides, which we estimate causes 96% of outpatient cases and all hospitalizations and deaths (see S1 Appendix for details).
Considering the impact on QALYs caused by all four parasites, the most effective treatment option is “Overseas Albendazole and Ivermectin”, which would reduce disease burden by an estimated 80%. The screening programs would be less effective (~72%), primarily because of the sensitivity of available diagnostics. For the base case analysis, “Overseas Albendazole and Ivermectin” is both less expensive and results in better health outcomes relative to either of the programs incorporating domestic screening and treatment.
Relative to “No Program”, “Overseas Albendazole and Ivermectin” would cost an estimated $2,146 per case, $76,606 per hospitalization and $458,718 per death averted. The cost per life year and QALY gained are $24,036 and $2,219 respectively. In comparison, the cost per QALY gained is $18,312 for “Overseas Albendazole and Domestic Screening for Strongyloides” and $32,706 for “Domestic Screening and Treatment” relative to “No Program”.
Fig 1 shows the expected cost per QALY gained as functions of Strongyloides prevalence and QALY decrement for infected refugees (holding all other parameters at base case values). At very low QALY decrements, the cost per QALY gained is similar to the cost per life year gained, ~$32,000, and decreases rapidly as the QALY decrement increases. For Strongyloides prevalence, the cost per QALY gained starts at about $40,000 at 1% prevalence and decreases to less than $18,000 at prevalence rates greater than 3%. A value of $50,000 per QALY gained, has commonly been cited as a reasonable threshold for identifying cost-effective interventions.[36] Considering the cost per QALY gained is $2,219 at the baseline prevalence estimate of 20% and remains below $50,000 at prevalence rates as low as 1%, this sensitivity analysis suggests that “Overseas Albendazole and Ivermectin” will probably be considered cost-effective.
Fig 2 is a tornado diagram showing the minimum and maximum cost per QALY gained based on changing one parameter at a time according to the minimum and maximum values shown in S1 Table. After the QALY decrement and Strongyloides prevalence, the parameters with the most influence (in decreasing importance) are 1) the probability that refugees arrive from a country where IOM provides presumptive treatment, 2) discount rate, 3) annual probability of strongyloidiasis hospitalization, 4) ivermectin efficacy, 5) cost of ivermectin presumptive treatment overseas, 6) annual probability of an outpatient strongyloidiasis case, and 7) an adjustment factor for differences in relative overseas and domestic drug efficacy. The maximum cost per QALY gained remains less than $10,000 for all of these parameters. Of note, the cost per QALY gained varies from $472 using the upper bound of the range for the annual probability of hospitalization (0.00012) given Strongyloides infection to $2,775 at the lower bound (0.0000066). The case fatality rate among hospitalized cases is less important, varying from $2,146 at 2% mortality during hospitalization to $2,366 at 25%. In case overseas presumptive treatment is less effective than domestic treatment, a correction factor of 75% is also assessed in the one-way sensitivity analyses and results in a cost per QALY gained estimate of $3,265. The uncertainty in efficacy of ivermectin against Strongyloides (57–99%) results in a range of cost per QALY estimates from $1,943 to $3,959.
If the presumptive treatment program were not available to all refugees and some refugees had to go through domestic screening and treatment, the combined cost of the overseas presumptive treatment and domestic screening and treatment would increase compared to the ideal “Overseas Albendazole and Ivermectin” program in which presumptive treatment is possible for all refugees. Assuming that presumptive treatment could only be provided to 75% of refugees before they relocate to the United States and that the remaining 25% would go through domestic screening and treatment, the cost per QALY gained would increase from $2,219 to about $9,300.
Credibility intervals (95%) can be estimated from the results of the multivariate Monte Carlo Simulations. Relative to “No Program”, the cost per QALY gained varies from $600 to $29,500 for “Overseas Albendazole and Ivermectin” and from $6,000 to $168,000 for “Domestic Screening and Treatment.” A cost-effectiveness acceptability curve is shown in Fig 3. This figure shows the probability that each option is preferred as a function of decision makers’ willingness to pay per QALY gained. At low willingness to pay (<$3,000), the “No Program” option is preferred because this is always the lowest cost option. At willingness to pay per QALY gained > $12,000, the “Overseas Albendazole and Ivermectin” program is preferred in over 90% of the simulations.
Overseas presumptive treatment with albendazole and ivermectin is a cost-effective intervention for improving the health of refugees prior to arrival in the United States. Across a range of parameter estimates, the cost per QALY gained is less than $10,000 compared with waiting for symptomatic refugees to present with illness. Compared with domestic screening programs, overseas presumptive treatment is both less expensive and more effective.
Although all intestinal parasites can detrimentally affect health in refugees, Strongyloides infection presents the greatest threat to refugees due to 1) high infection prevalence in adults, 2) the potential for lifelong infection, and 3) serious sequelae including death, if infected refugees develop dissemination or hyperinfection syndrome due to an immunocompromised state. Assuming the same QALY decrement and probability of outpatient strongyloidiasis treatment as for other parasitic diseases, the potential health burden (cases, QALYs) is more than 20 times greater than for hookworm, Trichuris, and Ascaris infections combined. However, until recently, CDC’s presumptive treatment program has relied on a single dose of albendazole, which would have little effect on Strongyloides prevalence. Further, Stongyloides is not currently targeted as part of any soil transmitted helminth control programs in endemic countries.
This analysis has some limitations. It is difficult to quantify the QALY burden associated with chronic intestinal parasitism, especially in people outside an endemic area. Most people would not be aware that they were infected and would have vague symptoms from conditions associated with infection (e.g., anemia with hookworm, abdominal complaints with strongyloidiasis).[4] Although this study emphasizes Stronglyloides, Trichuris, Ascaris and hookworm infections, albendazole and ivermectin may impact other infections such as other helminths (e.g. enterobius) and ectoparasites (e.g. scabies). This would lead to underestimation of the health impacts and QALYs gained relative to “No Program”. However, as shown in Fig 1, even limiting the analysis to four helminths and using very conservative QALY decrements <0.0005 would make presumptive treatment cost-effective.
In addition, although there are CDC recommendations, we had no actual data regarding the types and frequencies of diagnostic tests given to refugees at domestic follow-up examination. In accordance with CDC guidelines we assumed that at least two stool ova and parasite examinations and one Strongyloides serologic test would be necessary, but examinations may include fewer, or more likely, more extensive diagnostic testing. Finally, we have very limited data on the probability that infected persons seek treatment either as outpatients or inpatients. Strongyloidiasis-related hospitalizations may be underreported because physicians unfamiliar with strongyloidiasis may report acute respiratory failure or gram-negative sepsis as hospitalization causes. [10] When data were unavailable, we tried to use conservative estimates. Nonetheless, these parameters had a small effect on the uncertainty in cost per QALY estimates as shown in Fig 2. Another limitation is that this analysis is limited to scenarios in which it is possible to schedule refugee departures from countries with high prevalence rates of helminth infections to the United States. Such programs cannot be implemented for persons already in the United States seeking asylum status. Similarly, such programs could not be implemented among the unprecedented numbers of refugees entering Europe during 2015–16. [37]
The analysis assumes that domestic physicians will use documented overseas presumptive treatment to forego refugee screening. This requires both adequate documentation of overseas treatment and physician knowledge and acceptance of treatment as sufficient (assuming refugees have no other symptoms consistent with intestinal parasitosis at the post-arrival comprehensive examination). While significant progress has been made, the transmission of overseas refugee medical treatment documentation to domestic physicians remains a work in progress especially as intervention coverage expands and programs evolve. Documentation issues should become easier over time.
In conclusion, the high cost of drugs and diagnostic tests in the United States, inconsistencies in the domestic approach to intestinal parasitosis, and limited provider knowledge about these neglected tropical diseases make overseas presumptive treatment of U.S.-bound refugees a good investment. The addition of ivermectin to albendazole presumptive treatment will improve the health of newly arriving refugees and reduce their long-term risk of complicated strongyloidiasis and hospitalization.
|
10.1371/journal.pcbi.1006412 | Maintaining maximal metabolic flux by gene expression control | One of the marvels of biology is the phenotypic plasticity of microorganisms. It allows them to maintain high growth rates across conditions. Studies suggest that cells can express metabolic enzymes at tuned concentrations through adjustment of gene expression. The associated transcription factors are often regulated by intracellular metabolites. Here we study metabolite-mediated regulation of metabolic-gene expression that maximises metabolic fluxes across conditions. We developed an adaptive control theory, qORAC (for ‘Specific Flux (q) Optimization by Robust Adaptive Control’), and illustrate it with several examples of metabolic pathways. The key feature of the theory is that it does not require knowledge of the regulatory network, only of the metabolic part. We derive that maximal metabolic flux can be maintained in the face of varying N environmental parameters only if the number of transcription-factor binding metabolites is at least equal to N. The controlling circuits appear to require simple biochemical kinetics. We conclude that microorganisms likely can achieve maximal rates in metabolic pathways, in the face of environmental changes.
| To attain high growth rates, microorganisms need to sustain high activities of metabolic reactions. Since the catalysing enzymes are in finite supply, cells need to carefully tune their concentrations. When conditions change, cells need to adjust those concentrations. How cells maintain high metabolism rates across conditions by way of gene regulatory mechanisms and whether they can maximise metabolic activity is far from clear. Here we present a general theory that solves this metabolic control problem, which we have called qORAC for specific flux (q) Optimisation by Robust Adaptive Control. It considers that external changes are sensed by internal “sensor” metabolites that bind to transcription factors in order to regulate enzyme-synthesis rates. We show that such a combined system of metabolism and its gene network can self-optimise its metabolic activity across conditions. We present the mathematical conditions for the required adaptive control for robust system-steering to optimal states across conditions. We provide explicit examples of such self-optimising coupled metabolism and gene network systems. We prove that a cell can be robust to changes in K parameters, e.g. external conditions, if at least K internal metabolite concentrations act transcription-factor binding sensors. We find that the optimal relation of the enzyme synthesis rates of self-optimising systems and the concentration of the sensor metabolites can generally be implemented by basic biochemistry. Our results indicate how cells are able to maintain maximal reaction rates, even in changing conditions.
| Microbes need to grow fast to outcompete others. They therefore have to maintain high growth rates in changing environments. To achieve this specific metabolic fluxes (metabolic rates per unit of expended enzyme) need to be kept as high as possible. Since metabolic enzymes are a limited resource, cells should behave economically: synthesise the right enzymes in the right amounts, and adapt their levels when conditions change. In this paper we show how cells can achieve this in the case when the growth rate itself is fixed, but a limited protein pool needs to be optimally distributed over metabolic pathway reactions to maximise its steady-state rate.
Experimental evidence is mounting that cells are indeed able to tune enzyme levels to maximise the growth rate (Fig 1; [1, 2, 3, 4, 5, 6, 7, 8, 10]). Efficient enzyme allocation has also recently been shown explain measured flux values [11], and to underlie a surprising number of other general physiological phenomena [12, 13], such as the bacterial growth laws [14, 15, 16], overflow metabolism (the Crabtree or Warburg effect; [13, 17]), and catabolite repression [18]. Except perhaps for the case of optimal ribosomal synthesis [15, 16], it is not clear in any of these examples how cells can find the optimal protein expression state out of all possible ones.
Finding optimal states is difficult for microorganisms. They generally do not have sensor proteins in their membranes to alert them of the presence or absence of nutrients or stresses, because their membrane space is limited. It needs to be filled with transporters and respiratory proteins that directly contribute to fitness. Thus cells have to decide how to allocate their resources from internal cues only. Cells are evidently able to accomplish this feat, but that raises the question how they are able to achieve such “blind optimisation”.
Gene expression regulation is largely achieved by transcription factors that are either affected by direct binding of metabolites, or signal transduction cascades, as readouts of environmental and cellular states. Even though transcription factor binding by sensor metabolites is widely accepted in the field [19, 20], the identity of the sensors is only known in a handful of cases (Fig 2). In E. coli, fructose-1,6-bisphosphate (FBP), a glycolytic intermediate, binds to the transcription factor Cra to regulate genes involved in glycolysis [18, 21]; in yeast, the galactose catabolic pathway is induced by intracellular galactose [22]; in E. coli, uncharged-tRNAs induce synthesis of ppGpp when amino acids are limited, leading to the adjustment of ribosome expression [15, 16]; like most most amino acid pathways, the amino acid L-tryptophan regulates the transcription of several enzymes involved in its own biosynthetic pathway [23]; perhaps the best known example is the lactose operon, which is induced by allolactose, an intermediate of the pathway [24]. There is even very recent experimental evidence that E. coli’s central metabolism is in fact controlled by just three such sensor metabolites (cyclic AMP (cAMP), FBP and fructose-1-phosphate (F1P); [25]).
What remains unexplained is why certain sensor metabolites bind to transcription factors and others do not. How many sensors can we expect to be functioning? When do cells rely on just a few sensors? What are the design criteria for regulating circuits that maintain optimal metabolism in fluctuating environments? Does this regulation require complex, hard to evolve, biochemistry or it is almost gratuit? We derive a universal theory, called qORAC (for Specific Flux (q) Optimisation by Robust Adaptive Control), that gives answers to these questions. Understanding how growth rate itself is maximised is beyond the scope of this paper. Instead we focus on the important case of maximising specific rates of metabolic subnetworks at fixed growth rate.
In order to achieve maximal metabolic rates without direct knowledge of those external conditions and how they change, a controlling gene regulatory network must work as follows. At each point in time, internal sensor metabolites must influence a gene regulatory network, causing changes in gene expression. The strength of this signal depends on the concentration of the sensor metabolite. The crucial ingredient is that the gene network must be made in such a way that it expresses proteins at optimal steady state rates with respect to the current sensor metabolite concentration. The network thus ‘assumes’ a steady-state optimum at each point in time. As long as there is a mismatch between the enzyme synthesis rates and the external concentration, so that the metabolic system is not in an optimal state, the system will display dynamics. The sensor metabolite concentration will therefore continue to change and the enzyme synthesis rates will change with it. However, when the enzyme rates are optimal, given the current external environment, a steady state should be reached, which is then necessarily also optimal. In this way, the cell has achieved an optimal state without direct information about it. Its allocation of limited biosynthetic resources for protein synthesis will then be optimal. A gene network, informed by sensor metabolites, that causes optimal steady state enzyme levels in different conditions therefore must necessarily implement some form of qORAC control (Fig 3).
The experimental evidence presented in Fig 1 indicates that a qORAC-like control mechanism is active in cells. If cells are able to reach maximal growth rates (and hence maximal metabolic rates to attain this) in different conditions, at different optimal enzyme concentrations, then the gene regulatory network responsible must necessarily cause the correct enzyme synthesis rates (or must approximate these to a good degree). If this gene network works on the basis of internal metabolic information (rather than on information from signalling pathways, for instance), the control is adaptive, and indeed a form of qORAC control.
Remarkably, the qORAC theory we present here shows that a metabolic system, coupled to its controlling gene network, has a unique optimal steady state, no matter what the environmental conditions are—even though that optimum changes with those conditions. We prove that the dynamics of enzyme synthesis that is required for attaining optimal metabolic states can be inferred from the kinetic rate laws from metabolic enzymes alone. This is in direct agreement with a celebrated engineering principle, the internal model principle [26]. Our results also suggest that the optimising enzyme dynamics of a gene circuits circuit can be achieved with basic biochemistry.
The qORAC theory predicts which metabolites may act as sensors. A fundamental insight is that maintenance of optimal metabolism in the face of N parameters requires N sensor metabolites. The qORAC theory indicates that recent findings, such as the pervasive optimisation of enzyme levels in yeast [9], or the small number of sensor metabolites found in E. coli’s central metabolism [25], should necessarily be seen as surprising.
The phenotypic plasticity of microorganisms is a marvel of evolution. What would be even more remarkable is that cells can maximise their performance in changing conditions, without direct information about those changes. This appears almost impossible in view of the bewildering biochemical complexity of the cell. Part of what we achieve in this paper is to show that this skepticism is most likely unfounded: cells can do this. The insight can explain the robustness to human interventions in metabolic engineering and medicine, and provide opportunities to circuit design in synthetic biology.
We will first introduce the control problem that a cell faces. We consider a well-understood example: the regulation of galactose metabolism in yeast (Figs 1A and 2B). We aim to characterise the dynamics of a controlling gene circuit that always maximises the steady-state flux per unit invested enzyme in this pathway (the specific flux) upon an environmental change, such as in the extracellular galactose concentration. The controlling gene network has to distribute a finite amount of biosynthetic resources for enzyme synthesis over the four pathway enzymes to maximise the steady state pathway flux.
Depending on the external galactose concentration, less or more enzymatic resources should be invested in the galactose import reaction. This leaves a correspondingly smaller or larger pool of enzymatic resources for the remaining pathway reactions. An increase in [Galout] will cause an increase in [Galin], which is therefore indicative for the external change. Galin can thus act as a signal for the adjustment of enzyme concentrations of the pathway: the transporter concentration should decrease and the others should increase.
In yeast, Galin plays the role of metabolic sensor [27]. It relays information to the GAL operon by binding to gal3p, a regulatory protein that can activate transcription factors, such as gal80 and gal4. The key question is how the concentration of Galin should influence the gene network in order to steer the galactose pathway to maximal specific flux.
We refer to the relation between the steady-state concentrations of the metabolic sensor ([Galin]) and the metabolic enzymes as the input-output relation of the gene circuit. qORAC specifies this relation for robust maximisation of specific pathway flux. Whether a gene circuit with realistic biochemical kinetics can be found that can implement this input-output relation then still needs to be determined. Since the gene network for the galactose pathway in yeast is known, the optimal input-output relation may be found by fitting parameters in this network, which we achieved in an earlier paper [28]. In the current paper, however, we show that the problem of finding optimal input-output relations for a given metabolic pathway has a general solution, applicable to all examples shown in Fig 2A–2D. This indicates that cells can implement qORAC using simple regulating circuits.
The qORAC theory starts with the dynamics of the intracellular metabolite concentrations xI = (x1, …, xn) of a metabolic network,
x I ˙ = N v ( x I ; x E ) - μ x I . (1)
Here, N is the stoichiometry matrix, v(xI; xE) is the vector of reaction rates, xE are fixed external concentrations, and μ is the cellular growth rate. It is generally assumed that the dilution rate of concentrations by growth, −μxI, is negligible for metabolism. We take the same view here, and consider
x I ˙ = N v ( x I ; x E ) . (2)
The qORAC framework couples this metabolic pathway to enzyme dynamics, by choosing
e ˙ = E ( x S ) - μ e . (3)
Since enzyme dynamics occur at time scales of similar order as the growth rate, the dilution by growth cannot be neglected this time. Throughout the paper, the growth rate is a predefined parameter, and not part of the optimisation problem (see the Discussion for more information). E(xS) denote the enzyme synthesis rates for all the different enzymes involved in the pathway. These functions may only depend on internal sensor metabolite concentrations, as explained in the Introduction. The task is to define these functions in such a way that the combined dynamical metabolic-enzyme system converges to a steady state in which flux through the pathway is maximal.
As explained in the Introduction, qORAC relies on allocating resources on the basis of sensor metabolite information alone. The optimal allocation must therefore be uniquely defined for each set of sensor concentrations. By considering the optimisation problem in detail, we show that this requires several steps:
We now consider these four steps in detail.
We aim to maximise a steady-state specific flux vr/eT through the network where vr is some chosen output flux (e.g. in mM/hr) and eT (e.g. in grams) is total amount of invested enzyme. The optimisation problem we study is
max x I , e { v r e T | N v = 0 , ∑ j e j = e T } , (4)
with ej as the concentration of enzyme j. Thus, we wish to maximise a given output flux vr per unit of total invested enzyme eT of a metabolic network at steady state.
The optimisation problem stated in Eq (4) is equivalent to minimising the amount of enzyme necessary to sustain a given steady-state flux vr at rate Vr,
min x I , e { ∑ j e j v r | N v = 0 , v r = V r } . (5)
A crucial observation is now that since reaction functions generally are of the form vj = ejfj(xI; xE) [31], we may prescribe vr = 1. After all, if we can solve that problem then we can solve it for vr = Vr as well by multiplying all the enzyme concentrations by Vr, because the specific flux vr/eT remains the same. Hence, we simplify (5) to
min x I , e { ∑ j e j | N v = 0 , v r = 1 } . (6)
The relation vj = ejfj(xI; xE) may also be used to write ej = vj/fj(xI; xE) and rewrite (6) to
min x I { ∑ j v j f j ( x I ; x E ) | N v = 0 , v r = 1 } . (7)
Observe that the enzyme concentration vector e has disappeared from the problem. (Note also that this optimisation is not a stoichiometric-model optimisation, such as flux balance analysis [32]. The qORAC method takes into account the kinetics of the metabolic enzymes and the metabolite concentrations are the variables in this approach. The outcome of qORAC is the definition of a self-optimising dynamical system; this has nothing to do with the optimisation associated with stoichiometric modelling.)
It has recently been shown that the flux profiles that solve (7) (and therefore also the equivalent original problem (4)) are always subnetworks with a particularly simple structure, called Elementary Flux Modes (EFMs; [30, 29]). Such EFMs are one-degree-of-freedom flux vectors satisfying Nv = 0 that cannot be simplified further by deleting reactions without violating the steady state assumption [33, 34]. A given EFM is thus characterised by λ(V1, …, Vm), where λ is a free parameter and the flux vector (V1, …, Vm) is fixed.
If we want to optimise specific flux within a given EFM with flux vector (V1 …, Vm), we still need to find a vector xI for
min x I { ∑ j λ V j f j ( x I ; x E ) } . (8)
This motivates the introduction of the objective function O ( x I ) : = ∑ j λ V j f j ( x I ; x E ) , (9)
which is to be minimised, for given external concentrations xE, by suitably choosing internal concentrations xI. This function is convex for pathways with many kinds of reaction kinetics [11], and in the Supporting Information (SI) we show that it is in fact strictly convex, for an even larger class of rate laws. Hence, the optimum is uniquely specified by the external concentrations xE.
Note that the objective function has a lower value if the values of fj(xI; xE) are higher. Maximising specific flux may thus be reinterpreted as maximising the values of all fj’s simultaneously. These fj are closely associated to the saturation levels of enzyme j with its reactants (and effectors). This optimisation can be done by making as little enzyme as possible, so that the enzymes are used at their maximal capacity.
If we find the vector x I o which minimises O(xI), then we can infer the corresponding optimal enzyme concentrations eo by setting
e j o = λ V j f j ( x I o ; x E ) . (10)
It is clear that we may choose λ = 1 in O(xI): having found the minimiser of O(xI) for λ = 1, we have found it for all λ: the corresponding enzyme levels e j o just scale with λ. In hindsight, we may also for instance normalise the enzyme concentrations such that they sum to a given total concentration eT.
At this stage, the optimal enzyme concentrations that maximise the specific flux at steady state are still defined in terms of external concentrations xE: for each choice of xE, the objective function (9) needs to be minimised to find x I o, and subsequently eo needs to be calculated. In order to characterise gene regulatory networks that produce the right concentrations of enzymes in steady state, robustly with respect to changes in external concentrations but without direct knowledge of those changes, we need to understand the defining characteristics of optimal solutions.
Steady-state optimisers x I o are minima of O(xI), and are dependent on (i.e., parameterised by) xE. So, x I o is a (in fact, the) critical point of O(xI) = O(x1, …, xn), satisfying the optimality relations 0 = ∂ O ∂ x i = ∂ ∂ x i ∑ j V j f j ( x I ; x E ) , i = 1 , … , n . (11)
So instead of minimizing O(x) for given external conditions xE, we could solve (11) by prescribing xE and solving for the remaining variables, the internal concentrations xI. However, the gene network does not have access to xE. Eq (11) should be solved with knowledge of the current sensor concentrations only. We therefore solve (11) by prescribing a subset of the internal metabolite concentrations, sensor values xS, and solving for all remaining concentrations, namely all other internal concentrations, but now also the (unknown) external concentrations. The solution is denoted by ξ = (ξI, ξE), and is the estimated optimal concentration vector, under the assumption of steady state and optimality of the sensor values. In short, we call ξ the optimum as predicted by the sensors. Here, ξE are the external concentrations for which the current sensor values would have been optimal if the pathway had been in steady state. The part of ξI corresponding to sensor metabolites, ξS, of course coincides with the real concentrations xS, by construction. Since ξ is defined by xS, we denote it by ξ(xS).
To solution of ∂O(x)/∂xi = 0 for different sensor values is well-defined mathematically if the Implicit Function Theorem (IFT) holds (see SI for a more detailed exposition). In essence, this means that it is then possible to calculate the optimal allocation by varying the sensors appropriately. The sensors are able to “track” the optima. Any choice of sensor metabolites for which the IFT holds is a candidate for the proposed adaptive control. An immediate consequence of the IFT is that the number of sensor metabolite concentrations must equal the number of changing external metabolite concentrations to which the system needs to be robust. This makes intuitive sense: to track changes (and hence achieve robustness) in N parameters, the gene network should be influenced by (at least) N (independent) internal sensors. Examples of parameters are environmental nutrient concentrations, temperature, pH and toxin concentrations.
With ξ(xS), we can define corresponding predicted optimal enzyme levels, analogous to (10), by setting
e j o = V j f j ( ξ ( x S ) ) . (12)
At these enzyme concentrations, the pathway is either in steady state or not. If not, the metabolic concentrations are still changing, including the sensor concentrations. Hence, the predicted optimal enzyme levels also change. This argument indicates that the only steady state of the metabolic network steered in this fashion is the optimal one.
In the SI we prove that an EFM metabolic pathway with added qORAC control has a unique steady state, the optimum. The proof is fully worked out for linear chains of enzymatic reactions (Theorem 3 in SI), but the techniques of the proof extend to a much larger class of pathways. All one needs to require is that for each choice of enzyme concentrations, the metabolic pathway has a unique steady state (a common enough assumption), and that the sensors are a few reaction steps away from the external concentrations (which makes intuitive sense). This result therefore ensures that when the qORAC-controlled pathway has reached a steady state, it necessarily must be optimal.
We now finish by implementing the enzyme synthesis rate functions Ej in
e ˙ j = E j - μ e j .
By setting
E j = μ V j f j ( ξ ( x S ) ) , (13)
we have ensured that at steady state the enzyme levels are optimal. The complete construction is termed qORAC, and is summarised in Definition 1. A fully-worked out example for the small pathway shown in Figs 3 and S2 is specified in Example 1.
Definition 1 (qORAC): The following differential-algebraic system of equations implements Specific Flux (q) Optimisation by Robust Adaptive Control (qORAC) through an EFM with flux vector (V1, …, Vm) in a cell culture growing at fixed growth rate μ. Let I be the index set of internal metabolite concentrations, E the index set of external concentrations, and S the index set of sensor concentrations. Let furthermore O ( x I ) = ∑ j = 1 m V j / f j ( x I ; x E ) be the objective function. Then we consider for i ∈ I, and j = 1, …, m,
x ˙ i= ∑ j = 1 m N i j v j = ∑ j = 1 m N i j e j f j ( x I ; x E ) , (14)
e ˙ j= E j ( x S ) - μ e j , (15)
E j ( x S )= μ V j / f j ( ξ ( x S ) ) ∑ l = 1 m V l / f l ( ξ ( x S ) ) , (16)
where ξ(xS) = (ξI(xS), ξE(xS)) is the predicted optimum, and is the (time-dependent) solution of
ξ S= x S , (17)
∂ O ∂ ξ i ( ξ )= 0 . (18)
The rescaling of Ej(xS) in (16) by the sum of all the inverses of 1/fj implies that total enzyme concentration is chosen to be equal to 1. Other rescalings give identical results, up to the chosen scaling factor. The choice above, however, is particularly useful, since it ensures positive synthesis rates both for positive and negative metabolic rates through the pathway, and it ensures that it is well-defined also at thermodynamic equilibrium (see SI for details).
Example 1: qORAC for a simple pathway The example qORAC-controlled metabolic pathway from Figs 4 and S2 is specified by the following set of equations for the metabolite concentrations x = (x1, …, x4) = ([C], [C′], [N′], [C3N2]). Note that x1 = [C] is an external concentration which may change value periodically, as shown in Fig 4.
x˙1=0,x˙2=v1−v3,x˙3=v2−v3,x˙4=v3−v4,
where vi = eifi(x), i = 1, …, 4, and the kinetics functions fi(x) are defined by
f 1 ( x )= 0 . 6 x 1 - 0 . 75 x 2 ( 0 . 2 x 1 + 1 . 0 ) ( 0 . 33 x 2 + 1 . 0 ) , f 2 ( x )= [ N ] - 3 . 0 x 3 ( 0 . 5 x 3 + 1 ) ( [ N ] + 1 ) , f 3 ( x )= 0 . 2 x 2 x 3 - 0 . 17 x 4 ( 0 . 2 x 3 / 5 + 1 . 0 ) ( 0 . 33 x 2 + 0 . 17 x 4 + 1 . 0 ) , f 4 ( x )= x 4 - 0 . 0025 0 . 33 x 4 + 1 . 0025 .
The objective function is given by O ( x ) = 1 f 1 ( x ) + ⋯ + 1 f 4 ( x ). The enzyme dynamics are given e ˙ j = E j ( x 2 ) - e j, j = 1, … 4, where
E j ( x 2 ) = 1 / f j ( ξ ( x 2 ) ) ∑ k 1 / f k ( ξ ( x 2 ) ) , j = 1 , … , 4
and the predicted optimum ξ(x2) is defined by
{ ξ 2 = x 2 , ∂ O ∂ ξ 2 ( ξ 1 , … , ξ 4 ) = ∂ O ∂ ξ 3 ( ξ 1 , … , ξ 4 ) = ∂ O ∂ ξ 4 ( ξ 1 , … , ξ 4 ) = 0 .
A toy metabolic network, with two external parameters and one output flux, is shown in Fig 4 (see Box 2 for the mathematical implementation). In this example, only the external [C] concentration is allowed to vary, so one internal sensor metabolite is required. Upon changes in this external concentration, the sensor concentration changes, causing changes in enzyme synthesis, which finally result in adaptation to the new optimum. The optimal enzyme synthesis relations of the gene network are also shown. They are simple curves, suggesting that small gene circuits are sufficient for optimal steering of this pathway.
To illustrate the general applicability of qORAC, consider the complicated branched example network in Fig 5. It has two inputs and two outputs and two allosteric interactions; by employing four sensors, it can be made robust to changes in all four external concentrations.
The qORAC framework is able to start from nearly any initial condition. As an extreme example, with no enzymes present, and only the sensor concentration and no other internal metabolite, the qORAC-controlled pathway still steers to optimum (S1 Fig). Similarly, if the sensor concentrations are ‘wrong’, such that they predict a metabolic flow in the opposite direction to the one dictated by external concentrations, the combined controlled system nevertheless converges to the correct optimum (S2 Fig).
The qORAC control does not guarantee that a metabolic pathway is actually steered towards the optimum. In an example in which one of the periodically changing parameters is a Km parameter of a rate law, the choice of sensors matters critically (Figs 6 and S3). With one choice, the system robustly steers to the optimal specific flux steady state, but with another choice it does not. In both cases, the technical requirements to use the internal metabolites as sensors are met.
In each of the pathways shown in Fig 2A–2D, the sensor metabolite(s) and transcription factor(s) have been identified. Specifying the kinetics for each enzymatic step in the pathway now directly gives the corresponding objective function (9) and the qORAC framework can be set up. The case of galactose uptake (Fig 2B) in yeast has been studied theoretically in detail by [28], including fitting the parameters of the well-characterised GAL gene network to approximate optimal input-output relations. Recent experimental evidence moreover shows that yeast cells are indeed able to tune the levels of these enzymes to optimise growth rate ([9]; Fig 1A).
Experimental evidence is accumulating that suggests that cells can tune their enzyme resources to maximise growth rate [1, 2, 3, 4, 5, 6, 7, 8, 10]. We addressed whether cells growing at a fixed rate can tune limited enzyme resources to steer metabolism to optimal flux states, given only limited information about the current metabolic state of the cell in the form of sensor-metabolite concentrations. We demanded robustness of optimality in the face of environmental changes. We logically derived the qORAC framework, which implements such control for Elementary Flux Modes, the minimal steady state pathways that maximise specific flux [29, 30]. Maximisation of specific fluxes is a requirement for maximisation of the specific growth rate of cells.
We use the term Specific Flux (q) Optimisation by Robust Adaptive Control (qORAC) to describe the regulatory mechanism that we study. ‘Robust’ signifies that attaining optimal states is independent of (environmental) parameter values—the system is robust to them. ‘Adaptive’ means that the control system steers the metabolic system to optimality without direct knowledge of external changes, contrary to the more widely studied problem of ‘optimal control’, in which the steering mechanism works using external changes as inputs to the controller [35].
It is important to note that the growth rate itself is not optimised in our approach. Maximising steady state growth rate rather than specific flux requires a fundamentally different approach. The modelling framework should be extended to Metabolite-Enzyme models in which enzymes are made from precursors [36, 37]. In such models, the growth rate features quadratically rather than linearly, in the resulting steady state and optimality equations. EFMs therefore no longer apply, and the objective function O(x) is also absent. Our approach is therefore more suitable to isolated pathways then to all of metabolism. For such smaller pathways, it is more reasonable to assume that there is a fixed amount of enzyme resources to distribute, and that the cellular growth rate is considered constant. Recent work does suggest, however, that the objective function O(x) studied here in fact matters to cells also on a more global metabolic level [11].
An important finding of our work is that the number of sensor metabolites must be (at least) equal to the number of parameters for which the metabolic pathway is robustly optimal. In other words, if the metabolic pathway always achieves states of maximal specific flux, regardless of the values of three (independently changing) environmental parameters, such as, for example, osmolarity, temperature and some nutrient concentration, then the number of sensors is expected to be three. This is a general result that follows from the associated mathematics of this control problem. Finding the sensors experimentally is difficult, and the number of known sensors is still quite small. However, it is telling that the whole of central carbon metabolism in E. coli seems to be controlled by just three sensors, FBP, cAMP and F1P [25].
The identity of suitable sensors does not follow immediately from the optimisation problem. In general, one needs to make sure that the Implicit Function Theorem applies to the optimum Eq (11), and this is not a trivial matter. However, a different argument shows that sensors near the beginning or ends of the pathway would work in most cases. The reason is that for all metabolites in between a set of fixed concentrations, their optimal value is uniquely determined by minimising the corresponding optimisation problem (i.e. finding the minimum of a suitable objective function O(x; xS) with x the set of metabolites between the sensors xS). The remaining variables, including the external concentrations, then need to be determined using the optimum Eq (11). This is easiest (it involves the smallest number of equations and unknowns to solve for) when sensors are close to the external metabolites. Also from a biological standpoint this makes sense: such sensors obviously provide the most information of any change in external concentrations.
An important question is whether the adaptive control can be achieved by molecular circuits, given our understanding of biochemical kinetics and molecular interactions. The explicit example from galactose metabolism in yeast [28] gives hope that this might be true in general. If the necessary gene network is small, then the optimal circuit is likely also evolvable. We cannot give definite answers about this, but the computational analyses of different networks, of which some are shown in this paper, indicate that qORAC-controlled networks show remarkably simple dynamics and input-output relations. One would expect that biochemical systems are capable of evolving those, and that synthetic biologists are capable of designing them.
The parameterisation of the optimising circuit is completely determined by the kinetics and the wiring of the metabolic pathway that it controls, since the objective function (9) contains only this information. This interdependence between the controller and the controlled is sometimes called the ‘internal model principle’ in engineering [26] which roughly states that the control system should have knowledge of the dynamic behaviour of the system in order to be able to control it. Additional control mechanisms may then prevent for instance undesired oscillations or slow responses.
The internal model principle, applied to metabolic pathway control, suggests a new perspective on the larger problem of understanding metabolic regulation. The theory presented here indicates that knowledge of the metabolic pathway, including properties of catalysing enzymes, is sufficient to understand how this pathway needs to be controlled to maximise flux. It is not necessary to know the controlling regulatory pathway in advance. This offers hope for situations in which this circuit has not been characterised yet, or for which it needs to be designed synthetically.
Technological advances have spurred recent interest in studying control properties of gene regulatory networks in cellular metabolism. One line of work involves characterising a particular gene control system and studying its theoretical properties. Examples are the perfect adaptation in the chemotaxis network in E. coli [38, 39], the robustness properties of the heat-shock response system [40] and of the circadian clock [41]. Several authors have considered dynamic optimisation of resources in pathways from a mostly computational perspective, e.g. to minimise the time of adaptive response [42], deFBA [43], and for other objectives than maximal specific flux, such as detecting equilibrium regimes of pathways [44], robustness to flux perturbations [45], and noise propagation [46]. In many studies, the control is not adaptive, but optimal; the objective is then usually to maximise the long term production of biomass [47, 42, 48, e.g.].
The approach taken here differs principally from most previous works in the following respect. The objective (maximal specific flux) is defined in advance, and the optimal input-output relations are characterised later. The framework is also analytic rather than computational: the input-output relations are obtained by solving the optimum equations (11) for the pathway, rather than by using a numerical optimisation routine. The latter is impossible, since this would require knowing the external concentrations.
A few recent papers have used adaptive controls similar to ours. So-called Flux Control Regulation (FCR; [49]) comes closest, and uses the same type of adaptive control as proposed in qORAC. FCR also explicitly relies on making estimates at each time point under the assumption of steady state. When the system is in fact in steady state, it has reached the desired objective. The principle difference between FCR and qORAC lies in the objective. The input-output relations in FCR come from measurements and ensure steady state properties only. qORAC, however, solves a steady state optimisation problem, and constructs input-output relations directly from the kinetic rate laws of the metabolic pathway itself. Another recent example of a coarse-grained model of cellular physiology including gene expression control can be found in [50]. Two other examples using adaptive control are from the context of optimal ribosomal allocation to maximise the growth rate in E. coli. The free amino acid concentration acts as a sensor to ppGpp, which downstream influences gene expression. Two models have been proposed that are based on optimal synthesis of ribosomes so as to maximise growth rate [15, 16]. The input-output relations used in these models are not derived from kinetic properties as in qORAC, but are designed by hand to approximate maximal growth rates in different conditions.
The choice of sensors sometimes matters for the control to steer the pathway to optimum (Figs 6 and S4). This example already indicates that, although the qORAC control follows logically from the design objective, it is not easy to decide which intermediate metabolites make it controllable. We cannot expect completely general mathematical theorems. Apparently, some choices of sensors do work, and others do not, for the same pathway, using the same initial conditions. A second, mathematical reason why one cannot expect convergence to optimal states is that if time would be reversed, the control would remain the same, but dynamics would be reversed. The control is based on steady state properties of the system, and these do not change upon time reversal.
qORAC has direct applications in synthetic biology. To achieve maximal production rates in a biotechnological-product producing pathway requires a controller that qORAC provides. The only ingredient to design such a controller are the enzymatic rate laws in the pathway. qORAC then immediately makes predictions about the optimal enzyme synthesis rates, as a function of one or more intermediate metabolites. As the synthetic biology field advances, synthetic circuits with the required input-output relationships for the constituent enzymes of the pathway can be designed and built. qORAC therefore does not only contribute to the general understanding of steering mechanisms to optimal states, but provides direct operational relevance for microbiology, synthetic biology and biotechnological applications.
|
10.1371/journal.pntd.0001302 | Meningococcal Factor H Binding Proteins in Epidemic Strains from Africa: Implications for Vaccine Development | Factor H binding protein (fHbp) is an important antigen for vaccines against meningococcal serogroup B disease. The protein binds human factor H (fH), which enables the bacteria to resist serum bactericidal activity. Little is known about the vaccine-potential of fHbp for control of meningococcal epidemics in Africa, which typically are caused by non-group B strains.
We investigated genes encoding fHbp in 106 serogroup A, W-135 and X case isolates from 17 African countries. We determined complement-mediated bactericidal activity of antisera from mice immunized with recombinant fHbp vaccines, or a prototype native outer membrane vesicle (NOMV) vaccine from a serogroup B mutant strain with over-expressed fHbp. Eighty-six of the isolates (81%) had one of four prevalent fHbp sequence variants, ID 4/5 (serogroup A isolates), 9 (W-135), or 74 (X) in variant group 1, or ID 22/23 (W-135) in variant group 2. More than one-third of serogroup A isolates and two-thirds of W-135 isolates tested had low fHbp expression while all X isolates tested had intermediate or high expression. Antisera to the recombinant fHbp vaccines were generally bactericidal only against isolates with fHbp sequence variants that closely matched the respective vaccine ID. Low fHbp expression also contributed to resistance to anti-fHbp bactericidal activity. In contrast to the recombinant vaccines, the NOMV fHbp ID 1 vaccine elicited broad anti-fHbp bactericidal activity, and the antibodies had greater ability to inhibit binding of fH to fHbp than antibodies elicited by the control recombinant fHbp ID 1 vaccine.
NOMV vaccines from mutants with increased fHbp expression elicit an antibody repertoire with greater bactericidal activity than recombinant fHbp vaccines. NOMV vaccines are promising for prevention of meningococcal disease in Africa and could be used to supplement coverage conferred by a serogroup A polysaccharide-protein conjugate vaccine recently introduced in some sub-Saharan countries.
| Epidemics of meningococcal meningitis are common in sub-Saharan Africa. Most are caused by encapsulated serogroup A strains, which rarely cause disease in industrialized countries. A serogroup A polysaccharide protein conjugate vaccine recently was introduced in some countries in sub-Saharan Africa. The antibodies induced, however, may allow replacement of serogroup A strains with serogroup W-135 or X strains, which also cause epidemics in this region. Protein antigens, such as factor H binding protein (fHbp), are promising for prevention of meningococcal serogroup B disease. These proteins also are present in strains with other capsular serogroups. Here we report investigation of the potential of fHbp vaccines for prevention of disease caused by serogroup A, W-135 and X strains from Africa. Four fHbp amino acid sequence variants accounted for 81% of the 106 African isolates studied. While there was little cross-protective activity by antibodies elicited in mice by recombinant fHbp vaccines from each of the four sequence variants, a prototype native outer membrane vesicle (NOMV) vaccine from a mutant with over-expressed fHbp elicited antibodies with broad protective activity. A NOMV vaccine has the potential to supplement coverage by the group A conjugate vaccine and help prevent emergence of disease caused by non-serogroup A strains.
| For more than 100 years devastating epidemics of meningococcal disease have occurred in sub-Saharan Africa [1]. In the decade 1988 to 1997, more than 700,000 cases and over 100,000 deaths were reported. Public health responses were limited by scarce resources [2]. Further, the only vaccines available, un-conjugated (plain) polysaccharides, elicited incomplete and short duration of protection in young children [3], [4], and had a minimal effect on decreasing transmission of the organism [3], [5], [6]. Control of epidemic meningococcal disease in Africa, therefore, remains an unsolved public health challenge.
Most meningococcal disease in industrialized countries is caused by strains producing capsular serogroups B, C or Y, whereas most disease in sub-Saharan Africa is caused by serogroup A strains. After more than ten years of work [7], [8], a promising serogroup A polysaccharide-protein conjugate vaccine recently was developed for sub-Saharan Africa [9], [10]. As of January 21, 2011, nearly 20 million people had been immunized as part of demonstration projects in three countries (http://www.path.org/menafrivac/index.php). While this vaccine has the potential to eliminate serogroup A epidemics, widespread vaccination may result in selective pressure for replacement of strains with other capsular serogroups such as X or W-135, which have caused epidemics in this region [11]–[14]. However, with the possible exception of Spain [15], there is little evidence of serogroup replacement after widespread use of monovalent serogroup C meningococcal conjugate vaccines in Europe [16], [17]. Pneumococcal serotype replacement, in contrast, has been a problem in many countries where pneumococcal polysaccharide-protein conjugate vaccines were introduced [18]–[22].
A number of protein antigens are being developed for prevention of meningococcal serogroup B disease (Reviewed in [23], [24]). These antigens also are prevalent in meningococcal strains with other capsular serogroups [7], [25], [26]. Therefore, the vaccines have the potential to prevent disease caused by non-group B strains. One of the most promising of the new protein vaccine candidates is factor H binding protein (fHbp, which was previously referred to as GNA1870 [27] or LP2086 [28]). Recombinant fHbp is part of two vaccines in clinical development for prevention of serogroup B disease [28]–[30]. Native outer membrane vesicle vaccines from meningococcal mutants with over-expressed fHbp also are under investigation [31]–[36]. The fHbp antigen is a surface-exposed lipoprotein that binds complement fH [37], which down-regulates complement activation and enhances the ability of the organism to escape complement-mediated bacteriolysis [37]–[39]. In immunized mice and humans, antibodies to recombinant fHbp vaccines elicited complement-mediated serum bactericidal activity [27], [28], [30], [40]–[45], which in humans is the hallmark of protection against meningococcal disease [46]–[48]. The present investigation was undertaken to determine the vaccine-potential of fHbp for control of meningococcal epidemics in Africa caused by serogroup A, W-135 and X isolates.
In a previous study, we characterized fHbp sequence variants in a small collection of serogroup A, W-135 and X isolates from patients in sub-Saharan Africa [49]. The objectives of the present study were to determine fHbp sequence variants in an expanded panel of case isolates from Africa, to measure levels of fHbp expression, which in previous studies had been reported to be important for predicting susceptibility of serogroup B strains to anti-fHbp bactericidal activity [50], and to investigate the immunogenicity in mice of recombinant fHbp vaccines representative of sequence variants prevalent among invasive African strains. The recombinant vaccine immunogenicity results were compared to that of a prototype native outer membrane vesicle vaccine (NOMV) prepared from a serogroup B mutant strain engineered to over-express fHbp. Our hypothesis was that the NOMV would elicit broader serum bactericidal antibody responses against the strains from Africa than the recombinant fHbp vaccines since in previous studies, mutant NOMV vaccines with over-expressed fHbp elicited broader bactericidal activity against serogroup B strains [33]–[36], [51].
Vaccine immunogenicity was evaluated in CD 1 mice in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Children's Hospital & Research Center at Oakland Institutional Animal Care and Use Committee. Blood collection was performed under anesthesia, and all efforts were made to minimize suffering. The human complement source for measuring serum bactericidal activity was serum from an adult who participated in a protocol that was approved by the Children's Hospital Oakland Institutional Review Board (IRB). Written informed consent was obtained from the subject.
Antibody concentrations were transformed (Log10). For calculations of geometric mean antibody concentrations, concentrations below the limit of the assay were assigned a concentration half of the lower limit. Two-tailed Student's t tests were used to compare the geometric mean antibody concentrations between two independent groups of mice. All statistical tests were two-tailed; probability values of less than or equal to 0.05 were considered statistically significant.
The distributions of the prevalent clonal complexes, fHbp variant groups, and fHbp sequence variants are shown in Figure 2. The serogroup A isolates were derived from three clonal complexes: CC 1 (three isolates from 1963 to 1967 and one from 1989), CC 4 (ten from 1966 to 1990), and CC 5 (three from 1988–1999, and fourteen from 2003–2007). In contrast, the 53 serogroup W-135 isolates were predominately from CC 11 (N = 47 from 1994 to 2007). Among the 22 serogroup X isolates, four were members of a new CC (designated CC 181 (http://pubmlst.org/neisseria/). Of the remaining 18 isolates, three from the 1970′s were sequence type (ST) 3687, and 15 from 2006–2007 were ST 5403. These two STs differed by only a single nucleotide change in one of the seven loci (fumC) and, thus, may represent an undesignated CC, which accounted for 82% of the serogroup X isolates.
There also were a limited number of PorA VR types (Figure S2). Among the serogroup A isolates, P1.20,9 was present overall in 55 percent, and in 89 percent of 18 serogroup A isolates obtained since 1990. Among the serogroup W-135 isolates, P1.5,2 and related types such as 5-1,2-2 predominated (98%), and among the serogroup X isolates, P1.19,26 and related type P1.19,26-4 accounted for 68%. The PorA VR typing results are consistent with previous studies of strains from sub-Saharan Africa [58], [59].
One hundred percent of the serogroup A isolates, 95% of the X isolates, and 34% of the W-135 isolates had fHbp in the variant 1 group (Figure 2, middle column). The remaining W-135 isolates had fHbp variant 2 (58%) or 3 (8%); one serogroup X isolate had fHbp variant 3.
The distribution of the major fHbp amino acid sequence variants is shown in Figure 2 (right column). All of the group A isolates had fHbp sequence variants ID 4 or 5, which differed from each other by a single amino acid. The most prevalent sequence variants in the serogroup X isolates were ID 74 (67%) and ID 73 (12%); two of the serogroup X isolates in the category “other” had fHbp ID 4, which was prevalent among serogroup A isolates. With only a few exceptions the W-135 isolates were clonal based on having a common clonal complex (CC 11) and PorA VR type (P1.5,2, see Figure S2). This clone could be subdivided into multiple subclones based on genes encoding fHbp ID 9 (variant 1), 22/23 (variant 2, and which differed from each other by one amino acid), or 349 or 111 (variant 3; included in the category “other”, Figure 2). Collectively, four fHbp sequence variants (or related variants, each differing from the respective prevalent variant by 1 amino acid) were present in 81% of the 106 isolates. These were fHbp ID 4/5, 74 or 9 (variant 1 group), or ID 22/23 (variant 2 group). The percent amino acid identities between each of these sequence variants, and between fHbp sequence variants ID 1, 28 and 77, which were used as control vaccines, are summarized in Table 1.
We determined fHbp expression levels in bacterial cells from 44 isolates (Figure 3) using a quantitative Western blot. Six of the 16 group A isolates tested (Panel A), and nine of the 14 W-135 isolates tested with fHbp in the variant 1 group (Panel B), had low fHbp expression (defined by ≤33% of fHbp expressed by the group B reference strain H44/76, which is a naturally high expresser of fHbp ID 1 in variant group 1). All but one of seven group W-135 isolates tested with fHbp in the variant group 2 expressed ≤33% of the group B reference strain 8047, which is a naturally high expresser of fHbp ID 77 in variant group 2 (Panel D). In contrast, all but one of the seven group X isolates tested with fHbp in variant 1 group had high fHbp expression (>100% expression compared with H44/76, Panel C). The seventh isolate had ∼80% of the fHbp expressed by H44/76, which was considered intermediate.
We prepared isogenic mutants with different levels of fHbp expression from three isolates: a serogroup B reference strain, and serogroup A and W-135 African isolates. Figure 4 panels A, B and C illustrate relative expression of fHbp on the surface of live bacteria as measured by flow cytometry. Panels D, E and F show the corresponding fHbp expression measured in solubilized bacterial cells by a quantitative Western blot [50]. For each set of mutants, there was a 5- to 10-fold range between lowest and highest fHbp expression.
The wildtype serogroup B reference isolate had naturally high expression of fHbp ID 1. Sera from mice immunized with each of the recombinant fHbp ID 1, 4, 9, or 74 (variant 1 group) vaccines had high bactericidal titers against this strain (Figure 4, Panel G). As expected, this strain was resistant (titer <1∶10) to the serum from mice immunized with the recombinant fHbp ID 22 vaccine (variant group 2). Against the isogenic mutant of H44/76 with 58% fHbp expression by Western blot as that of the wildtype H44/76 strain, the respective anti-fHbp ID 1, 4, 9, or 74 bactericidal titers were ∼10-fold lower than the corresponding titers measured against the higher fHbp expressing wildtype strain (Panel G). In contrast, none of these antisera was bactericidal against the H44/76 mutant with low fHbp expression (10% of the wildtype isolate). The H44/76 wildtype strain and two fHbp mutants were equally susceptible to bactericidal activity of a positive control mAb against PorA (Panel G, yellow horizontal hatched bars). Similar respective results were observed with the mutants of the serogroup A (A3) and serogroup W-135 isolate (W13). For example, the W-135 wildtype strain with 45% fHbp expression relative to that of the reference strain was susceptible to anti-fHbp bactericidal activity only by the antiserum prepared to the recombinant fHbp vaccine ID 9 that matched that of the target isolate (titer >1∶1000, Panel I). In contrast, the isogenic mutant with increased fHbp expression (126% relative to H44/76) was susceptible to anti-fHbp bactericidal activity by antisera to any of the recombinant fHbp vaccine sequence variants in variant group 1. While the data from the mutants do not permit a precise definition of the level of fHbp expression required for homologous and cross-reactive anti-recombinant fHbp bactericidal activity, collectively the results indicated that fHbp expression below 30% of H44/76 was associated with resistance and, with increasing fHbp expression, the isolates became more susceptible to anti-fHbp cross-reactive bactericidal activity.
Figure 5 shows the anti-fHbp bactericidal titers of antisera prepared to the different recombinant fHbp sequence variants in variant group 1 when measured against wildtype serogroup A, W-135 and X isolates with fHbp in variant group 1. The isolates were generally susceptible to anti-fHbp antibodies elicited by the recombinant fHbp vaccine that matched the sequence variant of the vaccine (blue bars) but not to antibodies elicited by mismatched recombinant fHbp variants. The lack of cross-reactive bactericidal activity was most notable for the anti-fHbp ID 1 and ID 74 antisera (Panels A and D, respectively). This result was surprising since these vaccines had 92 to 96% amino acid identity with the fHbp expressed by the test strains (Table 1). In Figure 5 the order of the isolates from left to right is with increasing fHbp expression (as shown in Figure 2). While there were trends for increased susceptibility to anti-fHbp bactericidal activity with increasing fHbp expression (for example, the W-135 isolates and the respective anti-fHbp ID 9 titers), the relationship is not linear, being confounded by other variables. The lack of linearity is most evident with the serogroup A isolates where some of the lowest fHbp expressers (i.e., A1 and A2) were killed by the anti-fHbp ID 4 antiserum that matched that of the isolates, while some of the highest expressers (i.e., A13 and A14) were resistant.
Figure 6 shows the anti-fHbp bactericidal titers of antisera prepared to different recombinant fHbp sequence variants in variant groups 2 or 3 when measured against wildtype serogroup W-135 isolates with fHbp ID22/23 in variant group 2. The anti-fHbp ID 22 antiserum was bactericidal against all of the isolates, even though nearly all of these isolates were low fHbp expressers. Although these isolates also were killed by the control mismatched anti-fHbp ID 77 (variant 2) or fHbp ID 28 (variant 3) antisera (84 to 94 percent amino acid identity with fHbp ID 22/23, Table 1), the respective titers were 10 to 100-fold lower than those of the anti-fHbp ID 22 antiserum (compare anti-ID 22 bactericidal titers in Panel B to those in Panels C and D, Figure 6).
We measured susceptibility of 12 African isolates to bactericidal activity of an antiserum from mice immunized with a prototype NOMV vaccine prepared from a mutant of group B strain H44/76 with over-expressed fHbp ID 1 (Figure 7). As controls, we tested antisera from mice immunized with an NOMV vaccine from an isogenic fHbp knock-out mutant (NOMV fHbp KO) or a recombinant fHbp ID 1 vaccine. Against the wildtype serogroup B H44/76 strain, which was a high expresser of fHbp ID 1 that matched the recombinant fHbp vaccine, the bactericidal titers of the control antisera to the NOMV vaccine from the fHbp knock-out mutant, or the recombinant fHbp ID 1 vaccine, were both ∼1∶10,240.
Only one of the 12 heterologous African isolates (X4) was susceptible to bactericidal activity of the antiserum to the recombinant fHbp ID 1 vaccine (white bars). In contrast, 11 of the 12 isolates were killed by the antiserum from mice immunized with the NOMV vaccine from the serogroup B mutant with over-expressed fHbp ID 1 (orange bars), which included both serogroup A isolates resistant to bactericidal activity of the antiserum to the recombinant fHbp ID 4 vaccine that matched that of the isolate (gray bars). The one W-135 strain, W1, which was not killed by any of the antisera, was the lowest expresser of fHbp (Figure 3). The remaining three serogroup W-135 tested and all four serogroup X isolates were killed by sera from the mice immunized with the NOMV vaccine from the mutant with over-expressed fHbp ID 1, but the respective bactericidal titers were lower than the corresponding titers elicited by the recombinant fHbp vaccine with a sequence variant that matched that of the strain. With one exception (X5), the bactericidal titers elicited by the control NOMV vaccine from the fHbp knockout mutant were negative (titers <1∶10, gray hatched bars). Although not shown in Figure 7, when we mixed this antiserum with the antiserum to the recombinant fHbp ID 1 vaccine, the serum bactericidal titers remained negative (<1∶10, 11 isolates) or unchanged (titer 1∶50, isolate X5). Thus, there was no evidence of cooperative bactericidal activity between anti-fHbp antibodies elicited by the recombinant fHbp ID 1 vaccine and antibodies elicited to other antigens in the NOMV. These results are in contrast to a previous report of cooperative bactericidal activity observed between human antibodies to recombinant fHbp ID 1 and Neisserial Heparin binding antigen, which individually lacked bactericidal activity [60].
By ELISA, the mice immunized with the NOMV vaccine from the mutant with over-expressed fHbp ID 1 had higher serum anti-fHbp ID 1 antibody concentrations than mice immunized with the recombinant fHbp ID 1 vaccine (respective geometric means of 2203 and 746 U/ml, P<0.02, Figure 8, Panel A). By ELISA, the sera from the mice immunized with the mutant NOMV vaccine also showed greater inhibition of binding of fH to fHbp ID 4, which was the sequence variant expressed by serogroup A strains (Panel B, P<0.05 at each dilution tested). The increased fH inhibition was not only a result of the higher serum anti-fHbp concentrations in the mutant NOMV vaccine group, but also appeared to be from a different anti-fHbp antibody repertoire, since on average the anti-fHbp antibody concentration required for inhibition of fH in this group was nearly 4-fold lower than in the recombinant fHbp vaccine ID 1 group (respective geometric means of 1.17 vs. 4.04 U/ml, P<0.05, Panel C).
In this study, we investigated 106 serogroup A, X and W-135 isolates from Africa to assess the vaccine-potential of fHbp for prevention of meningococcal epidemics. Our results confirm previous observations that epidemic serogroup A, W-135 and X isolates from sub-Saharan Africa are derived from a limited number of clonal complexes and PorA VR sequence variants [59], [61], [62]. With respect to fHbp, 95 percent of the isolates overall had genes encoding fHbp from variant groups 1 or 2. When we considered fHbp sequence variants that differed by a single amino acid to be a single sequence variant, 81 percent of the had one of four prevalent fHbp amino acid sequence variants: three in variant 1 group and one in variant 2 group.
All of the serogroup A strains obtained over a 45 year period expressed a nearly invariant fHbp ID 4/5 in variant group 1 group. In a recent report, fHbp ID 4/5 (referred to in the article as B16 and B22, respectively) also were prevalent among serogroup A isolates from South Africa, which were from clonal complexes ST-1 and ST-5 [26]. In the present study, two of the three serogroup X isolates from South Africa from the 1970s also had fHbp ID 4 (HF24 and HF78, Table S1) [26]. Further, an identical fHbp variant was prevalent among recent serogroup B isolates from the United Kingdom [63], and serogroup B and Y isolates from South Africa [26]. Collectively, these results illustrate remarkable stability of this fHbp sequence over time. At the opposite extreme with respect to fHbp variability were the serogroup W-135 isolates in our study, which were clonal with respect to sequence type (ST-11) and PorA (P1.5,2), but expressed fHbp sequence variants from variant group 1 (ID 9), 2 (ID 22/23) or 3 (ID 349). Two of these fHbp sequence variants, ID 9 (variant group 1) and ID 22 (variant group 2) also were prevalent among the recent ST-11 W-135 isolates from South Africa (referred to as B45 and A10, respectively) [26]). Therefore, with certain clones, such as ST-11, there is the potential for recombination at the fHbp gene locus, which can result in rapid changes in the fHbp antigenic variant group.
Several previous studies reported broad serum bactericidal activity against serogroup B strains by vaccination with recombinant fHbp antigens from the respective variant group [64] or sub-family [30]. In the present study, however, we found limited cross-reactive serum bactericidal activity in mice immunized with recombinant fHbp vaccines when measured against serogroup A, W-135 or X isolates that did not match the amino acid sequence of vaccine antigen. For the serogroup A and W-135 isolates, low fHbp expression appeared to contribute to resistance to anti-fHbp bactericidal activity. When fHbp is sparsely-exposed on the bacterial surface, the ability of two IgG anti-fHbp antibodies to bind to appropriately space epitopes, engage C1q and activate classical complement pathway bacteriolysis may be limited [45]. Our data from isogenic mutant strains with different levels of fHbp expression directly demonstrated that by increasing fHbp expression, a resistant wildtype serogroup A or serogroup W-135 isolate became more susceptible to anti-fHbp bactericidal activity (Figure 4). Conversely, by decreasing fHbp expression, a susceptible wildtype serogroup B strain became resistant. Among the serogroup X wildtype isolates, however, low expression was not a factor for resistance to bactericidal activity since with one exception these isolates were high fHbp expressers. Low expression also did not appear to explain resistance of two of the wildtype serogroup A isolates resistant to bactericidal activity by antibodies to all of the recombinant fHbp sequence variant vaccines, including the fHbp ID 4 vaccine that matched fHbp in the isolates. These two isolates were killed by control antibodies to the serogroup A capsule and PorA, and by antibodies elicited in mice by the NOMV vaccine from the mutant with increased expression of fHbp ID 1. Further studies are needed to define the basis for resistance of these isolates to bactericidal activity by serum antibodies to the recombinant fHbp vaccines.
The lack of broad cross-reactive bactericidal antibody activity to the African isolates in sera from mice immunized with different recombinant fHbp sequence variants was consistent with recent data from two clinical trials in the UK showing limited breadth of serum bactericidal responses after immunization of infants and toddlers with a multicomponent vaccine containing recombinant fHbp ID 1 [42], [43]. In these trials, broader bactericidal antibody responses were observed when the recombinant proteins were combined with a detergent-treated outer membrane vesicle vaccine, which elicited protective antibodies against PorA and which also had an adjuvant effect that augmented the serum antibody responses to the recombinant proteins.
In previous studies, NOMV vaccines from mutants with over-expressed fHbp elicited broad serum bactericidal antibody responses in mice against genetically diverse serogroup B strains [35], [36], [51], [54]. Broad serum bactericidal responses against serogroup B strains also were recently described in an infant non-human primate model [33]. In the present study, mice immunized with an NOMV vaccine from a group B mutant with over-expressed fHbp ID 1 developed serum bactericidal activity against all but one of the isolates tested from Africa. In contrast, all but one of the isolates were resistant to the antiserum to the recombinant fHbp ID 1 vaccine. For the NOMV vaccines, each mouse received 1.25 µg of total protein, which was adsorbed with 170 µg of aluminum hydroxide. For the recombinant protein vaccines, the mice were immunized with 20 µg of recombinant protein mixed with Freund's complete adjuvant for the first dose and Freund's incomplete adjuvant for doses 2 and 3. If anything, the broad serum bactericidal titers elicited by the lower dose of NOMV vaccine given with the aluminum adjuvant, which is suitable for use in humans, underscores the greater vaccine potential of this approach to elicit broad protective immunity. At least three factors appeared to contribute to the enhanced serum bactericidal antibody responses to the NOMV vaccine with increased fHbp expression. First, was over-expression of fHbp, which in a recent study was shown to be required for broad serum anti-fHbp bactericidal responses [34]. Second, was the presence of natural adjuvants in the NOMV such as lipooligosaccharide or PorB [65]. Third, was the possible effect of fHbp antigen presentation in the NOMV on anti-fHbp antibody repertoire as evidenced by greater ability of the anti-fHbp antibodies to inhibit of binding of fH to fHbp than anti-fHbp antibodies elicited by the recombinant fHbp ID 1 vaccine (Figure 8). Greater inhibition of binding of fH to the surface of N. meningitidis also would be expected to decrease down-regulation of complement activation by fH, and enhance susceptibility of the organism to bactericidal activity [37], [38].
In summary, the present data from studies in mice immunized with a prototype NOMV vaccine with increased fHbp expression suggest that this vaccine approach could supplement coverage conferred by the serogroup A polysaccharide conjugate vaccine recently introduced in Africa, and extend coverage against strains with other serogroups. An important question, however, is whether a mutant NOMV vaccine that requires several doses for protection is likely to be practical in a resource poor region such as Africa as compared with conjugate vaccines against serogoups X and W-135. In older children and adults, a single dose of a meningococcal polysaccharide conjugate vaccine can elicit protective serum antibodies. Protection, however, elicited by conjugate vaccines is serogroup specific whereas NOMV vaccines with over-expressed fHbp elicit protective antibodies against strains with different serogroup capsules. Also, in infants immunized with a conjugate vaccine, more than one primary dose usually is necessary for protection [66], and protective serum antibodies last less than a year or two [67], [68]. Thus, periodic boosting with a conjugate vaccine is required to maintain immunity [68], [69]. Defining an optimal NOMV vaccine schedule will require studies in humans. That multiple NOMV doses may be needed to elicit protective antibodies, however, is not necessarily different from the requirements for eliciting and maintaining protection by meningococcal conjugate vaccines in infants and children.
There are several important limitations to the present study. First, we investigated only 106 case isolates. Although these isolates were from 17 countries, 30 percent were from Burkina Faso. Africa is a large and diverse continent with a complex ecology affecting meningococcal transmission and disease. Development of a mutant NOMV fHbp-based vaccine for sub-Saharan Africa will require ongoing surveillance of meningococcal strains to assure that the vaccine antigens match those of the prevalent strains. Second, for investigation of anti-fHbp antibody activity elicited by the recombinant fHbp vaccines, we prepared hyperimmune antiserum pools in mice immunized with the vaccines given with Freunds complete and incomplete adjuvant. This adjuvant is unsuitable for humans and the high anti-fHbp titers in the hyperimmune mouse serum pools are unlikely to be achieved in humans. The poor cross-reactive bactericidal activity of these mouse antisera is, therefore, likely to be even lower in humans immunized with recombinant vaccines given with aluminum adjuvants. The broad serum bactericidal antibody responses to the mutant fHbp NOMV vaccine, however, were in mice given the vaccine with aluminum hydroxide.
A third limitation was that while the data from the isogenic mutants with different levels of fHbp showed a clear relationship between fHbp expression levels and susceptibility to anti-fHbp bactericidal activity, the relatively small number of wild type strains tested, and the presence of potential confounders such as capsular serogroup [70], LOS [71], [72] or alternative fH binding ligands [73], did not provide sufficient statistical power for a formal analysis of the relationship between fHbp expression and anti-fHbp bactericidal activity.
A fourth limitation of the present study is that we used a prototype NOMV vaccine from a mutant group B strain with over-expressed fHbp ID 1. We chose this vaccine since it had worked well in previous studies of group B strains, and an NOMV vaccine from a similar mutant African strain was not yet available. Neither the PorA VR type of the group B vaccine strain nor the fHbp sequence variant was present among the African isolates. We would anticipate that even higher serum bactericidal antibody responses would be elicited by NOMV vaccines prepared from mutant African isolates where the antigens in the vaccine would be matches to the Africa strains. Finally, although the serum bactericidal titers of the control mice immunized with the NOMV from the fHbp knockout were negative, we had insufficient sera from mice immunized with the NOMV vaccine from the mutant with over-expressed fHbp to prove that the bactericidal antibodies were directed against fHbp. In several previous studies however, we demonstrated that bactericidal activity against serogroup B strains was greatly or completely diminished after depletion of anti-fHbp antibodies by solid phase adsorption [34], [36], [51].
|
10.1371/journal.ppat.1002648 | The Transcription Factor AmrZ Utilizes Multiple DNA Binding Modes to Recognize Activator and Repressor Sequences of Pseudomonas aeruginosa Virulence Genes | AmrZ, a member of the Ribbon-Helix-Helix family of DNA binding proteins, functions as both a transcriptional activator and repressor of multiple genes encoding Pseudomonas aeruginosa virulence factors. The expression of these virulence factors leads to chronic and sustained infections associated with worsening prognosis. In this study, we present the X-ray crystal structure of AmrZ in complex with DNA containing the repressor site, amrZ1. Binding of AmrZ to this site leads to auto-repression. AmrZ binds this DNA sequence as a dimer-of-dimers, and makes specific base contacts to two half sites, separated by a five base pair linker region. Analysis of the linker region shows a narrowing of the minor groove, causing significant distortions. AmrZ binding assays utilizing sequences containing variations in this linker region reveals that secondary structure of the DNA, conferred by the sequence of this region, is an important determinant in binding affinity. The results from these experiments allow for the creation of a model where both intrinsic structure of the DNA and specific nucleotide recognition are absolutely necessary for binding of the protein. We also examined AmrZ binding to the algD promoter, which results in activation of the alginate exopolysaccharide biosynthetic operon, and found the protein utilizes different interactions with this site. Finally, we tested the in vivo effects of this differential binding by switching the AmrZ binding site at algD, where it acts as an activator, for a repressor binding sequence and show that differences in binding alone do not affect transcriptional regulation.
| The bacterium Pseudomonas aeruginosa causes a variety of human infections and is the leading cause of death in patients with cystic fibrosis. The main reason for the severity of these infections arises from the ability of P. aeruginosa to express virulence factors that protect it from the host immune system. Several of these processes are controlled by a transcription factor called AmrZ, a potential target for anti-microbial therapeutics. AmrZ is unusual in that it has the ability to both activate some genes, such as for alginate biofilm, and repress others, as with flagellum and itself. Here we determine the three dimensional structure of AmrZ bound to DNA containing a repressor sequence. Our structure shows the specific interactions the protein makes with the DNA for binding and repression. It also reveals that both the sequence and shape of the DNA are important for tight association. We next examined the binding of the protein to DNA containing an activator sequence and found that it has different interactions. However, by switching the AmrZ binding site at algD, where it acts as an activator, for a repressor binding sequence in P. aeruginosa, we show that differences in binding alone do not account for transcriptional regulation.
| Pseudomonas aeruginosa is an opportunistic, Gram negative bacterium that causes a variety of infections, mainly in immune-challenged patients [1]–[3]. More notably, chronic lung infection by P. aeruginosa is the leading cause of death in patients with the autosomal recessive disorder cystic fibrosis (CF) [4]. The underlying cause of the severity of these infections is due in part to the arsenal of virulence factors P. aeruginosa has at its disposal, including type III secretion systems, production of biofilms, phospholipase, exotoxin A, motility, and lipopolysaccharide. In alginate producing strains isolated from CF patients, the transcription factor AmrZ (Alginate and Motility Regulator Z, formerly AlgZ) is highly expressed [5]. Our previous work has shown AmrZ functions as both a transcriptional activator and repressor of several virulence factors. AmrZ is necessary for alginate production, via the activation of algD, which is the first gene in the alginate biosynthetic operon [6]. Reciprocal to this, AmrZ represses fleQ, which encodes an activator of flagellum expression [7]. AmrZ is also required for the regulation of genes responsible for type IV pili localization and twitching motility, through the interaction with a currently unknown gene target [8]. Finally, AmrZ also represses its own transcription by binding to two sites on the amrZ promoter, amrZ1 and amrZ2 [9].
The 108 amino acid, 12.3 kD AmrZ protein is a member of the ribbon-helix-helix (RHH) family of DNA binding proteins, sharing highest sequence similarity to the Arc and Mnt repressors from bacteriophage P22 [10]. Sequence analyses predict that there are over 2300 proteins containing RHH domains found in bacteria, Archaea, and bacteriophages; however, less than twenty of these proteins have been studied with structural or biochemical techniques [11]. Structural information from RHH proteins both in the presence [12]–[19] and absence [20]–[27] of operator DNA, show that they exist as dimers, formed by a hydrophobic core created by the two α-helices. The majority of RHH proteins are transcriptional repressors. AmrZ and Helicobacter pylori NikR are currently the only characterized RHH proteins known to function as both transcriptional activators and repressors [23]. DNA binding by RHH proteins occurs by the insertion of the anti-parallel β-sheet formed by one β-strand from each monomer into the major groove of DNA. The interactions between the protein and the recognition site are very specific, and mutations to either the DNA binding β-sheet, or the operator site often have a negative effect on DNA binding [27], [28]. In addition to binding DNA as a dimer, RHH proteins also assemble as tetramers, which are stabilized by other domains in the protein, such as occurs with the C-terminal domain of Mnt [29].
Information from sequence alignments and structural predictions define three regions of the AmrZ protein, an extended N-terminus spanning residues 1–16, the RHH domain, located from residues 13–66, and a C-terminal domain from residues 67–108 [30]. Both the extended N-terminus and the C-terminal domain do not share any sequence similarity to other proteins, and their exact function has remained an open question. The extended N-terminus has been hypothesized to play a role in DNA binding, and is conserved in other AmrZ orthologs of P. putida and P. syringae [31]. Extended N-termini of other RHH proteins have been examined, although their functions vary between cofactor binding, oligomerization and protein-protein interactions, ATP hydrolysis, in addition to having roles in DNA recognition. The C-terminal domain of AmrZ is proposed to be involved in protein oligomerization, which is supported by glutaraldehyde cross linking assays that show AmrZ forms oligomeric species consistent with the molecular weight of dimers and tetramers in solution [30].
Of the genes that are regulated by AmrZ, the specific locations of the binding sites are only known for two of them. AmrZ functions as a transcriptional activator at the algD promoter and binds 282 base pairs upstream of the transcriptional start site. Additionally, AmrZ acts as a transcriptional repressor of its own gene and recognizes two sites on the amrZ promoter (amrZ1 and amrZ2) at positions −93 and −161. Interestingly, DNA foot-printing has been performed at each of these three sites, and little sequence consensus is shared among them [6], [9]. Both the algD operon and amrZ are under the control of the alternative sigma factor AlgT (AlgU/σ22) [32]. Expression of the algD operon requires additional factors, including the response regulators AlgB and AlgR, the nucleosome proteins IHF and AlgP, and the AlgQ protein, each being necessary, but not sufficient to activate transcription on their own [5]. To date, only the AmrZ and AlgT proteins are known to interact with the amrZ promoter region.
Open questions have remained as to the exact strategies employed by AmrZ to function as both a transcriptional activator and repressor. It is unclear what specific interactions the protein makes with both activator and repressor sequences within DNA in order to carry out these functions. To answer these questions, we determined the crystal structure of an AmrZ C-terminal truncation mutant, Δ42 AmrZ, in complex with an 18 bp oligonucleotide containing the amrZ1-binding site. This structure defines the specific recognition site as two half sites separated by a linker region, and provides evidence that the extended N-terminus of AmrZ interacts with the DNA in a sequence independent manner. Site directed mutagenesis experiments of the amrZ1 DNA reveal that recognition is not only based on the direct readout of the nucleotide sequences, but also relies on recognition of the intrinsic shape of the DNA. These data allow for the creation of a model for transcriptional repression by AmrZ, where a combination of specific base recognition at two half sites and recognition of intrinsic DNA structure allow for binding. We also examined the interaction of AmrZ with the algD-binding site, where AmrZ binding functions as an activator of alginate biosynthesis. The results from these assays only identify one AmrZ binding site on algD and also suggest that the protein may utilize an additional residue in DNA binding. Finally, we demonstrate that while there are different protein interactions at the activator and repressor sequences, these differences alone do not account for the activator and repressor activity of AmrZ.
We determined the structure of a C-terminal truncation mutant of AmrZ, Δ42, in complex with an 18 bp oligonucleotide containing the amrZ1 site to 3.1 Å resolution (Table 1 and Figure 1A). The Δ42 variant of AmrZ (residues 1–66) contains the extended N-terminus and the RHH DNA binding domain, but has a truncated C-terminal domain. Many C-terminal truncation variants of AmrZ were used in crystallization experiments, and Δ42 was the only AmrZ construct tested that crystallized either in the presence or absence of DNA. The Δ42 AmrZ protein was tested for DNA binding affinity, and compared with the wild type protein, no reduction in affinity to any of the three known AmrZ binding sequences (amrZ1/amrZ2/algD) was observed (data not shown). The structure reveals AmrZ binds the amrZ1 site as a dimer-of-dimers, and DNA recognition occurs by the interaction with two half sites on the DNA, separated by five base pairs. There are no major structural differences between each AmrZ dimer (Cα RMSD = 0.381 Å) (Figure 1B). The dimer-dimer interface, which occludes approximately 290 Å2 of surface area on each dimer, is formed by a series of interactions between specific residues located on the loop connecting α-helix 1 and α-helix 2 on chains B and C (Figure 1C). The interactions in this region are symmetric, with the backbone carbonyl of His38 of one protomer forming a hydrogen bond to Arg40 of the opposing protomer, and the side chain of His39 forming a salt bridge to Glu51, also across the interface. The relatively small interface between each dimer, in combination with evidence that AmrZ forms higher order oligomers in solution [30], suggests there are likely additional dimer-dimer interactions mediated by the C-terminal domain of the protein.
The interface between AmrZ monomers to form a dimer is primarily composed of α-helix 1 and α-helix 2 of each monomer that come together to form a hydrophobic core. The dimer interface is quite extensive, composed of 25 residues (Figure 1D, underlined residues) and buries approximately 1600 Å2 of each monomer. Each AmrZ dimer interacts with the amrZ1 binding site through sequence dependent interactions (see below), mediated by the insertion of the anti-parallel β-sheet, formed by dimerization, into the major groove of DNA. Additionally, a number of sequence independent interactions to the phosphate backbone are formed, further supporting the protein-DNA complex (Figure 1E). The protein - DNA interactions exclude a total surface area of 1469 Å2 and are symmetric on both halves of the DNA. The one exception to this is the α-helix 2 N-terminus of chain D, which is not positioned to interact with the phosphate backbone; however, this is most likely due to the lack of a 5′ phosphate group on the nucleotide A1, an artifact of chemical DNA synthesis.
The structure allows us to determine the specific nucleotide sequence recognized by AmrZ, as well as other factors that contribute to DNA recognition. The insertion of the anti-parallel β-sheet from each AmrZ dimer into the major groove of the amrZ1 site provides for the recognition of two half sites, each with the sequence 5′-GGC (Figure 1E, orange bases). Sequence dependent binding by AmrZ occurs via the interaction of three residues, Lys18, Val20, and Arg22, with the nucleotides. Lys18 from one AmrZ monomer is positioned where it can form hydrogen bonding interactions to the O6 and N7 atoms of the two guanine nucleotides, G23 and G24 (G4 and G5 on the other half site) (Figure 2A). The DNA binding β-sheet also orients the residue Arg22, from the other monomer of the dimer, to form a bidentate hydrogen bond to both the O6 and N7 atoms of the nucleotide G12 (G31 on the other half site). This is on the opposite strand of the two bases with which Lys18 interacts. Bidentate hydrogen bonding, specifically between arginine residues and guanine nucleotides, are a major determinant in the selectivity of DNA bases [33]. Another relevant residue located on the DNA binding β-sheet is Val20. Interestingly, among other RHH proteins, this position in the DNA binding β-sheet is generally conserved as a neutral hydrophilic residue. One other exception to this is the Neisseria gonorrhoeae FitAB protein in which there is also a valine at this position [18]. The structure of FitAB in complex with DNA shows the valine forms a van der Waals interaction with the C5 methyl group of a thymine base. In the AmrZ structure, it appears that residue Val20 is poised to select for cytosine bases C13 and C14 (C32 and C33 on other half site) via a hydrophobic interaction (Figure 2B). Purine nucleotides would not be favorable in these locations since the N7 atom of the purine base would interfere with the hydrophobic pocket that is formed by Val20, while a thymine nucleotide in this position would sterically clash with the isopropyl side chain of the valine residue.
To confirm the requirement for each half site in amrZ1 recognition by AmrZ, a series of mutations were created to the amrZ1 DNA binding site. Both 1 and 2 nucleotides in each amrZ1 half site were mutated, and the affinity of WT AmrZ to each of these mutant sequences was measured using fluorescence anisotropy (Table 2). Mutating one nucleotide in each 5′-GGC AmrZ recognition half site to 5′-GTC caused a 9.8 fold reduction in affinity, while mutating two of the nucleotides in each half site to 5′-TTC caused a 12.9 fold reduction in affinity compared to binding to the native amrZ1 sequence. These results confirm the observations from the structure that sequence dependent recognition occurs through the interactions with two half sites, each with the sequence 5′-GGC.
We previously evaluated the contribution of Lys18 and Arg22 to AmrZ activity using in vitro DNA binding assays at amrZ1 and transcriptional reporter assays to measure amrZ repression [30]. The mutation of Lys18 to an alanine (K18A) resulted in a drastic reduction in the DNA binding activity, causing a 274-fold increase in the dissociation constant (Kd), compared to WT AmrZ. When amrZ was replaced in the P. aeruginosa chromosome with a gene encoding K18A AmrZ, amrZ derepression was observed, which was similar in magnitude to that observed in strains harboring null amrZ alleles. Similar results were obtained for the R22A mutant of AmrZ, which had a 44-fold increase in Kd compared to WT AmrZ in vitro, and comparable effects of amrZ transcription in vivo. When the effects of mutating the valine at position 20 to an alanine (V20A) were tested in vitro, a 10-fold increase in Kd was observed. Even with a smaller reduction in DNA binding ability compared to the K18A and R22A mutants, V20A AmrZ was unable to repress amrZ transcription in vivo.
We observe electron density for the extended N-terminus starting at residue 10 on chains A and C. This density is only observed on the side of the AmrZ dimer that makes the specific contacts to the amrZ1 binding site. The lack of electron density of the extended N-terminus on the side of the AmrZ dimer that does not contact the DNA suggests that the N-terminus is disordered in solution, and becomes structured upon DNA binding. Residues 10–17 of the N-terminus form a looped structure, allowing the amino acids Ser13 and Arg14 in the major groove to interact with the DNA (Figure 3). This looped structure is supported by the residue Tyr11, which forms a hydrogen bond to the backbone of Glu25 from the other monomer in the dimer, and by the head-on orientation of the carboxyl side chain of Glu25 perpendicular to the aromatic ring of Tyr11. The side chain of Ser13 forms a hydrogen bond to a phosphate in the DNA backbone, and also positions the residue Arg14 into the major groove of the DNA; however, no contacts between Arg14 and the DNA bases are observed in the structure. This is consistent with previous studies of an AmrZ R14A mutant, which has no change in binding affinity for the amrZ1 DNA when compared to WT protein in vitro, as well as no effect on amrZ repression in vivo [30]. Other RHH proteins contain extended N-termini that contribute to DNA binding. The Staphylococcus aureus pSK41 plasmid-encoded ArtA protein has a 16 residue N-terminal domain that is necessary for recognition of at least one of the binding sites of the protein [19]. Additionally, the seven residue extended N-terminus of the Arc repressor is disordered in solution, but adopts a tandem-turn structure upon binding DNA [13], and mutations to the N-terminus result in decreased binding to operator sites [34]. Although mutations to the extended N-terminus in AmrZ do not reduce affinity to the amrZ1 repressor site, AmrZ may act in a manner similar to ArtA, where the extended N-terminus may provide specificity for DNA binding at other sites.
There are a number of interactions between AmrZ and the phosphodiester backbone of the DNA that act to position the DNA binding β-sheet in the major groove. The majority of these sequence independent interactions occur with residues located in the N-terminus of α-helix 2, which points down towards the DNA backbone (Figure 2C). The positioning of this helix allows the formation of hydrogen bonding interactions between the side chains of Ser41 and Ser44, and the backbone amide nitrogens of Met42 and Asn43 to the phosphate groups of the DNA. This interaction is further bolstered by the positive dipole of the N-terminal end of α-helix 2 and the negatively charged phosphate backbone of the DNA. Another sequence independent interaction between AmrZ and the DNA occurs via the side chain of Arg28, from α-helix 1, to the backbone of the DNA. Interestingly, the location of α-helix 2 also allows the side chain of Asn43 to form two hydrogen bonding interactions to the backbone amide nitrogen and carbonyl oxygen of the DNA binding residue, Arg22, in the opposite monomer. This also helps position Arg22 for interaction with the DNA bases. Sequence independent interactions formed by the N-terminus of α-helix 2 are one of the main structural features of RHH proteins [11]. These contacts are often observed to anchor the protein onto the DNA; however, in the case of AmrZ, this electrostatic interaction may play an additional role in recognition of the intrinsic shape of the DNA, particularly in the linker region.
Analysis of the amrZ1 DNA in the structure reveals a significant narrowing of the minor groove to 2.8 Å in the A/T rich region between the two amrZ1 half sites (Figure 4A). In addition to the narrow minor groove, there is an increase in the width of the major grooves where AmrZ interacts with each half site, most likely to accommodate the width of the anti-parallel β-sheet in this region (Figure 4B). A-tract DNA, as in the amrZ1 site, has specific properties in that each ApA base pair step exhibits a negative roll, and bifurcated hydrogen bonds between each adenine and two thymine nucleotides on the opposite strand lead to propeller twisting and minor groove narrowing; A-tract DNA is also thought to be less flexible due to the extra stabilization provided by the additional bifurcated hydrogen bonds [35]. Based on this we investigated the region between the two amrZ1 half sites for any role in AmrZ binding, and whether the binding of AmrZ causes distortions to the amrZ1 DNA, or if the amrZ1 DNA is intrinsically distorted, allowing for AmrZ recognition.
To test if the linker region between each AmrZ binding half site contributes to AmrZ affinity at amrZ1, the native A/T rich linker sequence 5′-AAAAC was mutated to a G/C rich linker region with the sequence 5′-CGCGC, which resulted in a 7.5 fold reduction in binding (Table 2). It is important to note that this reduction in binding is not caused by the removal of specific protein - nucleotide interactions, since there are no contacts between the AmrZ protein and amrZ1 DNA in this region. Combining the mutations in the AmrZ binding half site with the mutations to the linker region (TTC/GC amrZ1) caused a severe aberration in binding affinity (244-fold reduction). The results show that binding affinity is regulated by both the sequence dependent interactions between AmrZ and amrZ1 and the linker region separating these binding sites.
Additional binding experiments were performed to determine if the intrinsic structure of the A/T rich linker contributes to binding affinity. The five base pair linker region on the native amrZ1 binding site was mutated to three sequences, each having their own unique properties. A sequence with the linker region mutated to 5′-TTTTC resulted in a 5.0-fold reduction in AmrZ binding, when compared to the WT amrZ1 sequence (Table 2). TpT base pair steps have the same properties of ApA base pair steps, including a narrow minor groove and less flexibility [35]. Mutating the amrZ1 sequence to 5′-AATTC caused a 7.0-fold reduction in affinity when compared to AmrZ binding to the WT amrZ1 sequence (Table 2). Molecular dynamics simulations of the interactions between the papillomavirus E2 transcription factors and their binding sites have shown that the 4 nucleotide sequence AATT has similar minor groove and propeller twist properties to A-tract DNA [36]. The last amrZ1 mutant binding site tested had a linker region containing the sequence 5′-ATATC, and AmrZ binding to this site was also altered compared to the WT amrZ1 sequence, causing a 4.7-fold reduction in affinity (Table 2). This site was designed to test if flexibility in the linker region allowed AmrZ to distort the DNA and form a complex. The TpA step in this sequence permits variations in roll, twist and slide due to poor stacking between these base pairs, and DNA containing these steps contain wider minor grooves, caused by the steric clashing of cross strand adenines [37]. It should be noted that the properties described for these sequences are average parameters derived from structures and that individual structures show a range of properties, specifically minor groove width [38]. Although AmrZ had reduced affinity for each of these three sequences, the most dramatic effect was mutation of the linker region to 5′-CGCGC. Binding of AmrZ to the sequence 5′-ATATC was reduced suggesting that A/T content of the sequence, which is usually thought to impart flexibility to DNA, was not the main contributor to AmrZ specificity. AmrZ binding to the two sequences harboring mutant linker regions with similar properties to the A-tract sequence (5′-TTTTC and 5′-AATTC) was decreased, suggesting that there are properties unique to the 5′-AAAAC linker sequence in the native amrZ1 binding site that allow for binding specificity. Taken together, these data allow us to propose that binding specificity is directed by intrinsic distortions to the DNA, rather than the flexibility conferred by the A/T rich sequence composition. Recognizing a physical feature of the DNA rather than a specific sequence introduces degeneracy in the recognition sequence that would influence the number of potential recognition sites for AmrZ. We queried the P. aeruginosa PAO1 genome [31] for the number of binding sites with the exact amrZ1 repressor sequence and found 5 sites. If we allow the A/T linker region to be degenerate, the number of potential binding sites increases to 77. Further biological studies will be required to determine how many of these sites function as actual regulators.
Narrowed minor grooves of DNA have a strong correlation between the width and increased electronegative potential of the minor groove [38]. There are many examples of transcription factors that recognize local distortions of the minor groove in addition to sequence specific recognition in both prokaryotic and eukaryotic organisms. The Listeria monocytogenes helix-turn-helix (HTH) transcription factor MogR recognizes two half sites on the flaA operator site [39]. The minor groove between the two half sites is distorted, and contributes to MogR specificity for this site. Another example is the myocyte enhancer factor-2 (MEF2), a member of the MADS-box superfamily, which recognizes a narrowed minor groove on the consensus sequence to bind and activate transcription [40]. These two examples, in addition to others, use positively charged residues, specifically arginine, to recognize and form contacts with the enhanced electronegative potential of the narrow minor groove [38]. However, there are examples of proteins similar to AmrZ that recognize minor groove shape, but do not make any contacts to the minor groove. The classical example is the bacteriophage 434 repressor recognition of six binding sites on the two operator regions, OR and OL, which is greatly modulated by the sequence composition of the central region of these sites [41]. These variations in binding affinities have been shown to be biologically important in directing the lysogenic or lytic fate of bacteriophage 434 [42]. Although the 434 repressor positions an arginine residue near the minor groove, there are no specific contacts by the protein to this region, and mutational analysis shows that this arginine does not contribute to binding affinity [41]. Recently, recognition of the intrinsic structure of narrowed minor grooves has been studied with the DNA bending protein Fis, which is responsible for the compaction of bacterial DNA [43]. An A/T rich (5′-AATTT) narrowed minor groove, located between two Fis binding sites is compressed, allowing for the insertion of two HTH domains into the adjacent major grooves of DNA. Mutations to this narrow minor groove sequence cause changes in binding, with the biggest change occurring by mutating the sequence to a G/C rich sequence (5′-GGCGC). Narrowed minor grooves between binding half sites have been observed in other RHH protein - DNA structures. In the structure of Arc in complex with DNA, the minor groove between half sites is narrowed to 1.2 Å, and the sequence in this region has the sequence 5′-GTGCT [13]. Likewise, in the Streptococcus sp. CopG-DNA structure the minor groove is narrowed to 1.9 Å and has the sequence 5′-TTGAG [14]. The DNA in complex with the inc18 plasmid encoded omega protein has a minor groove width of 2.7 Å, and the A/T rich sequence 5′-AAAT. Also, due to a fortuitous packing arrangement in the crystal, both bound and free DNA were observed, with the free DNA having similar secondary structure as the omega bound DNA [16]. For each of these proteins, the contributions of the linker region between the two half sites to binding have not been determined.
Alterations in the sequence specific half sites, the linker region, or both can modulate affinity for AmrZ to amrZ1; however, the exact mechanism by which AmrZ recognizes the distorted structure of the minor groove remains enigmatic. In the Δ42AmrZ-amrZ1 structure (Figure 1A), there are no positively charged amino acids that contact the minor groove. The extended N-terminus of AmrZ contains an arginine at position 2 which might make these contacts; however, in vitro DNA binding assays performed with various N-terminal truncation mutants of AmrZ showed no decrease in binding to amrZ1 [30]. The phosphate backbone on either side of the narrow minor groove of amrZ1 is contacted on each side by the N-terminus of α-helix 2 from chains A and C (Figure 1A, 2C). This attraction is most likely enhanced due to the positive dipole formed by the N-terminus of the α-helix and the increased electronegative potential of the narrowed minor groove.
In addition to functioning as a repressor when bound to amrZ1, AmrZ binding to the algD site is necessary for the activation of genes responsible for alginate biosynthesis. Interestingly, there are significant divergences between the activator and repressor sequences, and AmrZ affinity to the algD binding site is approximately 24 fold reduced compared to the amrZ1 binding site (Tables 2 & 3). Using the information from the AmrZ interaction with the repressor amrZ1 binding site, we asked if we could predict how AmrZ interacts with the binding site on the algD promoter (Figure S1). We set out to determine the features of the algD sequence necessary for AmrZ recognition and activation. By aligning the left half AmrZ binding site on amrZ1 (5′-GGC) to the algD sequence (positions 5–7), it became apparent that there is no similar right half binding site on algD, and additionally, the sequence of the linker region is also different (Figure 5A). In order to probe the interaction between AmrZ and algD, multiple single nucleotide mutations of the algD site were created, and binding affinity of AmrZ to each of the mutant algD sequences was measured with fluorescence anisotropy.
Through mutagenesis of nucleotides in the proposed left half binding site in algD, we show that AmrZ recognizes the sequence 5′-GGC at this site. The guanine nucleotides at positions 5 and 6 on one strand of the algD binding site and positions 7 and 8 on the other strand (Figure 5A) were mutated to thymine bases, and the binding affinity of AmrZ to each of these mutant sequences was measured (Table 3, Figure 5B). The mutation to position 5 resulted in a slight increase to AmrZ affinity, while mutations to positions 6, 7, and 8 each resulted in significant reductions in affinity. To determine the nucleotides AmrZ interacts with on the right half of the algD binding site, guanine bases at position 16 on one strand, and positions 13, 14, and 17 on the opposite strand of algD were mutated to thymine residues. No significant differences in binding affinity to these sequences are observed (Table 3, Figure 5B), suggesting that AmrZ interacts with this half site in a different manner than what is observed at amrZ1.
Our previous binding experiments show the same residues, Lys18, Val20, and Arg22 are involved in the sequence dependent interactions with algD [30]. In addition, we found that Arg14 is also necessary for binding, with the R14A mutant of AmrZ exhibiting a 5 fold reduction in binding affinity at algD. This arginine residue is also required for transcriptional activation of algD, where R14A AmrZ only retains 3% of WT activity in vivo. From the AmrZ - amrZ1 structure, the extended N-terminus forms a looped structure which positions Arg14 into the major groove of DNA (Figure 3); however, no specific contacts between this residue and the amrZ1 DNA are observed, and mutations of this residue have no effects on in vitro and in vivo activity at the repressor site.
The differences we observe in interactions of AmrZ with the activator algD binding sequence versus the repressor amrZ1 binding sequence led us to ask if these different binding modes alone could account for activation or repression activity. To test this hypothesis we switched the AmrZ binding site in the algD promoter (activator) with the amrZ1 binding site (repressor) and introduced this variant into an algD::lacZ transcriptional fusion, which was stably integrated into the genome of the mucoid P. aeruginosa strain FRD1 (FRD1 palgDamrZ1-lacZ). The position and length of the switched binding site were the same as in the native algD promoter. With this construct we measured relative activation at algD with a β-galactosidase activity assay compared to an algD::lacZ transcriptional fusion containing the wild type algD AmrZ binding site (FRD1 palgD-lacZ). The results of this experiment (Table 4) reveal activation of algD remains unchanged when the amrZ1 binding site replaced the native site. Cell lysates from the FRD1 palgDamrZ1-lacZ strain had 527.7 units of β-galactosidase activity compared to 536.1 units for FRD1 palgD-lacZ. Expression of both palgD-lacZ and palgDamrZ1-lacZ were significantly reduced in amrZ mutant P. aeruginosa strains (Table 4), indicating that the reporter fusion faithfully reproduced what has been observed previously regarding AmrZ activation of algD [5], [10], [30]. The activation of algD with the amrZ1 repressor site at its promoter supports a model in which AmrZ binding alone does not regulate activation or repression of transcription, but rather interactions of AmrZ with other regulators at the amrZ and algD promoters likely contribute to repression or activation, respectively. This is consistent with the previous evidence that the AlgB, AlgR, IHF, and CysB regulators are known to bind on the algD promoter and are necessary for activation [44]–[48], suggesting a possible interaction of AmrZ with one of these proteins. To date, no other regulators have been identified to bind the amrZ promoter; however, it is possible one of these same regulators may also interact with AmrZ there as well. An additional determinant likely dictating activation versus repression is the position of the AmrZ binding site relative to the start of transcription, which differs for algD (−282) and amrZ1 (−93).
AmrZ functions as both a transcriptional activator and repressor of P. aeruginosa virulence genes. We have determined the structure of Δ42 AmrZ in complex with an 18 base pair oligonucleotide containing the amrZ1 binding site. AmrZ binding to this site results in the repression of amrZ transcription. By combining structural and biochemical data, we developed a model for AmrZ recognition at amrZ1. Using the suggested terminology from the recent review by Rohs et al. [49], the protein-DNA specificity of AmrZ can be classified by major groove base readout through protein residues in the β-sheet with two GGC half sites in the DNA. This is combined with local shape readout utilizing minor groove distortions in the linker region between the half sites. We also probed the interaction of AmrZ with another biologically important binding site algD, which leads to the activation of alginate biosynthesis. In contrast, we observed stark differences in the physical interactions that AmrZ makes with the algD sequence that suggest the protein likely utilizes a different mode of recognition at this site. AmrZ binds the algD sequence with lower affinity, and mutagenesis of the algD sequence shows that only one half site contributes to AmrZ binding. However, these differences in protein binding at the promoter sequences are alone not sufficient to account for the activator or repressor activity of AmrZ, and likely the position of AmrZ binding at the promoter and/or protein interactions with other regulators are also necessary for biological function.
The gene encoding WT AmrZ was PCR amplified from the P. aeruginosa strain PAO1 with the primers amrZ_F (5′-CGCCATCACATATGCGCCCACTGAAACAGGC) and amrZ_wt_R (5′-CGCCATCAGGATCCTCAGGCCTGGGCCAGCTC). The resulting gene product was then inserted into a modified pET19 expression vector (Novagen) which encodes an N-terminal poly-Histidine tag, followed by a Rhinovirus 3C protease cleavage site, which permits the removal of the affinity tag (PreScission Protease, GE Healthcare). The pET19-amrZ vector was transformed into E. coli C41(DE3) cells for expression. One liter of LB-Broth (Luria-Bertani) supplemented with 50 µg/ml of ampicillin was inoculated with 10 ml of an overnight culture of the C41 cells containing the pET19-amrZ vector. The cells were grown at 37°C to an OD600 = 0.5, and induced with 1 mM isopropyl β-D-thiogalactopyranoside (IPTG) at 16°C for 20 hours. Prior to induction with IPTG, cells were rapidly cooled on ice to 20°C to bring the temperature of the culture close to the induction temperature. Induction of the cells at low temperature was necessary for protein solubility during overexpression. Cells were harvested by centrifugation, resuspended in lysis buffer (100 mM KH2PO4 pH 7.5, 500 mM NaCl, 10% glycerol, 4 M urea), and lysed using an EmulsiFlex C-5 cell homogenizer (Avestin). Cell debris was removed at 30,000× g and the supernatant was passed over a 10 ml Ni-NTA (Qiagen) column equilibrated with lysis buffer. This column was washed with 20 column volumes of wash buffer 1 (100 mM KH2PO4 pH 7.5, 500 mM NaCl, 10% glycerol, 35 mM imidazole, 3 M urea), followed by 10 column volumes of wash buffer 2 (100 mM KH2PO4 pH 7.5, 500 mM NaCl, 10% glycerol, 50 mM imidazole, 2 M urea). Bound AmrZ was eluted with elution buffer (100 mM KH2PO4 pH 7.5, 500 mM NaCl, 10% glycerol, 500 mM imidazole, 1 M urea), treated with PreScission Protease according to the manufacturer's directions, and dialyzed over night at 4°C against 100 mM Bis-Tris pH 5.5, 100 mM NaCl, 5% glycerol, 2 mM dithiothreitol (DTT), and 0.5 mM EDTA. The partial denaturing conditions introduced by the 4 M urea were necessary for protein solubility and affinity to the Ni-NTA column, and no change in secondary structure or DNA binding affinity was observed compared to protein purified without urea present. AmrZ was then passed over a MonoS cation exchange column, and eluted with a 0.1 M–1 M gradient of NaCl. Purity of the peak fractions was verified by SDS-PAGE, and fractions containing pure WT AmrZ were pooled. For crystallization experiments, AmrZ was dialyzed against 100 mM Bis-Tris pH 5.5, 100 mM NaCl, 2% glycerol, while for DNA binding assays, AmrZ was dialyzed against a buffer containing 100 mM Bis-Tris pH 6.5, 150 mM NaCl, 5% glycerol. WT AmrZ was then concentrated to 20 mg/ml for crystallization experiments, or 1 mg/ml for DNA binding assays, aliquoted, flash frozen in liquid nitrogen, and stored at −80°C. Concentration of WT AmrZ was measured using the BCA assay (Thermo Scientific) using a standard curve of lysozyme as a reference.
The Δ42 C-terminal truncation mutant of AmrZ was amplified from the P. aeruginosa strain PAO1 using the primers amrZ_F and amrZ_Δ42_R (5′-CGCCATCAGGATCCTCAAACACCGAGATTGTCTTG). Expression and purification of this protein was carried out using the procedures outlined for WT AmrZ.
Crystallization trials of AmrZ were carried out by screening multiple AmrZ C-terminal deletion constructs against a library of double stranded DNA oligonucleotides containing the amrZ1 binding site (Integrated DNA Technologies). Initial crystals were obtained only with the Δ42 AmrZ C-terminal truncation and an 18 bp oligonucleotide in a condition containing 6% PEG 8 K, 0.1 M MES pH 6.0, 0.1 M CaCl2, 0.1 M NaCl. For experimental phasing, selenomethionine (Se-Met) derivatized Δ42 AmrZ was prepared using published methods [50]. Purification of this protein was performed using the methods described above, with the only exception being the addition of 5 mM DTT in the final dialysis buffer. Crystals of the Se-Met Δ42 AmrZ - 18 bp amrZ1 complex were obtained by mixing the protein and DNA in a 1∶1.5 molar ratio (810 µM AmrZ: 607.5 µM amrZ1) in the presence of 50 mM MgSO4. This complex was crystallized by the hanging drop vapor diffusion method at 25°C at a 1∶1 ratio with reservoir solution containing 3% PEG8K, 0.1 M MES pH 6.0, 0.15 M NaCl, and 2 mM TCEP pH 8.0. Crystals grew within 2–3 weeks and were soaked in a solution containing 20% 2-methyl-1,3 propanediol for cryo-protection before being frozen in liquid nitrogen for data collection.
Diffraction data for crystals containing the Δ42 AmrZ: 18 bp amrZ1 complex were collected on beamline X25 at the National Synchrotron Light Source (NSLS), Brookhaven National Labs. The dataset was collected at the selenium peak, with an X-ray wavelength of 0.9793 nm. Indexing, integration and scaling of the data were performed using HKL2000 program suite [51]. Phasing of the structure was performed using SAD methods with the program SOLVE [52], and density modification was performed using RESOLVE [53]. Manual model building was performed in Coot [54], and refinement was carried out using the programs REFMAC5 [55] within the CCP4 program suite [56], and CNS [57]. Data collection and refinement statistics are found in Table 1. The atomic coordinates and structure factors have been deposited in the Protein Data Bank under the PDB id 3QOQ.
Binding affinity of the various amrZ1 and algD binding site mutants were performed using fluorescence anisotropy as previously described [30], [58]. In brief, increasing concentrations of WT AmrZ were incubated in a reaction (25 µl) containing 1 nM 22-mer 5′-6-carboxy-fluorescein (6-FAM) labeled DNA oligonucleotide (IDT) containing either the amrZ1 or the algD sequences, 100 nM nonspecific DNA of random sequence, 100 µg/ml bovine serum albumin (BSA), 100 mM Bis-Tris pH 6.5, 150 mM NaCl, and 5% glycerol. DNA concentrations were kept 1 nM (<<Kd) to ensure equilibrium measurements of binding constants. Anisotropy measurements were recorded at 25°C on a Safire2 microplate reader with a fluorescence polarization module (Tecan Group, Ltd.), using an excitation wavelength of 470 nm and an emission wavelength of 525 nm. Anisotropy data were scaled and normalized using Equation 1 below:(1)In this equation, Aobs is the measured anisotropy value for each AmrZ concentration, A0 is the anisotropy of the unbound DNA, and Amax is the maximum anisotropy observed in each experiment. The dissociation constant (Kd) was calculated by fitting the data to the equation for a single state binding model (Equation 2).(2)Fitting of the data to Equation 2 was performed using SigmaPlot. Raw data and fits for AmrZ binding to each DNA sequence can be found in Figures S2 and S3, with the results from these experiments being presented in Tables 2 and 3. Results presented are the averages of four independent experiments.
The program CURVES+ [59] was used to measure the major and minor groove widths of the amrZ1 DNA. The buried surface areas formed by protein-DNA interactions and by protein-protein interactions were measured using the programs AREAIMOL in the CCP4 suite [56] and PDBsum [60], respectively. Ramachandran statistics found in Table 1 were calculated with PDBsum [60]. Electrostatic surface representations of the protein and DNA were created by first generating a PQR file, which contains charge and radius information for each atom, with the program PDB2PQR [61], followed by visualization of the electrostatic surface using the APBS program [62]. All figures of structural representations were prepared using the program PyMol [63].
The AmrZ binding site (ABS) (CCATTGGCCATTACCAGCCTCCC) in the algD promoter was replaced by the same-length amrZ1 ABS (GTACTGGCAAAACGCCGGCACGC) from the amrZ promoter by site-directed mutagenesis [64]. Mutagenesis was achieved by primers algD74 (GCGTGCCGGCGTTTTGCCAGTACATTACGCCGGAGATGGCATTTC) and algD75 (GTACTGGCAAAACGCCGGCACGCGCCATTACATGCAAATTACGATTGC), together with flanking primers algD65 (CCCCAAGCTTCTCTTTCGGCACGCCGAC) and algD66 (CCGGGATCCCCGACATAGCCCAAACCAAAG). PCR products of algD65/algD74 and algD66/algD75 were denatured and hybridized. The products were used as the template for the second PCR, with primers algD65 and algD66. With HindIII and BamHI sticky ends, the final PCR product was cloned into HindIII and BamHI double digested mini-CTX-lacZ transcriptional fusion vector [65], resulting in a new plasmid pBX8, which harbors modified algD::lacZ transcriptional fusion (palgDamrZ1-lacZ). The sequence of the palgDamrZ1-lacZ promoter was verified by PCR and sequencing.
The plasmid pBX8 was transferred into P. aeruginosa FRD1 using E. coli strain SM10. The modified palgDamrZ1-lacZ transcriptional fusion was integrated at the attB site within the chromosome of FRD1 and FRD1 ΔamrZ [30], and the unnecessary portion of the fragment was removed by pFLP2 [66], resulting in FRD1 palgDamrZ1-lacZ or FRD1 ΔamrZ palgDamrZ1-lacZ, respectively.
P. aeruginosa in mid-log phase were pelleted and washed with Z-buffer (110 mM Na2HPO4, 45 mM NaH2PO4, 10 mM KCl, 2 mM MgSO4, pH7.0). Cells were lysed through three rounds of fast freezing at −80°C then thawing at 37°C, followed by mild sonication. Samples were centrifuged at 18 k× g and 4°C for 10 min at 21,000× g. The supernatants were analyzed for β-galactosidase activity by mixing 10 µl of a sample supernatant with 80 µl of Z-buffer. To start the reaction 20 µl of 4 mg/ml orthonitrophenol was added. The color change in the reaction was monitored with time and reactions were stopped by addition of 40 µl 1 M Na2CO3 for reading. Wild type FRD1 palgD::lacZ transcriptional fusion was the positive control, and FRD1 lacZ with no promoter acted as the negative control. The absorbance at both 420 nm and 550 nm of each reaction solution was read in a Molecular Devices M5 microplate reader. Miller Units were calculated from different strains as outlined [67].
|
10.1371/journal.pmed.1002699 | Deep-learning-assisted diagnosis for knee magnetic resonance imaging: Development and retrospective validation of MRNet | Magnetic resonance imaging (MRI) of the knee is the preferred method for diagnosing knee injuries. However, interpretation of knee MRI is time-intensive and subject to diagnostic error and variability. An automated system for interpreting knee MRI could prioritize high-risk patients and assist clinicians in making diagnoses. Deep learning methods, in being able to automatically learn layers of features, are well suited for modeling the complex relationships between medical images and their interpretations. In this study we developed a deep learning model for detecting general abnormalities and specific diagnoses (anterior cruciate ligament [ACL] tears and meniscal tears) on knee MRI exams. We then measured the effect of providing the model’s predictions to clinical experts during interpretation.
Our dataset consisted of 1,370 knee MRI exams performed at Stanford University Medical Center between January 1, 2001, and December 31, 2012 (mean age 38.0 years; 569 [41.5%] female patients). The majority vote of 3 musculoskeletal radiologists established reference standard labels on an internal validation set of 120 exams. We developed MRNet, a convolutional neural network for classifying MRI series and combined predictions from 3 series per exam using logistic regression. In detecting abnormalities, ACL tears, and meniscal tears, this model achieved area under the receiver operating characteristic curve (AUC) values of 0.937 (95% CI 0.895, 0.980), 0.965 (95% CI 0.938, 0.993), and 0.847 (95% CI 0.780, 0.914), respectively, on the internal validation set. We also obtained a public dataset of 917 exams with sagittal T1-weighted series and labels for ACL injury from Clinical Hospital Centre Rijeka, Croatia. On the external validation set of 183 exams, the MRNet trained on Stanford sagittal T2-weighted series achieved an AUC of 0.824 (95% CI 0.757, 0.892) in the detection of ACL injuries with no additional training, while an MRNet trained on the rest of the external data achieved an AUC of 0.911 (95% CI 0.864, 0.958). We additionally measured the specificity, sensitivity, and accuracy of 9 clinical experts (7 board-certified general radiologists and 2 orthopedic surgeons) on the internal validation set both with and without model assistance. Using a 2-sided Pearson’s chi-squared test with adjustment for multiple comparisons, we found no significant differences between the performance of the model and that of unassisted general radiologists in detecting abnormalities. General radiologists achieved significantly higher sensitivity in detecting ACL tears (p-value = 0.002; q-value = 0.019) and significantly higher specificity in detecting meniscal tears (p-value = 0.003; q-value = 0.019). Using a 1-tailed t test on the change in performance metrics, we found that providing model predictions significantly increased clinical experts’ specificity in identifying ACL tears (p-value < 0.001; q-value = 0.006). The primary limitations of our study include lack of surgical ground truth and the small size of the panel of clinical experts.
Our deep learning model can rapidly generate accurate clinical pathology classifications of knee MRI exams from both internal and external datasets. Moreover, our results support the assertion that deep learning models can improve the performance of clinical experts during medical imaging interpretation. Further research is needed to validate the model prospectively and to determine its utility in the clinical setting.
| We wanted to see if a deep learning model could succeed in the clinically important task of detecting disorders in knee magnetic resonance imaging (MRI) scans.
We wanted to determine whether a deep learning model could improve the diagnostic accuracy, specificity, or sensitivity of clinical experts, including general radiologists and orthopedic surgeons.
Our deep learning model predicted 3 outcomes for knee MRI exams (anterior cruciate ligament [ACL] tears, meniscal tears, and general abnormalities) in a matter of seconds and with similar performance to that of general radiologists.
We experimented with providing model outputs to general radiologists and orthopedic surgeons during interpretation and observed statistically significant improvement in diagnosis of ACL tears with model assistance.
When externally validated on a dataset from a different institution, the model picked up ACL tears with high discriminative ability.
Deep learning has the potential to provide rapid preliminary results following MRI exams and improve access to quality MRI diagnoses in the absence of specialist radiologists.
Providing clinical experts with predictions from a deep learning model could improve the quality and consistency of MRI interpretation.
| Magnetic resonance imaging (MRI) of the knee is the standard-of-care imaging modality to evaluate knee disorders, and more musculoskeletal (MSK) MRI examinations are performed on the knee than on any other region of the body [1–3]. MRI has repeatedly demonstrated high accuracy for the diagnosis of meniscal and cruciate ligament pathology [4–7] and is routinely used to identify those who would benefit from surgery [8–10]. Furthermore, the negative predictive value of knee MRI is nearly 100%, so MRI serves as a noninvasive method to rule out surgical disorders such as anterior cruciate ligament (ACL) tears [11]. Due to the quantity and detail of images in each knee MRI exam, accurate interpretation of knee MRI is time-intensive and prone to inter- and intra-reviewer variability, even when performed by board-certified MSK radiologists [12]. An automated system for interpreting knee MRI images has a number of potential applications, such as quickly prioritizing high-risk patients in the radiologist workflow and assisting radiologists in making diagnoses [13]. However, the multidimensional and multi-planar properties of MRI have to date limited the applicability of traditional image analysis methods to knee MRI [13,14].
Deep learning approaches, in being able to automatically learn layers of features, are well suited for modeling the complex relationships between medical images and their interpretations [15,16]. Recently, such approaches have outperformed traditional image analysis methods and enabled significant progress in medical imaging tasks, including skin cancer classification [17], diabetic retinopathy detection [18], and lung nodule detection [19]. Prior applications of deep learning to knee MRI have been limited to cartilage segmentation and cartilage lesion detection [20–22].
In this study, we present MRNet, a fully automated deep learning model for interpreting knee MRI, and compare the model’s performance to that of general radiologists. In addition, we evaluate changes in the diagnostic performance of clinical experts when the automated deep learning model predictions are provided during interpretation. Finally, we evaluate our model’s performance on a publicly available external dataset of knee MRI exams labeled for ACL injury.
Reports for knee MRI exams performed at Stanford University Medical Center between January 1, 2001, and December 31, 2012, were manually reviewed in order to curate a dataset of 1,370 knee MRI examinations. The dataset contained 1,104 (80.6%) abnormal exams, with 319 (23.3%) ACL tears and 508 (37.1%) meniscal tears. ACL tears and meniscal tears occurred concurrently in 194 (38.2%) exams. The most common indications for the knee MRI examinations in this study included acute and chronic pain, follow-up or preoperative evaluation, injury/trauma, and other/not provided. Examinations were performed with GE scanners (GE Discovery, GE Healthcare, Waukesha, WI) with standard knee MRI coil and a routine non-contrast knee MRI protocol that included the following sequences: coronal T1 weighted, coronal T2 with fat saturation, sagittal proton density (PD) weighted, sagittal T2 with fat saturation, and axial PD weighted with fat saturation. A total of 775 (56.6%) examinations used a 3.0-T magnetic field; the remaining used a 1.5-T magnetic field. See S1 Table for detailed MRI sequence parameters. For this study, sagittal plane T2-weighted series, coronal plane T1-weighted series, and axial plane PD-weighted series were extracted from each exam for use in the model. The number of images in these series ranged from 17 to 61 (mean 31.48, SD 7.97).
The exams were split into a training set (1,130 exams, 1,088 patients), a tuning set (120 exams, 111 patients), and a validation set (120 exams, 113 patients) (Fig 1). To form the validation and tuning sets, stratified random sampling was used to ensure that at least 50 positive examples of each label (abnormal, ACL tear, and meniscal tear) were present in each set. All exams from each patient were put in the same split. Table 1 contains pathology and patient demographic statistics for each dataset.
We obtained a publicly available dataset from Štajduhar et al. [23] consisting of 917 sagittal PD-weighted exams from a Siemens Avanto 1.5-T scanner at Clinical Hospital Centre Rijeka, Croatia. From radiologist reports, the authors had extracted labels for 3 levels of ACL disease: non-injured (690 exams), partially injured (172 exams), and completely ruptured (55 exams). We split the exams in a 60:20:20 ratio into training, tuning, and validation sets using stratified random sampling. We first applied MRNet without retraining on the external data, then subsequently optimized MRNet using the external training and tuning sets. The classification task was to discriminate between non-injured ACLs and injured ACLs (partially injured or completely torn).
Reference standard labels were obtained on the internal validation set from the majority vote of 3 practicing board-certified MSK radiologists at a large academic practice (years in practice 6–19 years, average 12 years). The MSK radiologists had access to all DICOM series, the original report and clinical history, and follow-up exams during interpretation. All readers participating in the study used a clinical picture archiving and communication system (PACS) environment (GE Centricity) in a diagnostic reading room, and evaluation was performed on the clinical DICOM images presented on an at least 3-megapixel medical-grade display with a minimum luminance of 1 cd/m2, maximum luminance of 400 cd/m2, pixel size of 0.2, and native resolution of 1,500 × 2,000 pixels. Exams were sorted in reverse chronological order. Each exam was assigned 3 binary labels for the presence or absence of (1) any abnormality, (2) an ACL tear, and (3) a meniscal tear. Definitions for labels were as follows:
Independent of the MSK radiologists, 7 practicing board-certified general radiologists and 2 practicing orthopedic surgeons at Stanford University Medical Center (3–29 years in practice, average 12 years) labeled the internal validation set, blinded to the original reports and labels. These clinical experts’ labels were measured against the reference standard labels established by the consensus of MSK radiologists. The general radiologists were randomly divided into 2 groups, with 4 radiologists in Group 1 and 3 radiologists in Group 2. The 2 orthopedic surgeons were also in Group 1. Group 1 first reviewed the validation set without model assistance, and Group 2 first reviewed the validation set with model assistance. For the reviews with model assistance, model predictions were provided as predicted probabilities of a positive diagnosis (e.g., 0.98 ACL tear). After a washout period of 10 days, Group 1 then reviewed the validation set in a different order with model assistance, and Group 2 reviewed the validation set without model assistance. The Stanford institutional review board approved this study.
Performance measures for the model, general radiologists, and orthopedic surgeons included sensitivity, specificity, and accuracy. We also computed the micro-average of these statistics across general radiologists only and across all clinical experts (general radiologists and surgeons). We assessed the model’s performance with the area under the receiver operating characteristic curve (AUC). To assess the variability in estimates, we provide 95% Wilson score confidence intervals [35] for sensitivity, specificity, and accuracy and 95% DeLong confidence intervals for AUC [36,37]. A threshold of 0.5 was used to dichotomize the model’s predictions. The model performance on the external validation set was assessed with the AUC and 95% DeLong confidence intervals.
Because we performed multiple comparisons in this study to assess the model’s performance against that of the practicing general radiologists and also to assess the clinical utility of providing model assistance, we controlled the overall false discovery rate (FDR) at 0.05 [38] and report both unadjusted p-values and adjusted q-values. Roughly, FDR < 0.05 can be interpreted as the expected proportion (0.05) of false claims of significance across all significant results. Thus, instead of using the unadjusted p-value to assess statistical significance, a q-value < 0.05 properly accounts for these multiple comparisons. To assess model performance against that of general radiologists, we used a 2-sided Pearson’s chi-squared test to evaluate whether there were significant differences in specificity, sensitivity, and accuracy between the model and the micro-average of general radiologists. The orthopedic surgeons were not included in this comparison.
We assessed the clinical utility of providing model predictions to clinical experts by testing whether the performance metrics across all 7 general radiologists and 2 orthopedic surgeons increased when they were provided model assistance. There is natural variability when a clinical expert evaluates the same knee MRI study at different times, so it is not unexpected that a clinical expert’s performance metrics will be slightly better or slightly worse when tested on two occasions, regardless of model assistance. Thus, we performed robust hypothesis tests to assess if the clinical experts (as a group) demonstrated statistically significant improvement with model assistance. We used a 1-tailed t test on the change (difference) in performance metrics for the 9 clinical experts for all 3 labels. To assess whether these findings were dependent specifically on the orthopedic surgeons’ improvement, we performed a sensitivity analysis: we repeated the 1-tailed t test on the change in performance metrics across only the general radiologists, excluding the orthopedic surgeons, to determine whether there was still significant improvement.
The exact Fleiss kappa [39,40] is reported to assess the level of agreement of the 3 MSK radiologists, whose majority vote was used for the reference standard labels. Additionally, to assess if model assistance may improve inter-rater reliability, we report the exact Fleiss kappa of the set of 9 clinical experts with and without model assistance for each of the 3 tasks.
All statistical analyses were completed in the R environment for statistical computing [41], using the irr, pROC, binom, and qvalue packages [38,42–44], and R code was provided with submission.
The inter-rater agreement on the internal validation set among the 3 MSK radiologists, measured by the exact Fleiss kappa score, was 0.508 for detecting abnormalities, 0.800 for detecting ACL tears, and 0.745 for detecting meniscal tears.
For abnormality detection, ACL tear detection, and meniscal tear detection, the model achieved AUCs of 0.937 (95% CI 0.895, 0.980), 0.965 (95% CI 0.938, 0.993), and 0.847 (95% CI 0.780, 0.914), respectively (Fig 5). In detecting abnormalities, there were no significant differences in the performance metrics of the model and general radiologists (Table 2). The model specificity for abnormality detection was lower than the micro-average of general radiologists, at 0.714 (95% CI 0.500, 0.862) and 0.844 (95% CI 0.776, 0.893), respectively. The model achieved a sensitivity of 0.879 (95% CI 0.800, 0.929) and accuracy of 0.850 (95% CI 0.775, 0.903), while the general radiologists achieved a sensitivity of 0.905 (95% CI 0.881, 0.924) and accuracy of 0.894 (95% CI 0.871, 0.913) (Table 2).
The model was highly specific for ACL tear detection, achieving a specificity of 0.968 (95% CI 0.890, 0.991), which is higher than the micro-average of general radiologists, at 0.933 (95% CI 0.906, 0.953), but this difference was not statistically significant (Table 2). General radiologists achieved significantly higher sensitivity than the model in detecting ACL tears (p-value = 0.002, q-value = 0.019); the micro-average general radiologist sensitivity was 0.906 (95% CI 0.874, 0.931), while the model achieved a sensitivity of 0.759 (95% CI 0.635, 0.850). The general radiologists also achieved significantly higher specificity in detecting meniscal tears (p-value = 0.003, q-value = 0.019), with a specificity of 0.892 (95% CI 0.858, 0.918) compared to a specificity of 0.741 (95% CI 0.616, 0.837) for the model. There were no other significant differences in the performance metrics (Table 2). Summary performance metric estimates and confidence intervals can be found in Table 2, and individual performance metrics for the 7 board-certified general radiologists and 2 orthopedic surgeons in this study can be found in S2 Table.
The clinical utility of providing model predictions to clinical experts during the labeling process is illustrated in Fig 6, and numerical values provided in Table 3. When clinical experts were provided model assistance, there was a statistically significant increase in the clinical experts’ specificity in identifying ACL tears (p-value < 0.001, q-value = 0.006). The mean increase in ACL specificity was 0.048 (4.8%), and since the validation set contained 62 exams that were negative for ACL tear, this increase in specificity in the optimal clinical setting would mean potentially 3 fewer patients sent to surgery for suspected ACL tear unnecessarily. Though it appeared that model assistance also significantly increased the clinical experts’ accuracy in detecting ACL tears (p-value = 0.020) and sensitivity in detecting meniscus tears (p-value = 0.028), these findings were no longer significant after adjusting for multiple comparisons by controlling the FDR (q-values = 0.092 and 0.110, respectively). There were no other statistically significant improvements to clinical experts’ performance with model assistance. Individual results, unadjusted p-values, and adjusted q-values are provided in S3 Table.
To determine whether the statistically significant improvement in specificity in identifying ACL tears with model assistance was dependent on the orthopedic surgeons’ performance metrics, we assessed the improvement of general radiologists only, excluding orthopedic surgeons. This sensitivity analysis confirmed that even among only general radiologists, there was a significant increase in specificity in identifying ACL tears (p-value = 0.003, q-value = 0.019; see S4 Table). Additionally, we computed Fleiss kappa for the 9 clinical experts with and without model assistance, and while we did not assess statistical significance, we observed that model assistance increased the Fleiss kappa measure of inter-rater reliability for all 3 tasks. With model assistance, the Fleiss kappa measure for abnormality detection increased from 0.571 to 0.640, for ACL tear detection it increased from 0.754 to 0.840, and for meniscal tear detection it increased from 0.526 to 0.621.
The MRNet trained on Stanford sagittal T2-weighted series and Stanford ACL tear labels achieved an AUC of 0.824 (95% CI 0.757, 0.892) on the Štajduhar et al. validation set with no additional training. Additionally, we trained 3 MRNets starting from ImageNet weights on the Štajduhar et al. training set with different random seeds. We selected the MRNet with the lowest average loss on the tuning set and then evaluated this model on the validation set. This model achieved an AUC of 0.911 (95% CI 0.864, 0.958) on the Štajduhar et al. validation set. Štajduhar et al. recorded an AUC of 0.894 for their best model, a semi-automated approach using support vector machines, although it was evaluated using a 10-fold cross-validation scheme [23]. MRNet took less than 30 minutes to train on and less than 2 minutes to evaluate the Štajduhar et al. dataset with an NVIDIA GeForce GTX 12GB GPU.
The purpose of this study was to design and evaluate a deep learning model for classifying pathologies on knee MRI and to compare performance to human clinical experts both with and without model assistance during interpretation in a crossover design. Our results demonstrate that a deep learning approach can achieve high performance in clinical classification tasks on knee MR, with AUCs for abnormality detection, ACL tear detection, and meniscus tear detection of 0.937 (95% CI 0.895, 0.937), 0.965 (95% CI 0.938, 0.965), and 0.847 (95% CI 0.780, 0.847), respectively. Notably, the model achieved high specificity in detecting ACL tears on the internal validation set, which suggests that such a model, if used in the clinical workflow, may have the potential to effectively rule out ACL tears. On an external dataset using T1-weighted instead of T2-weighted series and a different labeling convention for ACL injury, the same ACL tear model achieved an AUC of 0.824 (95% CI 0.757, 0.892). Retraining on the external dataset improved the AUC to 0.911 (95% CI 0.864, 0.958). Our deep learning model achieved state-of-the-art results on the external dataset, but only after retraining. It remains to be seen if the model would better generalize to an external dataset with more MRI series and a more similar MRI protocol. We also found that providing the deep learning model predictions to human clinical experts as a diagnostic aid resulted in significantly higher specificities in identifying ACL tears. Finally, in contrast to the human experts, who required more than 3 hours on average to completely review 120 exams, the deep learning model provided all classifications in under 2 minutes. Our results suggest that deep learning can be successfully applied to advanced MSK MRI to generate rapid automated pathology classifications and that the output of the model may improve clinical interpretations.
There are many exciting potential applications of an automated deep learning model for knee MRI diagnosis in clinical practice. For example, the model described could be immediately applied for diagnostic worklist prioritization, wherein exams detected as abnormal could be moved ahead in the image interpretation workflow, and those identified as normal could be automatically assigned a preliminary reading of “normal.” With its high negative predictive value for abnormalities, the model could lead to quick preliminary feedback for patients whose exams come back as “normal.” Additionally, providing rapid results to the ordering clinician could improve disposition in other areas of the healthcare system. In this work we noticed that specificity for detecting ACL tears improved for both general radiologists and orthopedic surgeons, which implies that this model could help reduce unnecessary additional testing and surgery. Automated abnormality prediction and localization could help general radiologists or even non-radiologist clinicians (orthopedic surgeons) interpret medical imaging for patients at the point of care rather than waiting for specialized radiologist interpretation, which could aid in efficient interpretation, reduce errors, and help standardize quality of diagnoses when MSK specialist radiologists are not readily available. Ultimately, more studies are necessary to evaluate the optimal integration of this model and other deep learning models in the clinical setting. However, our results provide early support for a future where deep learning models may play a significant role in assisting clinicians and healthcare systems.
To examine the effect that a deep learning model may have on the interpretation performance of clinicians, our study deliberately recruited general radiologists to interpret knee MRI exams with and without model predictions. We found a statistically significant improvement in specificity for the ACL tear detection task with model assistance and, though not statistically significant, increased accuracy for ACL tear detection and increased sensitivity for meniscal tear detection. For both general radiologists and non-radiologist clinicians (orthopedic surgeons), we found improved sensitivity and/or specificity across all 3 tasks with model assistance (Fig 5; Table 3), although the group of surgeons was too small for formal analysis. Importantly, model assistance also resulted in higher inter-rater reliability among clinical experts for all 3 tasks, with higher Fleiss kappa measures with model assistance than without. To our knowledge, this is the first study to explore providing outputs of deep learning models to assist radiologists and non-radiologist clinicians in the task of image interpretation. More work will be needed to understand whether and how deep learning models could optimize the interpretation performance of practicing radiologists and non-radiologist clinicians.
A difficulty in deep learning for medical imaging is curating large datasets containing examples of the wide variety of abnormalities that can occur on a given imaging examination to train an accurate classifier, which is a strategy we employed for detecting ACL and meniscal tears. However, our other classification task was to distinguish “normal” from “abnormal” with the intention that if the model could learn the range of normal for a given population of knee MRI exams, then theoretically any abnormality, no matter how rare, could be detected by the model. An example is shown in Fig 3A of a relatively uncommon but serious complete rupture of the gastrocnemius tendon, which was correctly classified and localized as “abnormal” by the model, despite the fact that there were no other examples of this specific abnormality in the abnormal training data. It is possible that with a binary approach and enough “normal” training data, a model could detect any abnormality, no matter how uncommon. However, more work is needed to explore whether subtler abnormalities would require specific training data.
This study has limitations. Our validation set ground truth was not governed strictly by surgical confirmation in all cases. The deep learning model described was developed and trained on MRI data from 1 large academic institution. While MRNet performed well on the external validation set without additional training (AUC 0.824), we saw a substantial improvement (AUC 0.911) after training on the external dataset. This finding suggests that achieving optimal model performance may require additional model development using data more similar to what the model is likely to see in practice. More research is needed to determine if models trained on larger and multi-institutional datasets can achieve high performance without retraining. Power to detect statistically significant gains in clinical experts’ performance with model assistance was limited by the size of the panel, and a larger study that includes more clinical experts as well as more MRI exams may detect smaller gains in utility. Nevertheless, we have shown that even in this small set of clinical experts, providing model predictions significantly increased ACL tear detection specificity, even after correcting for multiple comparisons.
In conclusion, we developed a deep learning model that achieves high performance in clinical classification tasks on knee MRI and demonstrated the benefit, in a retrospective experiment, of providing model predictions to clinicians during the diagnostic imaging task. Future studies are needed to improve the performance and generalizability of deep learning models for MRI and to determine the effect of model assistance in the clinical setting.
|
10.1371/journal.ppat.1000063 | HtrA2/Omi Terminates Cytomegalovirus Infection and Is Controlled by the Viral Mitochondrial Inhibitor of Apoptosis (vMIA) | Viruses encode suppressors of cell death to block intrinsic and extrinsic host-initiated death pathways that reduce viral yield as well as control the termination of infection. Cytomegalovirus (CMV) infection terminates by a caspase-independent cell fragmentation process after an extended period of continuous virus production. The viral mitochondria-localized inhibitor of apoptosis (vMIA; a product of the UL37x1 gene) controls this fragmentation process. UL37x1 mutant virus-infected cells fragment three to four days earlier than cells infected with wt virus. Here, we demonstrate that infected cell death is dependent on serine proteases. We identify mitochondrial serine protease HtrA2/Omi as the initiator of this caspase-independent death pathway. Infected fibroblasts develop susceptibility to death as levels of mitochondria-resident HtrA2/Omi protease increase. Cell death is suppressed by the serine protease inhibitor TLCK as well as by the HtrA2-specific inhibitor UCF-101. Experimental overexpression of HtrA2/Omi, but not a catalytic site mutant of the enzyme, sensitizes infected cells to death that can be blocked by vMIA or protease inhibitors. Uninfected cells are completely resistant to HtrA2/Omi induced death. Thus, in addition to suppression of apoptosis and autophagy, vMIA naturally controls a novel serine protease-dependent CMV-infected cell-specific programmed cell death (cmvPCD) pathway that terminates the CMV replication cycle.
| Cellular suicide is an effective host defense mechanism to control viral infection. Host cells encode proteins that induce infected cell death while viruses encode proteins that prevent death and facilitate viral replication. Human cytomegalovirus encodes vMIA to suppress host-initiated death pathways. Cytomegalovirus infection is controlled by the evolutionarily ancient mitochondrial serine protease, HtrA2/Omi. HtrA2/Omi levels rise dramatically within mitochondria at late times during viral infection, eventually overcoming viral control of a cell death pathway that is dependent on this serine protease and independent of the well-studied apoptotic cell death pathway that conventionally depends upon a class of proteases called caspases. vMIA naturally counteracts HtrA2/Omi-dependent cell death and allows infected cells to survive and produce virus for several days. The natural inhibitory role of vMIA can be overwhelmed by overexpression of HtrA2/Omi in virus-infected cells, but uninfected cells are insensitive to HtrA2/Omi-induced death. The broad distribution of HtrA2/Omi within mammalian host species suggests this may represent an ancient antiviral response or a process of viral detente that establishes the timing of infection. Either way, the success of cytomegalovirus rests in the balance between cell death initiation and the viral cell death suppressor vMIA.
| Cell death is central to viral infection, as an evolutionarily-conserved means to eliminate intracellular pathogens and as a way that lytic viruses mediate release of progeny. Human cytomegalovirus (CMV), the major infectious cause of birth defects as well as an important cause of opportunistic disease worldwide [1], remains cell-associated during productive replication. Release of progeny virus depends upon the exocytic pathway [1] and continues until cells die via a poorly understood fragmentation process. CMV is well-armed to modulate cell-intrinsic as well as extrinsic innate and adaptive host clearance pathways [1]. The product of the UL37x1 gene, vMIA, a potent suppressor of apoptosis [2]–[4], also controls the timing of infected cell death [5]–[7]. Premature death in vMIA-mutant virus-infected cells reduces the period of progeny release by three to four days [5]–[7] without affecting cell-to-cell spread [5]. All vMIA-mutant viruses exhibit this premature death phenotype, but the involvement of caspases and the impact on viral yield varies with CMV strain. AD169varATCC strain (AD-BAC) depends upon vMIA to a greater extent [6],[7] than TownevarATCC (Towne-BAC), although vMIA prolongs the period of viral replication and release in both strains [5]. Importantly, vMIA from either strain retains the capacity to block caspase-dependent apoptosis [5]. The caspase-independent death pathway that is blocked by vMIA is not known. Other cell death suppressors are encoded by CMV [1], but, aside from vMIA, only UL38 has been implicated in control of infected fibroblast death to prolong replication [8],[9]. Studies to date reveal a complexity of infected cell death and a need for a more complete understanding of events that naturally terminate CMV infection.
Major pathways associated with death (apoptosis, necrosis, and autophagy) are triggered by specific host cell and immune system initiators and exhibit characteristic molecular events and cell morphological changes [10]–[12]. Proteases in the caspase, calpain, lysosomal cathepsin, and proteasomal serine protease classes are central to the execution of various death pathways. The fact that the premature death induced by vMIA-mutant CMV is resistant to inhibitors of caspases, cathepsins, and calpains [5] suggests a novel programmed pathway distinct from characterized death pathways [11],[13]. For those viral strains that have been characterized, infected cell death initiates approximately 7 to 10 days after infection of fibroblasts. In contrast, the premature death that occurs in vMIA mutant virus infection initiates 3 to 4 days postinfection [5]–[7]. The 12–24 h timing of individual cell fragmentation, association with cytopathic effect (CPE), and nominal impact of vMIA [5] all suggest that the final stages of productive replication terminate with a CMV infected cell-specific programmed cell death (cmvPCD).
Many DNA viruses encode antiapoptotic functions that sustain replication in the face of cell-intrinsic defenses [14]–[16]. vMIA equips CMV to counteract intrinsic host clearance pathways leading to cell death [5]–[7]. As an outer mitochondrial membrane protein, vMIA sits in a central position analogous to antiapoptotic Bcl-2 family members Bcl-2 and Bcl-xL [2], and prevents the formation of a mitochondrial permeability transition pore, release of cytochrome c and proapoptotic factors into the cytoplasm, and activation of executioner caspases. Unlike antiapoptotic Bcl-2 family members [2],[5],[17], vMIA lacks Bcl-2 homology domains but depends on an antiapoptotic domain that mediates interaction with GADD45 family members [18],[19]. vMIA also recruits BAX to mitochondria [20],[21] and disrupts mitochondrial networks [22]. This disruption normally accompanies BAX oligomerization at the outer mitochondrial membrane [23],[24], although vMIA mutants that fail to bind BAX continue to disrupt networks [25]. The vMIA-dependent recruitment of BAX does not lead to the formation of a transition pore complex or release of proapoptotic mediators [20],[21]. The contribution of BAX oligomerization or mitochondrial network disruption to cell death suppression remains to be investigated. Although both of these events are signs of apoptosis [23],[24],[26],[27] neither mitochondrial network disruption [28] nor BAX oligomerization [29],[30] are sufficient to induce apoptosis. These alterations are also associated with vMIA-mediated suppression of cell death during viral infection where the pathway(s) of death are not fully understood.
Consequences of mitochondrial release of proapoptotic mediators have been extensively studied [10], [31]–[33]. Cytochrome c controls apoptosome formation and downstream executioner caspase activation. Endonuclease G and apoptosis-inducing factor (AIF) promote nuclear events. Mitochondrial release of Smac/DIABLO and HtrA2/Omi overcomes the activity of inhibitor of apoptosis proteins (IAPs). The HtrA2/Omi proenzyme is processed within the mitochondria, removing a mitochondrial targeting sequence (amino terminal 33 amino acids) and a transmembrane domain [34]–[36]. Mature, active HtrA2/Omi resides in the intermembrane space, and is released into the cytoplasm through the transition pore complex at the same time as cytochrome c. Release of the serine protease HtrA2/Omi from mitochondria can result in two downstream effects: (1) cleavage of IAPs and an ultimate increase in caspase-dependent death and (2) trigger IAP-independent, caspase-independent death [37]. This latter pathway is also induced by extramitochondrial overexpression of HtrA2/Omi in the cytoplasm [35], [37]–[40]. The role of this serine protease as an inducer of cell death [38],[39] in mammalians seems opposite the role of the founding member of this protein family as a pro-survival serine protease in eubacteria [41]–[43]. Here, we demonstrate the central role of HtrA2/Omi executing a serine protease-dependent pathway that is controlled by vMIA during infection.
We evaluated cmvPCD during wild type (wt) virus (Towne-BAC, a GFP-expressing virus [44]) infection by scoring morphological changes in cells during replication (Fig. 1, supplemental Fig. S1). Termination of infection was associated with the accumulation of GFP-positive cell debris that remained associated with the monolayer (Fig. 1A). Cell fragmentation and death was first observed at 120 h postinfection in a small percentage of GFP+ foci (Fig. 1A, Fig. S1). GFP+ dead cell debris was observed only in foci (Fig. S1). Almost all (>90%) foci showed evidence of fragmentation by 240 h postinfection (Fig. 1C). Thus, cmvPCD occurred very late in infection, after maturation and release of progeny virus had peaked. GFP+ debris was observed much earlier during infection with vMIA null mutant virus, ΔUL37x1 (Fig. 1C), although the fragmentation process appeared similar to wt virus (Fig. 1B, Fig. S1). As previously reported [5], a majority (70%) of mutant virus foci contained debris by 120 h due to the single GFP+ cells that started to fragment between 72 and 96 h postinfection prior to the formation of foci (Fig. S1). There was a gradual increase in foci containing fragmented cells between 120 and 192 h postinfection such that, by 192 h, >90% of foci contained fragmented cells (Fig. 1C). This was consistent with our previous report showing both viruses spread with equivalent efficiency but that ΔUL37x1 induced premature caspase-independent death [5].
Intact GFP+ ΔUL37x1-infected cells appeared to fragment before virus spread to form foci (Fig. 1C, Fig. S1) [5]. To determine whether intact ΔUL37x1-infected cells released progeny virus before fragmenting, we evaluated the formation of immediate early nuclear antigen positive (IE+) foci by immunofluorescence. Infected cells became IE+ earlier than they became GFP+ (Fig. 2). IE+ foci surrounding single GFP+ cells were detected at 72 h postinfection (Fig. 2A–E), when a majority (>50%) of GFP+ cells were still intact (Fig. 2A, 2D). Whether intact or fragmented, >99% of ΔUL37x1-infected cells produced foci by 120 h postinfection [5]. Thus, most virus spread occurs before cells fragmented (Fig. 2A, 2D), suggesting that, like wt, mutant virus is released before infected cells die (Fig. 1, Fig. S1).
To confirm that the death of mutant virus-infected cells was dependent on late phase events, the DNA synthesis inhibitor phosphonoformate (PFA) or the DNA encapsidation inhibitor 2-bromo-5,6-dichlorobenzimidazole (BDCRB) was added at the time of CMV infection and cell fate was scored by staining for IE+ cells at 96 h postinfection (Fig. 2B, 2F). PFA inhibits late gene expression, including GFP, while BDCRB blocks virion maturation but allows late gene expression to proceed [45],[46]. Untreated control or BCDRB-treated infected cultures appeared similar, whereas PFA-treated cultures contained single IE+ cells that remained GFP-. Only about 30% of GFP+ cells or debris in untreated ΔUL37x1-infected cells were IE+ at 96 h postinfection (Fig. 2F). The remaining 70% were no longer IE+, suggesting that IE expression was lost as cells fragmented. BDCRB-treated cells exhibited a pattern similar to untreated cells, suggesting that fragmentation was triggered by events prior to DNA encapsidation. All infected cells in PFA-treated cultures remained IE+ (Fig. 2F) and did not fragment, suggesting that initiation of cell death was dependent on events that followed viral DNA synthesis. Thus, initiation of death was dependent on cellular changes associated with viral DNA replication and/or late phase gene expression.
We evaluated cell morphology [11] associated with death in ΔUL37x1 and wt virus-infected cells (Fig. 3). Late in CMV infection, inclusions form within enlarged cells coincident with replication and maturation processes that take place in the nucleus as well as in the cytoplasm. At 72 h postinfection, ΔUL37x1 and wt virus-infected cells exhibited similar nuclear and cytoplasmic inclusions [1],[47],[48] as well as enlarged cell CPE (Fig. 3A, 3F and Fig. S2) [5]. Stain for total nuclear DNA revealed a diffuse pattern (Fig. 3F, Fig. S2) that became distorted (Fig. 3G) and progressed through shrinkage and collapse and loss of nuclei (Fig. 3H–J) as cells fragmented (Fig. 3C–E). A similar process accompanied fragmentation in mutant or wt infected cells. The fragmentation process produced cell debris (Fig. 3E) lacking signs of DNA (Fig. 3J). Loss of nuclei scored by DNA stain or IE antigen (Fig. 2) was similar. Cell debris remained GFP+ and unstained by ethidium homodimer (Fig. S3), suggesting a non-necrotic death. Fragmentation was not synchronous in infected cultures, such that only 10% of infected cells exhibited intermediate fragmentation patterns (e.g. Fig. 3B–D) at any time (Fig. 3K and Fig. S1). The same types of morphological changes that started at 72 h in ΔUL37x1-infected cultures started at 120 h postinfection in Towne-BAC-infected cells (Fig. S1 and S4). The fragmentation of GFP+ cells (Fig. 1), loss of IE+ cells (Fig. 2) and loss of DNA+ nuclei (Fig. 3) were all characteristic of cmvPCD in wt and premature death in mutant virus-infected cells.
Previous work showing that caspases, cathepsins, or calpains were not involved in ΔUL37x1-initiated premature death [5], lead us to evaluate the contribution of cellular serine proteases to this process. We started by assessing the impact of a broad-spectrum inhibitor, TLCK [49]–[62], because this inhibitor does not affect the viral maturational serine protease at concentrations that are sufficient to block cellular serine proteases [63]. Addition of TLCK (11, 33, or 100 µM) to infected cultures at 30 h lead to a concentration-dependent reduction in cell fragmentation at 72 h postinfection (Fig. 4A). These concentrations of inhibitor did not reduce virus yields (Fig. 4B). Thus, TLCK inhibited premature cell death without any impact on virus replication. When TLCK was added at 30 h and fragmented cells were counted at 72, 96, and 120 h postinfection (Fig. 4C), cell death was reduced approximately twofold, suggesting that serine proteases play a central role in the timing of fragmentation. Despite experiment-to-experiment variability in the levels and rate of fragmentation death observed between 72 and 120 h postinfection, TLCK consistently inhibited this process (Fig. 4A and C) and increased the proportion of live, intact cells while absolute numbers of GFP+ cells or debris remained the same (Fig. 4D). This result implicated serine proteases early in the premature death induced by mutant virus.
Previously, we reported that the pan-caspase inhibitor zVAD.fmk had no effect on ΔUL37x1-induced premature death [5]. To determine whether caspases influenced death levels when serine proteases were inhibited, we applied zVAD.fmk alone as well as in combination with TLCK. The caspase inhibitor did not influence the serine protease-dependent process (Fig. 4E). In contrast, zVAD.fmk showed the expected [5],[6],[64] inhibition of apoptosis induced in CMV strain AD169varATCC infected cells (Fig. 4F). Thus, these data imply that cmvPCD is controlled by serine proteases that work independent of caspases.
To determine the timing of serine protease activity in controlling premature death, TLCK was added to ΔUL37x1-infected cells at 30, 54, or 78 h. Addition of TLCK at each of these times was found to dramatically reduce the level of death at 96 h postinfection (Fig. 5A). These results suggest serine proteases act within 24 h of fragmentation (Fig. 4D) and demonstrated the importance of these proteases late in infection. Taken together with data on timing of the death stimulus (Figs. 2 and 3, and [5]), serine proteases active late in CMV infection may either trigger or play an intermediary role in the cell death pathway. To determine the timing of serine protease activity in wt virus-induced cmvPCD, TLCK was added at 30, 54, and 78 h (Fig. 5B). Addition of TLCK at each of these times effectively reduced cell fragmentation at 144 h postinfection, suggesting that proteases active after the 78 h time period played a critical role during wt virus infection as well (Fig. 5B). These data demonstrate a common serine protease cell death pathway terminates mutant or wt virus infection, and demonstrate that the premature death in mutant virus infected cells follows a similar pathway as cmvPCD. Differences in timing show the importance of vMIA control in the timing of cmvPCD.
One mitochondrial serine protease, HtrA2/Omi, has been implicated in cell death pathways and exhibits sensitivity to TLCK [40], [65]–[67]. The specific HtrA2/Omi inhibitor UCF-101 [68] was added to ΔUL37x1- or Towne-BAC-infected cultures (Fig. 5C–E) at a concentration (10 µM) anticipated to minimize a previously recognized impact on other cellular targets [69]. UCF-101 reduced death when added to ΔUL37x1 or Towne-BAC-infected cultures at 30 or 54 h, implicating HtrA2/Omi as a mediator of cmvPCD (Fig. 5C–D). Although UCF-101 added at 54 h reduced death of ΔUL37x1-infected cells at 96 h postinfection, addition at 78 h was ineffective (Fig. 5C). Towne-BAC-associated death at 144 h was reduced by UCF-101 added as late as 102 h, but not when added at 126 h (Fig. 5D). UCF-101 treatment did not reduce viral yields (Fig. 5E). Most importantly, these data suggest that events over the 24 to 48 h preceding fragmentation of cells are influenced by HtrA2/Omi, regardless of whether considering the premature cmvPCD in mutant virus infected cells or cmvPCD in wt infection. The differences between UCF-101 and TLCK addition at 78 h (Fig. 5A) may be due to the effectiveness of these inhibitors on HtrA2/Omi or to additional serine proteases that contribute to cmvPCD. Overall, these data demonstrate that UCF-101 specifically reduces infected cell death and implicate the serine protease HtrA2/Omi in the pathway. Further, these data implicate HtrA2/Omi as a target of vMIA modulation.
To determine the impact of mutant or wt virus infection on HtrA2/Omi expression levels and subcellular localization as well as to investigate any impact of vMIA on HtrA2 expression, immunoblot analysis was carried out on Towne-BAC infected cells (Fig. 6). Levels of mature 36 kDa HtrA2/Omi levels were similar to uninfected cells at 24 h, but increased by 48 h and continued to accumulate over the course of infection (Fig. 6A–B). Comparisons of mutant and wt infected cells showed similar accumulation of HtrA2 by 48 h (Fig. 6B, Fig. S5). Premature cmvPCD initiated in mutant virus-infected cells prevented comparisons by immunoblot later in infection; however, immunofluorescence analyses at 96 h postinfection confirmed the dramatically increased HtrA2/Omi levels in mutant or wt virus-infected cells (Fig. 6C–H and Fig. S6). HtrA2/Omi colocalized with the mitochondrial membrane potential marker MitoTracker Red (Fig. 6C, 6F and Fig. S6) at late times of infection with either virus. These data indicate that HtrA2/Omi levels increase within mitochondria before the initiation of cmvPCD. vMIA does not alter expression pattern or mitochondrial localization of this protease but nevertheless prevents death.
Mitochondria in wt CMV infected cells followed the expected [22] reticular to punctate transition associated with disruption of mitochondrial networks (Fig. 6F, 6L, and Fig. S6) and mutant virus-infected cells retained a reticular pattern (Fig. 6C, 6I and Fig. S6) when stained for HtrA2/Omi, cytochrome c, mitochondrial HSP (mtHSP70), or MitoTracker Red. MitoTracker Red staining indicated that mitochondria retained a similar membrane potential despite this difference in morphology due to vMIA (Fig. S6). When the kinetics of the reticular to punctate transition was evaluated in Towne-BAC-infected cells, almost all (≥90%) of cells contained reticular mitochondria at 48 h, but transitioned to punctate by 96 h. In contrast, ΔUL37x1-infected cells retained a reticular morphology (≥80%) throughout infection. These data suggest that a vMIA-dependent process disrupts reticular mitochondria beginning at 48 h postinfection and this change in mitochondrial organization may contribute to cell survival. Despite this striking difference in mitochondria, the organelles of the secretory apparatus that form the viral assembly compartment at late times of infection [48],[70] were similar in either virus infection (Fig. S2, Fig. S6). Thus, disruption of mitochondrial networks by wt virus may contribute to control of HtrA2/Omi-dependent death and the failure of mutant virus to induce these changes may lead to premature HtrA2/Omi-dependent death.
We sought to determine whether mitochondria released cytochrome c prior to premature cmvPCD. Cells that had not yet started to fragment all showed reticular cytochrome c staining (Fig. 6 and Fig. S6) whereas diffuse staining was detected only as cells became highly fragmented (Fig. S7). These data suggest that release of cytochrome c follows the fragmentation that characterizes cmvPCD.
To directly address the impact of HtrA2/Omi overexpression on the cell fate, full-length HtrA2/Omi as well as a catalytic site mutant (HtrA2S306A) [34] were transiently expressed in uninfected and virus-infected cells. Initially, expression levels and impact on uninfected cell viability were assessed (Fig. 7A–H, Fig. S8). HtrA2/Omi (or mutant HtrA2/Omi) overexpression did not induce death in uninfected HFs (Fig. 7H) or HeLa cells (Fig. S8), consistent with published characterization of full-length protease [37]. Immunofluorescence patterns revealed the expected mitochondrial localization at 48 h post transfection (Fig. 7A–F), and immunoblot analyses using HeLa cells indicated equivalent expression levels of the wt and mutant protease (Fig. 7G). To determine the impact of HtrA2/Omi overexpression on infected cells, Towne-BAC or ΔUL37x1 were cotransfected with HtrA2/Omi or HtrA2S306A expression plasmids (Fig. 7I) and assessed for spread to form foci [71]. By 10 days posttransfection, wt and mutant BACmids had produced comparable numbers of plaques, as expected [5]. Cotransfection of HtrA2/Omi expression plasmid reduced the plaquing efficiency >10-fold compared to vector control or HtrA2S306A mutant (Fig. 7I). These data show that overexpression of catalytically active HtrA2/Omi prevents plaque formation independent of vMIA expression.
To determine whether the reduction in plaguing efficiency following overexpression of HtrA2/Omi was due to cell death induction, the fate of individual cells was monitored (Fig. S9). When Towne-BAC was cotransfected with HtrA2/Omi or HtrA2S306A, individual GFP+ cells were observed at 48 h, although even at this time the levels could be lower in cells receiving the protease active form (Fig. 7J–K). HtrA2/Omi/GFP+ cells began to fragment by 72 h posttransfection (Fig. 7J) and were lost from cultures by 168 h (Fig. S9). HtrA2/Omi overexpression-induced death required the active protease, based on the failure of HtrA2S306A to induce death (Fig. 7J) as well as on the ability of the inhibitor UCF-101 to block HtrA2/Omi overexpression-induced death (Fig. 7K). The numbers of GFP+ cells (Fig. 7J) or plaques (Fig. 7I) that formed following cotransfection of Towne-BAC with HtrA2S306A could not be distinguished from transfection of Towne-BAC with vector. These data show that overexpression of catalytically active HtrA2/Omi induces infected cell death that is independent of vMIA expression. The sensitivity of virus-infected cells and lack of impact on uninfected HFs (Fig. 7H) supports the specific role of HtrA2/Omi in a novel cell death pathway in CMV-infected cells.
A role of vMIA in HtrA2-induced death was investigated using the cotransfection assay carried out using lower doses of expression plasmids as well as using vMIA-expressing cells. Cotransfection of HtrA2/Omi expression plasmid at a 25 or 30-fold lower level revealed a differential impact on these viruses (Fig. 8A), where Towne-BAC exhibited a greater resistance to HtrA2/Omi-induced death. These conditions were also employed to demonstrate that vMIA overexpression overcame HtrA2/Omi-induced death (Fig. 8A). To address the role of vMIA further, HFs as well as HFs stably transduced with retroviruses expressing Myc-tagged vMIA or mutant protein [18] vMIAmut (Fig. 8B–C) were infected. As expected [5], vMIA-HFs suppressed the premature cmvPCD when assessed at 96 and 120 h postinfection, whereas vMIAmut-HF or nontransduced control HFs did not (Fig. 8B). These data suggest the intact antiapoptotic domain of vMIA is required to control premature cmvPCD. The experimental plating efficiency of ΔUL37x1 virus was the same on either cell line (Fig. S10 and [5]). These results were consistent with a role for vMIA in controlling kinetics of cmvPCD and suggest that similar functional domains of vMIA are required in suppression of apoptosis or HtrA2-dependent cmvPCD. Immunoblot analyses were used to compare transduced vMIA (or vMIAmut) levels relative to native viral expression (Fig. 8C). The lower levels of transduced gene expression likely contribute to the death suppression observed (Fig. 8B). Rescue viruses derived from ΔUL37x1 confirmed that an intact UL37x1 locus is sufficient to completely control premature death, mitochondrial organization, and viral yield (Fig. 8D, Fig. S10). Overall, these data confirm the critical role of vMIA as a determinant of cmvPCD when induced by overexpression of HtrA2/Omi transfection or during the late phase of infection. Thus, ΔUL37x1 infection sensitizes to the prodeath impact of HtrA2/Omi, and vMIA controls HtrA2/Omi prodeath pathways during wt CMV infection.
In order to determine whether the antiapoptotic activity of vMIA is preserved in cells where HtrA2/Omi is overexpressed, we performed experiments with HtrA2/Omi expression constructs in HeLa cells exposed to Fas-mediated apoptosis (Fig. 9) [2],[19]. Immunofluorescence analyses showed expected levels and localization of HtrA2/Omi and vMIA in transfected cells (Fig. 9A–I). These analyses indicated that vMIA and HtrA2/Omi (or HtrA2S306A) colocalize with mitochondria under all conditions. Introduction of HtrA2/Omi or mutant expression constructs did not influence the antiapoptotic activity of vMIA (Fig. 9J–L), consistent with previous work showing vMIA-dependent antiapoptotic function is active at late times of infection [5],[6]. Together, these data suggest that HtrA2/Omi does not interfere with vMIA-mediated control of apoptosis.
To directly visualize levels of serine proteases in infected cells, the fluorescent reagent sulforhodamine 101-leucine chloromethyl ketone (SLCK) was used to reveal the distribution and activity of serine proteases [72] in ΔUL37x1 or Towne-BAC infected live cell cultures (Fig. 10). By day 5, foci with brightly stained GFP+ debris was observed in cultures infected with either virus (Fig. 10E–F), although fragmentation was rare in wt virus-infected cultures at this time. The SLCK staining pattern was distinct in ΔUL37x1-infected cells and included bright SLCK+ debris (Fig. 10A) that was distinguishable from Towne-BAC infected cells by differences in the amount of staining as well as the size and distribution of debris (Fig. 10A–B). By 8 days after infection, most ΔUL37x1-infected cells in each plaque were brightly fluorescent (Fig. 10C) whereas cells infected with wt virus (Fig. 10D) showed only SLCK+ debris. SLCK staining patterns did not appear to be mitochondrial at any time in either virus infection. These patterns were distinct from HtrA2/Omi (Fig. 6C), suggesting that SLCK labeling detected serine proteases in addition to HtrA2/Omi. Addition of 0.1 or 1 mM TLCK reduced but did not eliminate SLCK binding to mutant virus-infected cells, consistent with the induction of serine proteases (Fig. S11). Overall, >50% of GFP+ cells in ΔUL37x1 plaques also labeled with SLCK. SLCK staining was reduced to ≤30% by addition of 100 µM TLCK. Thus, SLCK revealed a higher level of protease activation in CMV infected cells that were susceptible to premature cmvPCD. This data suggests that vMIA may control a broader serine protease-dependent death pathway by counteracting mitochondrial HtrA2/Omi during viral infection.
CMV replicates in the nucleus, matures in the cytoplasm and is released into the surrounding medium or adjacent cells over the course of a 7 to 10 day replication cycle [1]. Host cells are dramatically reprogrammed for production of progeny virus until death occurs via a process that begins late in CMV infection, associated with late gene expression that drives CPE and cell cycle dysregulation [73]–[76]. In an effort to define viral and cellular contributions to morphological and biochemical events that terminate CMV infection, we have discovered the key role of mitochondrial HtrA2/Omi and a novel cell death pathway. This cellular serine protease appears to be responsible for induction of cmvPCD following a pathway that is held in balance by the viral cell death suppressor, vMIA. vMIA resides in the mitochondrion where it is a potent suppressor of cytochrome c release, thereby preventing activation of executioner caspases during apoptosis [2]–[4]. In addition to suppression of apoptosis, vMIA carries out a distinct and nonoverlapping role suppressing death induced by HtrA2/Omi during the late phase of viral infection. This cmvPCD pathway is triggered only in the context of infection. Late phase infected cell events promote cell fragmentation together with collapse, shrinkage, and loss of nuclei in a pathway that is dependent on HtrA2/Omi protease activity and associated with the activation of cytoplasmic serine proteases that may act as executioners. HtrA2/Omi levels rise before induction of death, consistent with a central role of this protease in initiation of cmvPCD. Suppression of this death pathway, like suppression of apoptosis, is associated with global disruption of mitochondrial networks by vMIA. Unlike apoptosis, however, cmvPCD apparently does not require cytochrome c release from mitochondria to trigger downstream events. Further, HtrA2/Omi remains mitochondrial late during infection suggesting death may be initiated by the activity of the intramitochondrial protease, which raises an interesting question as to how transduction of the death signal occurs. Our data reveal a pathway that is triggered by high intramitochondrial HtrA2/Omi protease and controlled by vMIA.
Although vMIA-mutant virus undergoes premature cmvPCD, the fragmentation process is similar to cmvPCD in wt virus-infected cells. The difference appears to be in timing of cell death. vMIA delays death for several days beyond the initiating trigger which is coincident with the late phase of replication. Although induction of HtrA2/Omi is independent of vMIA, the impact of induction appears to be the target of vMIA function at the mitochondria where both reside. Suppression of cmvPCD benefits the virus by extending the period of virus production by infected cells [5], although cultured fibroblasts show only slight reduction in yield and cell-to-cell transmission in the absence of vMIA. A prolonged period of virus production increases the amount of virus released cell-free and potentially benefits transmission in natural settings. A delay in fragmentation would also delay phagocytosis and clearance of virus-infected cells [77]. HtrA2/Omi-dependent death may be viewed as an intrinsic host antiviral process analogous to apoptosis. vMIA control of HtrA2/Omi-mediated death is analogous to control of apoptosis, as both appear to be independent cell-intrinsic mechanisms of pathogen control. Importantly, vMIA appears to provide concurrent protection from both pathways.
vMIA disruption of reticular networks and organization of mitochondria [22] is independent of HtrA2/Omi accumulation within mitochondria, but does correlate with cell death suppression activity. Thus, ΔUL37x1-mutant virus-infection preserves mitochondrial networks throughout infection, during HtrA2/Omi accumulation and initiation of premature cmvPCD. In contrast, wt virus infected cells support the same accumulation of HtrA2/Omi and a vMIA-driven disruption of mitochondrial networks but survive. The correlation between this vMIA-dependent disruption and cell death protection suggests that punctate mitochondria may be more resistant to the stress induced by late phase events. Reticular mitochondria are known to rapidly disseminate Ca++ or ATP signals; whereas, punctate mitochondria have slower responses to changes in intracellular mediators [78]. Additional experiments will be needed to understand the mechanism underlying resistance of punctate mitochondria to death, whether mediated via caspases or HtrA2/Omi.
Emerging evidence suggests vMIA, viral strain differences, and cellular factors contribute to the control of mitochondria and death. Thus, AD-BAC kills cells earlier [6] and disrupts mitochondrial networks by 24 h postinfection [22] whereas Towne-BAC disruption occurs later, by 48 h postinfection and cells die later. vMIA associates with the outer mitochondrial membrane within 24 h [79],[80]. AD-BAC (or its parent AD169varATCC) depends upon vMIA to suppress caspase-dependent apoptosis that develops by 48 to 72 h postinfection. Towne-BAC (or its parent TownevarATCC) depends on vMIA to suppress caspase-independent, HtrA2-dependent cell death that develops by 72 to 96 h postinfection. It remains to be seen whether vMIA suppresses both pathways in cells infected with strains like AD-BAC. Accumulation of HtrA2/Omi occurs in other viral strains (McCormick, unpublished), underscoring the overall importance of the process described here. There are many potential factors contributing to qualitative or quantitative differences in the way characterized viral strains initiate and control death, with apoptosis apparently predominating in some settings and HtrA2-mediated death predominating in others. We focused here on dissecting the novel death pathway in Towne-BAC-infected cells, to characterize a novel HtrA2/Omi pathway that is independent of apoptosis.
cmvPCD may be influenced by or even associated with a number of additional modulatory effects of this virus that impact late times of infection, including dysregulation of the cell cycle [73]–[76], disruption of p53 activation [81], DNA damage response [82],[83] and unfolded protein response [84] that all remain incompletely understood. vMIA may reduce ATP levels during infection as it does in established cell lines. Although suggested to control late CPE in AD-BAC derivatives [85] vMIA has no impact on development of CPE in Towne-BAC derivatives. Any vMIA-specific reduction in ATP levels is likely highly coordinated with other viral processes contributing to late CPE. CMVs encode multiple factors that target mitochondria [86],[87], regulate expression of mitochondrial proteins [75] and even stimulate mitochondrial DNA synthesis [88] suggesting viral control of mitochondria functions is complex. The vMIA-specific impact on ATP levels as related to HtrA2/Omi remains unknown but may certainly be a feature of control. As an event that occurs very late in replication, cmvPCD is crucial to sustaining viral infection in individual cells. Our observations that either mutation of vMIA or premature overexpression of HtrA2/Omi levels dramatically alters the timing of death indicate that these two may balance each another in controlling cmvPCD.
Previously, pharmacological inhibitor and overexpression studies have implicated HtrA2/Omi as a regulator of death [35],[37],[40],[89],[90]. Genetic studies have suggested this protease functions primarily to ensure normal mitochondrial homeostasis [39],[91],[92], perhaps controlling protein quality and cellular stress responses [34] similar to the related bacterial protease HtrA [41]–[43]. The role of HtrA2/Omi in caspase-independent cell death has not previously been studied in detail, although the truncated, active form of HtrA2 drives death when released from mitochondria or expressed directly in the cytoplasm [35],[37],[40],[90],[93]. We have shown that the active form drives death specifically and only in CMV-infected cells, which we correlate with the fact that the enzyme remains mitochondrial throughout CMV infection. CMV infection is a unique setting that has unveiled a direct role for HtrA2/Omi in a caspase-independent cell death pathway analogous to apoptosis.
vMIA controls the programmed death of infected cells after a week or more of replication, following a period of persistent virus production. CMV infects many cell types in addition to HFs, and given that the timing of replication varies with cell type, vMIA control of HtrA2/Omi-dependent death may be critical in other cell types or in natural infection of the human host. Given the many functions that CMV has evolved to manage the virus:host standoff, we speculate that viral control of cmvPCD represents a benefit to the virus, potentially allowing infected cells to avoid sending alarm signals. Other examples of viral proteins acting together to control the type of cell death that follows replication can be identified. Thus, the adenovirus death protein, ADP, functions in the presence of E1B-19k, the viral Bcl-2 protein, and both contribute to the type of death that terminates infection [94]–[96]. Caspase-dependent apoptosis is itself a cell-intrinsic pathogen clearance process, minimizing inflammation and pathology while alarming the immune system to initiate cellular responses [77]. CMV-encoded cell death suppressors provide a means of evading cell death directed by host cell intrinsic, innate, and adaptive responses [97]. The benefit of controlling the mechanism and timing of cell death includes persistence, as well as the interface of virus-infected cells with the host immune system. In the host, cmvPCD may provide for greater success in attaining a foothold without evoking clearance. The presence of vMIA-like functions in other cytomegaloviruses [5],[17] as well as the broad distribution of mitochondrial cell death suppressors in other viruses suggests this novel serine protease pathway may occur in other biological settings.
The HtrA2 protease has MEROPS accession number S01.278 and the I.M.A.G.E. consortium clone obtained for these studies was identical in sequence to NCBI accession ID BC0000096. The vMIA [2] used to complement and repair ΔUL37x1 was obtained from AD169varATCC genomic DNA; NCBI accession ID X17403. The sequence of Towne-BAC was deposited to NCBI [8]; accession ID AY315197.
Human fibroblasts (HFs), vMIA-HFs, vMIAmut-HFs, and HeLas were cultured as previously described [5]. Viruses derived from the BACmid clones Towne-BAC and ΔUL37x1 [8] were maintained as DNA clones in E. coli or on complementing vMIA-HFs prior to experiments. AD169varATCC was maintained as previously described [22].
The kanamycin selection cassette in ΔUL37x1 was replaced with UL37x1 sequence derived from AD169varATCC to generate RC2707. Transfection of pON2707 [5] into HFs was followed by infection with ΔUL37x1 virus. Plaques that included cell death at a frequency similar to Towne-BAC virus [5] were isolated for further analysis. Sequencing of viral DNA from Towne-BAC and two independently derived isolates (ΔUL37x1R1, ΔUL37x1R2) confirmed replacement of the selection marker in ΔUL37x1 with UL37x1 nucleotide sequence identical to pON2707 and AD169varATCC while the control, Towne-BAC, was identical to the expected sequence [8],[98]. Expression of vMIA was confirmed by immunoblot analysis.
The HtrA2/Omi expression plasmid, pON601, was derived by restriction of the I.M.A.G.E. cDNA (HtraA2 clone #5344667, ATCC, Manassas, VA) with BsrGI, followed by removal of the single-stranded overhangs with Klenow DNA polymerase, and restriction with XhoI. The HtrA2/Omi encoding fragment was ligated to EcoRV and XhoI restricted pcDNA3.1+ (Invitrogen, San Diego, CA). The HtrA2S306A expression plasmid, pON602, was generated by PCR-site directed mutagenesis of the HtrA2 ORF to introduce the S306A mutation and a novel NaeI restriction enzyme site and utilized the mutagenic primer 5′-CTATTGATTTTGGAAACGCCGGCGGTCCCCTGGTTAAC-3′. Both clones were sequenced to confirm expected results. The vMIA and GFP expression clones and retroviral constructs used in these experiments were reported previously [2],[5],[18].
Immunodetection employed mouse monoclonal antibodies to c-myc epitope (9E10; Santa Cruz Biotechnology, Santa Cruz, CA), HtrA2/Omi (MAB1458; R&D Systems, Inc, Minneapolis, MN), cytochrome c (Clone 7H8.2C12, BD Pharmingen, San Jose, CA), β-actin (AC-74, Sigma, St. Louis, MO), golgin-97 (CDF4; Molecular Probes, Eugene, OR), mitochondrial heat shock protein 70 (mtHSP70) (a gift from Susan Pierce, Northwestern University), viral nuclear antigens IE1p72 and IE2p86 (MAB 810, Chemicon, Temeculah, CA), ICP36 (CH16) and pp28 (CH19) (both from Virusys Corporation, Randallstown, MD) or rabbit polyclonal antiserum to native vMIA [2] and peroxidase-conjugated horse anti-mouse IgG or goat anti-rabbit IgG, Texas Red-conjugated horse anti-mouse IgG (all from Vector, Burlingame, Calif.), or AlexaFluor 350-conjugated goat anti-mouse IgG (Molecular Probes, Eugene, OR). Immunoblot analysis of total protein from infected cells and immunofluorescence assays followed previously described methods [22]. MitoTracker Red CMXRos (Molecular Probes, Eugene, OR) staining of mitochondria followed previously described methods [22].
To assess morphological changes in infected cells and nuclei, cells grown on coverslips and infected for varying periods of time were fixed with 3.7% formaldehyde, permeabilized with Triton X-100 (EMD Biosciences, Darmstadt, Germany), stained with Hoechst 33258 (AnaSpec, San Jose, CA), and processed for microscopic evaluation as previously described [22]. Some cultures were stained with ethidium homodimer 1 (Molecular Probes, Eugene, OR), as previously described [5] to assess virus-induced cell death.
Images from live cell cultures were obtained as previously described [5] or utilized Simple PCI software, a Retiga Exi digital camera, and a Leica DM IRB microscope. Imaging of cultures grown on coverslip employed an AxioCam MRc5 camera attached to a Zeiss Axio Imager.A1 and AxioVision Release 4.5 software.
Replication inhibitors included phosphonoformate (PFA Sigma, St. Louis, Mo) dissolved in water and 2-bromo-5,6-dichlorobenzimidazole (BDCRB from L. B. Townsend, University of Michigan) dissolved in dimethyl sulfoxide (DMSO) (Sigma, St. Louis, MO). Protease inhibitors included TLCK, N-alpha-p-tosyl-L-lysine chloromethyl ketone, (Sigma, St. Louis, MO) in water, and UCF-101 or zVAD.fmk (both from Calbiochem, La Jolla, CA) dissolved in DMSO. Inhibitors were added by replacing culture medium with medium containing inhibitor while control medium included the appropriate solvent (DMSO) or no addition. Morphology and presence of viral nuclear antigens IE1p72 and IE2p86 were assessed as described above and viral yield was determined from total virus recovered on day 7 from supernatant and sonicated cells [5]. DMSO does not impact CMV death or CMV replication levels at the concentrations used (≤0.1%)[5].
Transfections of BACmid DNA have been described [5]. GFP-positive (GFP+) cells and GFP-positive foci (>2 GFP+ cell) were evaluated by live cell microscopy 2–10 days post transfection. Viral presence was confirmed by immunodection of viral nuclear antigens IE1p72 and IE2p86 and some experiments utilized a plasmid encoding GFP [5] for detection of transfected cells. Reported results were obtained from multiple DNA preparations.
Conditions for induced apoptosis of ADvarATCC infections and vMIA-dependent survival following transfection of HeLa have been described [2],[64]. Cell numbers were determined following Hoechst stain of surviving cells (HeLas) or following immunodetection of viral nuclear antigens IE1p72 and IE2p86 (ADvarATCC), comparing to untreated controls.
To assess the impact of HtrA2/Omi on HFs, GFP expression plasmid was cotransfected with control DNA (vector) or HtrA2/Omi or HtrA2S306A expression plasmids. By 48 h cells had recovered and were confluent. Images obtained of GFP fluorescence at 24 h intervals between 48–120 h post transfection were evaluated for numbers of GFP+ cells from 12 microscopic fields per day. Mean % survival (±standard deviation) was calculated from numbers of GFP+ cells compared to those at 48 h.
Sulforhodamine 101-leucine chloromethyl ketone, SLCK, (Immunochemistry Technologies, LLC, Bloomington, MN) was suspended in DMSO (Sigma, St. Louis, MO). For labeling, 2.5 µM SLCK was applied in the presence or absence of TLCK to live cultures for 30 minutes prior to fixation and imaging as described above. |
10.1371/journal.pgen.1001383 | Identification and Functional Validation of the Novel Antimalarial
Resistance Locus PF10_0355 in Plasmodium
falciparum | The Plasmodium falciparum parasite's ability to adapt to
environmental pressures, such as the human immune system and antimalarial drugs,
makes malaria an enduring burden to public health. Understanding the genetic
basis of these adaptations is critical to intervening successfully against
malaria. To that end, we created a high-density genotyping array that assays
over 17,000 single nucleotide polymorphisms (∼1 SNP/kb), and applied it to
57 culture-adapted parasites from three continents. We characterized genome-wide
genetic diversity within and between populations and identified numerous loci
with signals of natural selection, suggesting their role in recent adaptation.
In addition, we performed a genome-wide association study (GWAS), searching for
loci correlated with resistance to thirteen antimalarials; we detected both
known and novel resistance loci, including a new halofantrine resistance locus,
PF10_0355. Through functional testing we demonstrated that
PF10_0355 overexpression decreases sensitivity to
halofantrine, mefloquine, and lumefantrine, but not to structurally unrelated
antimalarials, and that increased gene copy number mediates resistance. Our GWAS
and follow-on functional validation demonstrate the potential of genome-wide
studies to elucidate functionally important loci in the malaria parasite
genome.
| Malaria infection with the human pathogen Plasmodium falciparum
results in almost a million deaths each year, mostly in African children.
Efforts to eliminate malaria are underway, but the parasite is adept at eluding
both the human immune response and antimalarial treatments. Thus, it is
important to understand how the parasite becomes resistant to drugs and to
develop strategies to overcome resistance mechanisms. Toward this end, we used
population genetic strategies to identify genetic loci that contribute to
parasite adaptation and to identify candidate genes involved in drug resistance.
We examined over 17,000 genetic variants across the parasite genome in over 50
strains in which we also measured responses to many known antimalarial
compounds. We found a number of genetic loci showing signs of recent natural
selection and a number of loci potentially involved in modulating the
parasite's response to drugs. We further demonstrated that one of the novel
candidate genes (PF10_0355) modulates resistance to the
antimalarial compounds halofantrine, mefloquine, and lumefantrine. Overall, this
study confirms that we can use genome-wide approaches to identify clinically
relevant genes and demonstrates through functional testing that at least one of
these candidate genes is indeed involved in antimalarial drug resistance.
| Plasmodium falciparum malaria is a major public health challenge
that contributes significantly to global morbidity and mortality. Efforts to control
and eliminate malaria combine antimalarial drugs, bed nets and indoor residual
spraying, with vaccine development a longer-term goal. Genetic variation in the
parasite population threatens to undermine these efforts, as the parasite evolves
rapidly to evade host immune systems, drugs and vaccines. Studying genetic variation
in parasite populations will expand our understanding of basic parasite biology and
its ability to adapt, and will allow us to track parasites as they respond to
intervention efforts. Such understanding is increasingly important as countries move
towards reducing disease burden and the ultimate elimination of malaria.
Given the potential impact of rapid evolution of P. falciparum in
response to control and eradication strategies, discovery and characterization of
P. falciparum genetic diversity has accelerated in recent
years. Since the first malaria genome was sequenced in 2002 [1], over 60,000 unique SNPs have
been identified by concerted sequencing efforts [2]–[4], and several genomic tiling arrays
[5]–[9] and low-density SNP arrays [10], [11] have been developed to query this
genetic variation. Recently the first malaria GWAS was published [11], in which 189
drug-phenotyped parasites from Asia, Africa and the Americas were genotyped using a
low-density array (3,257 SNPs); that study identified loci under positive selection
and found several novel drug resistance candidates.
For our study, we adopted two strategies for identifying genes involved in the
malaria parasite's adaptive response: searching for signals of recent or
ongoing natural selection, and searching for loci associated with one important
clinical adaptation—resistance to antimalarial drugs. To make these searches
possible, we began by sequencing 9 geographically diverse strains of P.
falciparum to identify novel variation, thereby doubling the number of
publicly available SNPs to 111,536 (all accessible at plasmodb.org), and used these
SNPs to develop a high-density genotyping array assaying 17,582 validated markers.
After characterizing linkage disequilibrium and population structure in our samples,
we used the arrays to search for evidence of both ongoing balancing selection and
recent positive selection, and to carry out a GWAS that sought loci associated with
resistance to thirteen antimalarial agents. We then followed up one of the novel
loci associated with drug resistance in order to verify that variation there was
biologically involved in modulating drug response.
We identified global population structure among malaria parasites using principal
components analysis (PCA) of 57 genotyped culture-adapted parasite samples
(Figure 1A, Table S1,
Figure
S1). African, American and Asian samples form three distinct
clusters, reflecting the likely independent introduction of P.
falciparum from Africa into Asia and the Americas. There was little
evidence for structure within Africa, suggesting high gene flow throughout the
region (Figure
S1). Asian and American parasites however both show substantial
internal structure.
There are also dramatic differences in linkage disequilibrium (LD) between
populations, with substantial LD extending less than 1 kb in Senegal, 10 kb in
Thailand, and 100 kb in Brazil (Figure S2). These observations are consistent
with previous findings, which showed that LD decays more rapidly in Africa, due
either to founder effects in other continents [12] or to elevated
outcrossing frequencies in Africa [12], [13], where higher transmission
intensity leads to a greater likelihood of sexual outcrossing rather than
selfing within the mid-gut of vector mosquitoes.
The short LD in malaria, driven by high levels of recombination, means that a
high density of markers is required to identify candidate loci in association
studies, since causal variants not on the array can seldom be tagged by
neighboring alleles (Table S2). On the other hand, short LD can
aid in fine-mapping candidate associations and greatly accelerates the search
for causal genes. Short LD also aids in identifying genomic regions under recent
positive selection with recombination-based methods, since the increased LD in
selected regions should stand out against the short-LD background.
We expect that many parasite proteins that interact with the host immune system
will be under balancing selection, because they will be under selective pressure
to maintain high levels of diversity. Indeed, previous studies have shown that
regions of the P. falciparum genome that are highly polymorphic
and appear to be under balancing selection encode antigens that are recognized
by the human immune system [4]. We examined evidence for balancing selection in our
data by searching for regions with high nucleotide diversity (as measured by SNP
π) and low population divergence (as measured by FST) (Figure 1B). When we examined
the loci lying in this region of the graph (Figure S3),
we found a number of known antigens and vaccine candidates. Loci in the same
region with unknown function are thus potential novel antigens that trigger
human immune response to malaria, and may prove useful as biomarkers or as
candidate vaccine molecules.
We carried out a similar search for loci under positive selection by identifying
regions with both low nucleotide diversity within Senegal and Thailand and high
population divergence between the two populations (Figure 1B). We observed throughout the genome
that nucleotide diversity was lower for nonsynonymous SNPs than for intergenic
SNPs (Figure
S4), a characteristic result of widespread purifying selection. At
the same time, nonsynonymous SNPs exhibited significantly greater divergence
than intergenic SNPs in all pairwise population comparisons, suggesting the
effect of positive selection in local P. falciparum
populations. Nonsynonymous SNPs with low diversity within a population and high
divergence between the two populations studied may represent polymorphisms
responsible for adaptive evolution.
We also carried out a genome-wide scan for recent positive selection using the
long-range haplotype (LRH) test [14], which identifies common variants that have recently
spread to high prevalence using recombination as a clock. Approximately 15 genes
were identified as having undergone recent positive selection by this approach,
including known drug resistance loci (pfcrt and
dhfr) as well as multiple members of the acyl-CoA
synthetase (ACS) and ubiquitin protein ligase families (Figures S5
and S6);
these latter genes also exhibit high divergence between Senegal and Thailand
(Figure 1B), evidence
for selection that is recent and population-specific. Taken as a group, the
genes identified by the LRH test show a significant enrichment for higher than
average population divergence (as measured by FST, Mann-Whitney
U = 1583, P = 0.0071). All of these
loci (Table
S3, Dataset S1), which include genes in the folate metabolism, lipid
biosynthesis and ubiquitin pathways, should be viewed as candidates for
functional follow-up and further characterization.
In order to directly assess the genetic basis for one important response to
antimalarial intervention, we carried out a GWAS to identify loci associated
with drug resistance in parasites. This same approach can potentially be applied
to many other clinically relevant malaria phenotypes, e.g. host response,
invasion, and gametocyte formation. Our first step was to measure drug
resistance (IC50 values) to 13 antimalarial drugs (amodiaquine,
artemether, artesunate, artemisinin, atovaquone, chloroquine,
dihydroartemisinin, halofuginone, halofantrine, lumefantrine, mefloquine,
piperaquine and quinine) in 50 culture-adapted parasites using a high-throughput
assay (Tables
S4 and S5, Text S1, Dataset
S1).
We performed the genome-wide association analysis using two statistical tests:
efficient mixed-model association (EMMA) and a haplotype likelihood ratio (HLR)
test (Figures
S7, S8, S9, S10, Methods). EMMA identifies
quantitative trait associations in individuals with complex population structure
and hidden relatedness; it has previously been shown to outperform both
PCA-based and λGC-based correction approaches in highly inbred
and structured mouse, maize, and Arabidopsis populations [15]. EMMA is particularly
applicable for small and structured sample sets: one of its first applications
was in a study of 24 diploid mouse strains [15], essentially the same sample
size as in our study (50 haploid strains). The HLR test is a multi-marker test
designed to detect the association of a single haplotype with a phenotype, and
is particularly powerful when the associated haplotype experienced recent strong
selection (and is therefore long) and occurs on a low-LD background [16]; it is
therefore particularly appropriate for this study. We addressed the effect of
population structure in the HLR test using population-specific permutation
(Methods). When used together, these
two complementary approaches provide a highly sensitive screen for association
signals within the P. falciparum genome.
The well-characterized chloroquine resistance locus, pfcrt,
served as a positive control for our GWAS methods (Figure 2A and 2C, Table S2),
an important test given our small sample size and the limited LD present in
P. falciparum. As expected, we found evidence for
association with resistance to chloroquine using both tests, consistent with
previous studies [11]; EMMA yielded evidence for association with
genome-wide signficance, while the signal from the HLR test fell just short of
genome-wide significance (Figure
2C).
Applying the same tests to the other drug phenotypes, we detected numerous novel
loci showing significant associations with drug resistance (Figure 2A and 2D, Table 1). Quantile-quantile plots for each
test demonstrate that we were able to effectively control for population
structure (Figure 2B).
Despite our small sample size and the low LD in P. falciparum,
in total eleven loci achieved genome-wide significance for association with
resistance to five different drugs: amodiaquine, artemisinin, atovaquone,
chloroquine and halofantrine. In most cases, the short extent of LD allowed
localization to individual genes. Among the loci identified were various
transporters and membrane proteins, as well as five conserved genes with unknown
function (Table 1, Dataset
S1).
Demonstrating that a signal of association actually reflects a causal molecular
process requires functional testing and validation of the candidate locus, both
because of concerns about power and reproducibility of genetic association
tests, and because even a robust statistical correlation need not imply
biological causation. To confirm the ability of GWAS to identify functionally
relevant candidates, we investigated one of our association findings,
PF10_0355, in greater depth. This gene contains multiple
SNPs associated with halofantrine resistance (Figure 2D), and encodes a putative
erythrocyte membrane protein (PlasmoDB.org) characterized by high genetic
diversity.
We set out to determine the role of PF10_0355 in halofantrine
resistance by transfecting halofantrine-sensitive Dd2 parasites with episomal
plasmids containing the PF10_0355 gene from a
halofantrine-resistant parasite (SenP08.04), a technique that is used routinely
for stable transgene expression [17]. Two independent transfectants overexpressing the
PF10_0355 gene from SenP08.04 both showed reduced
susceptibility to halofantrine when compared with the Dd2 parent or a
transfection control (Figure
3A), suggesting that this gene is indeed involved in modulating
parasite drug response.
Two independent transfectants overexpressing the endogenous
PF10_0355 gene from halofantrine-sensitive Dd2 also showed
reduced susceptibility to halofantrine (Figure 3A), however, pointing to a role of
overexpression in the observed resistance. Because PF10_0355 is
annotated as a putative erythrocyte membrane protein and belongs to the
merozoite surface protein 3/6 family, we tested the hypothesis that the observed
effect was the by-product of a growth or invasion-related process, rather than
resistance due to a direct interaction with the antimalarial itself. To that
end, we expanded our drug testing in the transfectant lines to include other
antimalarials, some structurally related and some unrelated to halofantrine.
Overexpression of PF10_0355 from either the Dd2 or the SenP08.04
parent caused increased resistance to the structurally related antimalarials
mefloquine and lumefantrine (Figure
3B and 3C), but had no effect on parasite susceptibility to the
structurally unrelated antimalarials chloroquine, artemisinin or atovaquone
(Figure 3D and 3E).
Indeed, we found evidence of cross-resistance between halofantrine and both
mefloquine and lumefantrine (Figure
4). We also observed cross-resistance between halofantrine and
artemisinin, which is expected as cross-resistance between aminoquinolines and
artemisinin compounds has been previously demonstrated [11], [18] and resistance to all
these drugs has been shown to be mediated by changes in pfmdr1
copy number [19], [20]. Overexpression of PF10_0355,
however, alters parasite susceptibility to the aminoquinolines but not to
artemisinin, suggesting that this effect is specific for that set of
structurally related compounds and distinct from the effect of
pfmdr1, which seems to exert a global effect of resistance
to unrelated compounds (i.e. both aminoquinolines and artemisinins). Using the
Dd2 parasite line, which has amplified pfmdr1 copy number, as a
background for PF10_0355 overexpression allowed us to
distinguish between cross-resistance to a structurally related class of
compounds (mediated by PF10_0355 overexpression) and
pan-resistance to multiple classes of drugs.
Given that overexpression of the PF10_0355 gene both from a
halofantrine-resistant and from a sensitive parasite conferred resistance to
halofantrine-related drugs, we investigated whether gene amplification might be
driving the observed resistance, as it often does for antimalarial drugs [21]–[26]. We
quantified PF10_0355 copy number in our transfectants and found
that the transfectant with the highest IC50 for all three drugs
(Dd2+P08B) also had the highest PF10_0355 copy number, as
measured by quantitative PCR (qPCR) (Figure 5A). Furthermore, when we examined the
PF10_0355 gene on our SNP array, we detected a substantial
increase in hybridization intensity at the PF10_0355 locus
compared to the genome average, suggesting that this gene is amplified in some
parasites (Figure 5B). The
amplified region appears only to contain the PF10_0355 gene
itself and not surrounding loci. We observed a similar pattern at
pfmdr1 on chromosome 5, where copy number variation is well
established (Figure S11). Follow-up qPCR analysis of 38 parasite lines confirmed
that parasites with amplified PF10_0355 have a greater mean
halofantrine IC50. (Figure 5C, Table S6, Dataset
S1). Copy number variation was further confirmed in a number of parasites
by quantitative Southern blotting (Figure S12).
In this study we used natural selection and genome-wide association methods to probe
the genetic basis of adaptation in P. falciparum. These approaches
are complementary: scanning for selected loci permits an unbiased search for unknown
adaptive changes, but provides little information about the processes at work, while
GWAS gives a focused look at one easily identified (and clinically critical)
adaptive phenotype. Results from both approaches open up new avenues for study, as
we seek to understand the biological significance of the findings.
The specifics of our strategy were designed to cope with two potential limitations in
applying genome-wide population genetic approaches to malaria: small sample sizes,
due to the difficulty in adapting parasites to culture and assessing drug and other
phenotypes; and a lack of correlation (LD) between nearby variants in the parasite
genome, which limits our ability to infer untyped SNPs from genotyped markers. The
second limitation we addressed by developing a high-density genotyping array (based
on new sequencing), to increase the fraction of genetic variation that we could
directly interrogate, while the effect of the first was mitigated by the phenotype
we targeted in our GWAS.
Drug resistance is a phenotype well-suited for GWAS because it is expected to be
caused by common alleles of large effect at few genomic loci [27]. If this is the case,
associations will be much easier to detect than in a typical human GWAS, in which
the phenotype is caused by alleles at many loci that are either rare or of small
effect. Additionally, the haploid nature of the intra-erythrocytic stage of
P. falciparum further heightens GWAS power by eliminating the
issue of allelic dominance. Finally, the increased LD caused by recent selection for
drug resistance counteracts the loss of power that comes from short LD, small sample
size, and the temporal and geographic stratification of the parasite population that
we examined. Thus, despite the potential limitations, we were able to detect a known
drug resistance locus (pfcrt), observed little
P-value inflation in our GWAS data (Figures S8,
S9, S10), and
identified a number of genome-wide significant loci associated with drug resistance.
Part of this success was likely due to specific tests we used to account for
population structure.
Going beyond these statistical tests, we went on to functionally validate one of
these loci, demonstrating that increased PF10_0355 copy number
confer resistance to three structurally related antimalarial drugs. This
demonstrates the feasibility of coupling GWAS and functional testing in the malaria
parasite for identifying and validating novel drug resistance loci and illustrates
the power of GWAS to find functionally important alleles.
Comparing our results to the recent GWAS described by Mu, et al. [11], which was also
directed at finding drug-resistance loci, we see that, beyond the well-known
pfcrt locus, there was no overlap between the associations
identified by each study. Differing sets of drugs tested and analytical methods
explain much of the disagreement. Of the eleven candidate associations in Table 1, one (that with
pfcrt) was found by both studies, eight were associations with
drugs not assayed in Mu, et al. (atovaquone and halofantrine), and two were found
only with a haplotype-based test, an approach not used by Mu, et al. Our candidate
locus at PF10_0355, in fact, would not have been detectable in the
Mu study because it was identified only by the multi-marker HLR test, because it
involved an association with halofantrine, and because the Mu, et al. genotyping
array lacked markers within 4 kb of the gene (plasmoDB.org).
Different parasite populations and marker sets probably explain many of the
dihydroartemisinin, mefloquine and quinine associations identified by Mu, et al. but
not seen in our data set. The studies used different parasite population
sets—theirs was weighted toward southeast Asian strains, and ours toward
African strains—and selection pressures and selected alleles can both vary
between populations. Our smaller sample size also means that we might lack power to
identify some associations accessible to Mu, et al. These difficulties are reflected
in human GWAS studies as well, where the ability to replicate associations using
multiple tests and in different sample sets has also been challenging to achieve
[28].
Ultimately, the disparities in loci identified point to the role of population
analysis as a tool for candidate gene discovery and not as a definitive study. Even
within each study, there is little overlap between the signals observed with
different methods—our study detects only one gene (pfcrt) by
both GWAS tests (EMMA and HLR), while Mu, et al. detected only two genes (unknowns,
not pfcrt) by both of their GWAS tests (Eigensoft and PLINK). Even
a well-designed GWAS serves only as a hypothesis-generating experiment, and it is
vital to empirically validate candidate loci associated with a phenotype of
interest. Especially given the small sample sizes and relatively sparse marker
density used in both malaria GWAS studies to date, functional validation of
candidates is necessary to address concerns about false positive results.
Our functional result, that increased PF10_0355 copy number confers
decreased susceptibility to halofantrine, mefloquine and lumefantrine, raises
additional questions for study. Further work will be needed to determine the precise
contributions of copy number variation and gene mutation to the parasite's
response to these drugs. The biological function of this gene's product is
unknown, but previous work indicates putative localization to the parasite surface
[29], as well
as it being a potential target of host immunity and balancing selection [30]. While the
protein itself does not appear to be a transporter, it is possible that it directly
binds drug or perhaps couples with transport proteins to modulate drug
susceptibility; interaction between membrane transporters and non-channel proteins
has been demonstrated in cancer, plant and yeast systems [31]–[33]. Additional
experiments are certainly required to determine the precise role of
PF10_0355 in modulating parasite response to this class of
compounds, including assessing its relevance to resistance in natural populations,
but it is clear that alteration of this locus can mediate drug resistance in
P. falciparum.
Although halofantrine, mefloquine and lumefantrine are not commonly used as primary
interventions, widespread halofantrine use has recently been documented in West
Africa. Notably, halofantrine was used to treat nearly 18 million patients between
1988 and 2005 [34], [35], and it remains in production and use today. Use of
halofantrine, mefloquine or lumefantrine as monotherapy may further explain how
mutations and copy number variation in the PF10_0355 gene were
selected. Lumefantrine is also currently used as a partner drug in the
artemisinin-based combination therapy (ACT) Coartem. The shorter half-life of
artemether allows lumefantrine to be present as monotherapy, making it vulnerable to
selection of drug resistant mutants. As genetic loci associated with drug responses
are identified and validated, these provide new molecular biomarkers to evaluate
drug use and response in malaria endemic settings. Thus, our findings have
implications for defining molecular biomarkers for monitoring partner drug responses
as intervention strategies, such as ACTs, are applied.
Beyond identifying a novel drug resistance locus, this study illustrates the general
utility of a GWAS approach for the discovery of gene function in P.
falciparum. Even with a small and geographically heterogeneous sample
of parasites, we identified a number of new loci associated with drug response and
validated one of them. Larger samples from a single population will have much
greater power to detect additional loci, including those where multiple and low
frequency alleles contribute to resistance. Future GWAS have the potential both to
provide greater insights into basic parasite biology and to identify biomarkers for
drug resistance and other clinically relevant phenotypes like acquired protection,
pathogenesis, and placental malaria.
Future GWAS will be able to counteract the loss of power caused by low LD, either by
focusing on parasite populations with reduced outcrossing rates, or by studying
cases of very strong selective pressure. This issue will soon become moot, however,
as the declining cost of whole-genome sequencing makes it practical to assay every
nucleotide in the genome on a routine basis. Culture-adapted parasites are amenable
to robust and reproducible phenotypic characterization, but their
limitations—the potential for artifactual mutations during adaptation and for
a biased selection of clones within a given infection—mean that genetic
changes identified using them require both functional validation and demonstration
that the changes are important during natural infection. As direct sequencing of
clinical isolates with demonstrable clinical phenotypes such as ex
vivo drug response or invasion properties becomes increasingly
feasible, sequencing will enable us to directly identify genetic changes in the
parasite associated with clinically relevant phenotypes. In the years ahead, genome
analysis of P. falciparum has the potential to identify genetic
loci associated with many phenotypes, enhance our understanding of the biology of
this important human pathogen, and inform the development of diagnostic and
surveillance tools for malaria eradication.
Parasite samples and origins are detailed in Text S1 and
Table
S1. Parasites were maintained by standard methods [36] and were
tested for their response to amodiaquine, artemether, artesunate, artemisinin,
atovaquone, chloroquine, dihydroartemisinin, halofuginone, halofantrine,
lumefantrine, mefloquine, piperaquine and quinine according to the methods
outlined by Baniecki, et al. [37] (Table S4, Figure S13,
Text
S1). Follow-up drug testing was done by measuring uptake of
3H-hypoxanthine [38]. Nucleic acids were obtained from parasite cultures
using Qiagen genomic-tips (Qiagen, USA). All DNA samples were evaluated by
molecular barcode [39].
We sequenced nine geographically diverse parasite isolates to 1.25x coverage,
nearly doubling the number of publicly available SNPs to 111,536 (Text S1).
These parasites had been previously sequenced to 0.25x coverage [2] and the
deeper sequencing allowed for more thorough SNP discovery. Using this combined
marker set, we created a high-density Affymetrix-based SNP array for P.
falciparum containing 74,656 markers. Arrays were hybridized to 57
independent parasite samples (Table S1), including 17 previously sequenced
strains used as a validation set. Genotype calls were produced using the BRLMM-P
algorithm [40]. Markers that did not demonstrate perfect concordance
between sequence and array data for the 17 strains were removed (Text S1).
The remaining 17,582 SNPs constituted the high-confidence marker set used
throughout this study (median marker spacing 444 bp, mean spacing 1,316 bp). All
genomic positions and translation consequences are listed with respect to the
PlasmoDB 5.0 assembly and annotation. SNP genotype data are publicly available
on plasmodb.org (release 6.0, July 2009) and dbSNP (Build B134, May 2011),
accessible by searching for submission batches Pf_0002 (sequencing of nine
isolates) and Pf_0003 (genotyping of 57 isolates) from submitter
BROAD-GENOMEBIO. Genotype data is also available as Dataset
S2.
Principal components analysis (PCA) was performed using the program SmartPCA
[41].
All single-infection samples were used for the analysis in Figure 1. Samples that tightly clustered with
the wrong continental population (A4, Malayan Camp and T2_C6) represented likely
cases of contamination and were thus omitted from all other analyses.
We measured diversity using a statistic we term ‘SNP π,’ which
quantifies the average number of pair-wise differences among samples from a
given population at assayed SNPs. Population divergence was measured using
FST, calculated using the method of Hudson, et al. [42].
Statistical evaluation of the significance of differences in SNP π and
FST among populations was performed using a bootstrapping
approach, where the SNP set was re-sampled with replacement and each statistic
recomputed 1000 times.
The statistic r2 was calculated within each population for all pairs
of SNPs sharing the same chromosome [43]; pairs were binned by
distance and averaged within each bin. The level of LD between unlinked markers
was estimated by calculating r2 between all pairs of SNPs on
different chromosomes. To determine the bias caused by small sample size, the
unlinked calculation was repeated, with the change that for each pair of SNPs,
the genotype for one was taken from one strain while the genotype for the second
was taken from another strain. This background value of r2 was
calculated separately for the possible pairs of different strains and then
averaged. Only single infections, as assessed by molecular barcode, were
used.
Because of the small number of samples, LRH results for individual continental
populations had a high level of variance. Thus, we pooled together samples from
Africa (n = 26) and Asia (n = 18,
excluding India), as suggested by our PCA analysis. SNPs included in the
analysis had a minor allele frequency of at least 0.05 and a call rate of at
least 0.8; missing genotypes were imputed using PHASE. LRH analysis was
performed using Sweep. Each SNP defined two core alleles, one base pair in
length. We calculated relative extended haplotype homozygosity (REHH) for each
core allele, to its left and right [44], yielding up to four REHH
scores per SNP locus. We standardized the REHH scores as a function of core
allele frequency, defined on a discrete grid from 0.05 to 0.95 with even spaces
of 0.025. This yielded a normally-distributed set of Z-scores
for which we calculated corresponding P-values and
Q-values.
We performed a GWAS for drug resistance to thirteen antimalarials across 50 of
our genotyped samples. 7,437 SNPs that had a minor allele count of five samples
as well as an 80% call rate under every phenotype condition were used for
GWAS. A Bonferroni significance threshold of
–log10(P-value) >5.17 was used for all
tests. See Text
S1 for more details on GWAS methods.
The Efficient Mixed-Model Association (EMMA) test [15] models quantitative trait
associations to a data set with complex population structure and hidden
relatedness. It calculates a genotype similarity matrix instead of discrete
categories and does not require a priori specification of
populations. The resulting P-value distributions demonstrate
little remaining effect from population structure (Figure S8)
while retaining power to find a number of associations at genome-wide
significance (Figure S8, Figure
2A, Table
1).
The Haplotype Likelihood Ratio (HLR) test [16] models the likelihood
that a single, resistant haplotype rose to dominance while all other haplotypes
proportionally decreased. PLINK [45] is used to produce sliding
window haplotypes across the genome and calculate haplotype frequencies for
input to the HLR test. We produced input for all 2-, 4- and 6-marker windows.
The LOD scores generated by the HLR test were converted to empirical pointwise
P-values by performing approximately 370,000 permutations
of the null model for each test condition, allowing us to calculate empirical
P-values up to a significance of 10−5.6.
We preserved population-specific phenotype frequencies by permuting only within
each of three populations defined by our PCA analysis (Table S1).
Resulting P-value distributions fit expectations well for the
vast majority of test conditions (Figures S9, S10) and
the test demonstrates power to detect a number of loci at genome-wide
significance (Figure 2A,
Table 1).
Copy number was assessed by evaluating the hybridization intensity at the
PF10_0355 locus on the high-density SNP array (Text S1).
Follow-up analyses were done by quantitative real-time PCR (qPCR) of the
PF10_0355 locus using the Delta Delta Ct method [46].
PF10_0355 was compared to the reference locus
PF07_0076 and 3D7 was used as a reference strain. A summary
of PF10_0355 copy number for all parasite strains tested is
provided in Table S6. Select resistant strains that were found to have multiple
copies of PF10_0355 were further analyzed by quantitative
Southern blotting and PF10_0355 copy number was compared to the
dhps gene from the 3D7 strain [47].
The full length ORF of PF10_0355 was amplified from either the
Dd2 (HFN sensitive) or SenP08.04 (HFN resistant) parasite isolate and cloned
into the pBIC009 plasmid under the expression of the Hsp86
promoter. Plasmid DNA was isolated, tranfected into the Dd2 parasite strain and
stable transfectants were selected with 2.5 nM WR99210 [48]. Parasites from two
independent experiments for each vector type (Dd2+Dd2 and
Dd2+SenP08.04) were isolated and successful transfection was confirmed by
plasmid rescue as well as episome-specific PCR and sequencing. Additionally, a
vector control strain was made by transfecting Dd2 parasites with the pBIC009
plasmid containing the firefly luciferase gene (EC 1.13.12.7).
|
10.1371/journal.pgen.1002011 | FUS Transgenic Rats Develop the Phenotypes of Amyotrophic Lateral Sclerosis and Frontotemporal Lobar Degeneration | Fused in Sarcoma (FUS) proteinopathy is a feature of frontotemporal lobar dementia (FTLD), and mutation of the fus gene segregates with FTLD and amyotrophic lateral sclerosis (ALS). To study the consequences of mutation in the fus gene, we created transgenic rats expressing the human fus gene with or without mutation. Overexpression of a mutant (R521C substitution), but not normal, human FUS induced progressive paralysis resembling ALS. Mutant FUS transgenic rats developed progressive paralysis secondary to degeneration of motor axons and displayed a substantial loss of neurons in the cortex and hippocampus. This neuronal loss was accompanied by ubiquitin aggregation and glial reaction. While transgenic rats that overexpressed the wild-type human FUS were asymptomatic at young ages, they showed a deficit in spatial learning and memory and a significant loss of cortical and hippocampal neurons at advanced ages. These results suggest that mutant FUS is more toxic to neurons than normal FUS and that increased expression of normal FUS is sufficient to induce neuron death. Our FUS transgenic rats reproduced some phenotypes of ALS and FTLD and will provide a useful model for mechanistic studies of FUS–related diseases.
| Amyotrophic lateral sclerosis and frontotemporal lobar degeneration are two related diseases characterized by degeneration of selected groups of neuronal cells. Neither of these diseases has a clear cause, and both are incurable at present. Mutation of the fus gene has recently been linked to these two diseases. Here, we describe a novel rat model that expresses a mutated form of the human fus gene and manifests the phenotypes and pathological features of amyotrophic lateral sclerosis and frontotemporal lobar degeneration. Establishment of this FUS transgenic rat model will allow not only for mechanistic study of FUS–related diseases, but also for quick development of therapies for these devastating diseases.
| Amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD) are two common neurodegenerative diseases [1], [2]. ALS is characterized by degeneration of motor neurons, denervation atrophy of skeletal muscles, and progressive paralysis of limbs [3], [4]. FTLD mainly affects cortical neurons and causes cortical dementia [5]. ALS patients may develop cortical dementia that overlaps with FTLD in pathology [2], [6]. ALS and FTLD share a common feature of pathology—ubiquitin-positive inclusion [7]–[10]. Although selective groups of neurons are primarily affected in each disease condition [2], increasing evidence suggests that ALS and FTLD may fall the same disease spectrum.
Fused in Sarcoma (FUS) has recently been linked to both ALS and FTLD [11], [12]. FUS is a highly conserved ribonucleoprotein that mainly resides in the nucleus while shuttling between the cytoplasm and the nucleus [13]–[15]. Fus was initially reported to translocate and fuse with one of several genes to form chimeric oncogenes in leukemia and liposarcoma [16], [17]. The N-terminus of the FUS protein is rich in glutamine, serine, and tyrosine residues, and may be responsible for transactivation activity of FUS oncogenic fusion [18], [19]. The C-terminal part of the FUS protein contains several structural motifs important for nucleic acid binding [18], [20], [21]. FUS may also play an important role in regulating mRNA [14], [22], [23]. Deletion of the fus gene results in chromosomal instability and perinatal death in inbred mice [24], but causes only male sterility in outbred mice [25]. FUS-positive inclusion is considered a hallmark of some sporadic FTLD [9], [26]. FUS, Tau, and TDP-43 are the important components of ubiquitinated proteins in FTLD, but exclude one another in ubiquitin-positive inclusion [8]–[10], [27]. Mutations in the fus gene segregate with ALS and FTLD [11], [12], [28], [29], implying a pathogenic role of FUS in these diseases.
Given the importance of FUS in human diseases, the consequences of mutation in the fus gene must be examined. Here we show that overexpression of a mutant, but not normal, human FUS in rats induced progressive paralysis resembling ALS. Mutant FUS transgenic rats developed severe axonopathy of motor neurons, denervation atrophy of skeletal muscles, and a substantial loss of cortical and hippocampal neurons. At advanced ages, normal FUS transgenic rats displayed deficits in spatial learning and memory, and a loss of cortical and hippocampal neurons. Neuronal loss was accompanied by ubiquitin aggregation and glial reaction. Our FUS transgenic rats recapitulated some features of ALS and FTLD.
To study the consequences of mutation in the fus gene, we created transgenic rats expressing the human fus gene with or without mutation (Table S1). Most mutations in the fus gene are a single amino acid alteration, as exemplified by the substitution of arginine for cysteine at residue 521 (R521C) that is identified in geographically unrelated patients [11], [12], [30]. We therefore chose R521C as an example of fus mutation for our transgenic studies. The ribonucleoproteins FUS and TDP-43 are both linked to ALS and FTLD [11], . FUS and TDP-43 are robustly and ubiquitously expressed in rodents during development [33], implying an important role for these genes in development. Constitutive expression of a mutant TDP-43 causes early death to transgenic founder rats [34], preventing transgenic lines from establishment. To overcome this potential difficulty, we used a tetracycline-inducible system to express human fus transgenes in a controlled manner [34]. From 26 transgenic founders carrying the normal (12 rats) or the mutant (14 rats) fus transgene, we established four transgenic lines (line number corresponding to copy number of the transgenes) that expressed human FUS, under tight control by Doxycycline (Dox), at substantial levels (Figure 1A, Figure S1, and Table S1). FUS transgenic rats were crossed with a CAG-tTA transgenic line to produce double transgenic offspring that expressed human FUS transgene in the absence of Dox [35]. Breeding female rats were given Dox in their drinking water until delivery such that expression of the fus transgenes would be recovered in the offspring after Dox withdrawal (Figures S1 and S2).
Immunoreactivity to human FUS was detected in the brain and spinal cord (gray and white matter) of FUS transgenic rats (Figure 1B, 1D, and 1E), but not in tissues of nontransgenic rats (Figure 1C and 1F). While transgenic rats of lines 16, 20, and 22 expressed human FUS at comparable levels (Figure 1A), only the mutant FUS transgenic rats (lines 16 and 22) developed paralysis resembling ALS (Figure 1G-1J and Videos S1 and S2). Similar disease phenotypes were observed in two independent lines of mutant FUS transgenic rats (Figure 1G–1J), suggesting that the disease phenotypes resulted from expression of the mutant fus gene.
Pathological analysis revealed that few motor neurons in the spinal cord were undergoing degeneration (Figure 2A–2E). Degenerating axons were detected in the dorsal corticospinal tracts (Figure 2G), the ventral roots (Figure 2I and 2M), and the dorsal roots (Figure 2K) of mutant FUS transgenic rats at paralysis stages. As a result of motor axon degeneration, groups of skeletal muscle cells were atrophied (Figure 2O), although there were some perimysial cells with small nuclei suggestive of inflammation. These pathological changes were not observed in nontransgenic rats (Figure 2A) and also not observed in age-matched, wild-type FUS transgenic rats (Figure 2B, 2D, 2F, 2H, 2J, 2L, and 2N) expressing human FUS at comparable levels (Figure 1A). Collectively, these findings suggest that mutation of the fus gene is pathogenic. Electromyography of the gastrocnemius muscle revealed fibrillation potential, a characteristic of denervation atrophy (Figure 2P). Confocal microscopy showed that a substantial number of neuromuscular junctions were denervated in paralyzed FUS transgenic rats (Figure 2Q and 2R). Through stereological cell counting, we estimated the number of spinal motor neurons and did not detect a significant loss of motor neurons, although a trend of neuron loss was observed in the mutant FUS rats at paralysis stages (Figure 2S). Our results suggest that degeneration of motor axons contributed to paralysis in the mutant FUS transgenic rats.
ALS and FTLD somewhat overlap in pathology [2], and mutation of the fus gene is linked to both ALS and FTLD [28], [29]. We therefore examined the pathology in the brains of mutant FUS transgenic rats. Through stereological cell counting (Figure S3), we detected a significant loss of neurons in the frontal cortex and dentate gyrus of mutant FUS transgenic rats at paralysis stages (Figure 3). This neuronal loss was not observed in age-matched, normal FUS transgenic rats of line 20, although they expressed human FUS at comparable levels (Figure 1A and Figure 3). While cortical neurons are the primary targets of degeneration in FTLD, hippocampal neurons could be affected particularly at advanced disease stages [36], [37]. Our results show that overexpression of mutant FUS induced a substantial loss of cortical and hippocampal neurons in FUS transgenic rats, a phenotype of FTLD in rat models.
FUS proteinopathy is a hallmark of some sporadic FTLD cases [9], [26]. How normal FUS is related to neurodegeneration in the disease remains to be examined. Our wild-type (line 20) and mutant (line 16) FUS transgenic rats expressed human FUS at comparable levels (Figure 1A), but only the mutant FUS transgenic rats developed paralysis at an early age (Figure 1G–1I). We further examined the normal FUS transgenic rats at advanced ages (Figure 4). Although the normal FUS transgenic rats were asymptomatic by the age of 1 year, they displayed a deficit in spatial learning and memory at the advanced age (Figure 4J and 4K). By stereological cell counting, we detected a moderate, but significant, loss of neurons in the frontal cortex and dentate gyrus of the normal FUS transgenic rats at advanced ages (Figure 4L and 4M). These findings suggest that increased expression of normal FUS is sufficient to induce neurodegeneration and that mutant FUS is more toxic to neurons than is normal FUS.
Ubiquitin-positive inclusion is a hallmark of ALS and FTLD [8]–[27]. Accumulated ubiquitin was detected in the cortex (Figure 5D–5F) and spinal cord (Figure 5J–5L) of mutant FUS transgenic rats at paralysis stages, but was not detected in the tissues of age-matched normal FUS transgenic rats (Figure 5A–5C and 5G–5I). In the normal FUS transgenic rats, ubiquitin aggregates were observed only when neuronal loss was detected at advanced ages (Figure 4), suggesting that ubiquitin aggregation accompanied neurodegeneration. Ubiquitin inclusions were detected only in FUS-expressing cells, but were not colocalized with FUS (Figure 4G–4I and Figure 5). Ubiquitinated aggregates were positive for the mitochondrial marker COXIV (Figure S4), suggesting that damaged mitochondria may be ubiquitinated for degradation. No typical FUS inclusion was detected in FUS transgenic rats (Figure 1B and 1E, and Figure 5). FUS mainly resided in the nucleus, but was also diffusely located in the cytoplasm (Figure 1E). The C-terminus of FUS contains a nuclear localization signal that is necessary for the nuclear import of FUS. Most mutations occur within the C-terminus of FUS and disrupt this nuclear localization signal [38], leading to cytoplasmic accumulation of FUS. The R521C mutation tested in our transgenic studies affects FUS distribution to a minimal extent [38], and may be less potent in eliciting redistribution and aggregation of FUS in transgenic rats. Glial cells are key players in neurodegeneration [39]. Here we found that astrocytes and microglia proliferated in the brain (Figure 6A–6F) and spinal cord (Figure 6H–6K) of FUS transgenic rats at paralysis stages. Our results indicate that neurodegeneration was accompanied by ubiquitin aggregation and glial reaction.
ALS and FTLD are two related neurodegenerative diseases [2], [6] and may fall within the same disease spectrum. While a subset of FTLD patients develop motor neuron disease [40], ALS patients may develop the symptoms and pathology of FTLD [41]–[43]. FUS and TDP-43 are two ribonucleoproteins and their mutant forms are linked to both ALS and FTLD [7]–[29]. We obtained two FUS transgenic lines expressing a mutant or normal human fus transgene at comparable levels. Transgenic rats expressing a mutant FUS developed progressive paralysis secondary to axonal degeneration and displayed a substantial loss of neurons in the cortex and hippocampus, reproducing some phenotypes of ALS and FTLD. While the mutant FUS transgenic rats developed some phenotypes of ALS and FTLD, the age-matched normal FUS transgenic rats were asymptomatic. Our findings in FUS transgenic rats confirm that mutation of the fus gene is related to these two diseases and suggest that mutation of the fus gene is pathogenic.
FUS proteinopathy characterizes a subset of sporadic FTLD, in which ubiquitin-positive inclusions are negative for TDP-43 and tau but positive for FUS protein [27], [44]. However, it is not known how normal FUS is related to neurodegeneration in these diseases. While overexpression of mutant FUS induced severe phenotypes in young animals, overexpression of the normal FUS also induced neuron death as well as learning and memory deficits in aged rats. Mutated FUS appeared more toxic in transgenic rats, but an increase in the expression or function of the fus gene may elicit neurotoxicity. The effects of gene mutation include gain-of-function, loss-of-function, and dominant-negative effects. Overexpression of either the mutant or wild-type FUS induced disease phenotypes in transgenic rats, suggesting that mutation of the fus gene may cause the disease by a gain of toxic properties. Since gain-of-function and dominant-negative mutations can induce similar effects in transgenic models, more sophisticated genetic approaches, such as gene knockin, may be required for determining the nature of FUS mutations.
FUS and TDP-43 show a similarity in disease induction. Mutant forms of these genes are more toxic than the normal genes [34], and increased expression of the normal genes is sufficient to induce neurodegeneration [45], [46]. Both FUS and TDP-43 are ribonucleoproteins and may have overlapping functions. Indeed, FUS and TDP-43 are found in one protein complex regulating HDAC6 mRNA [47]. Like TDP-43, FUS predominantly resides in the nucleus, but also shuttles between the nucleus and the cytoplasm to perform multiple functions [13]. Similar to results for TDP-43 [34], we found that FUS was diffusely located in the cytoplasm in transgenic rats. Possibly, redistribution of FUS may alter the functions of this multifunctional protein, incurring cellular toxicity.
In summary, our results suggest that mutant FUS is more toxic to neurons than normal FUS and that increased expression of normal FUS is sufficient to induce neuron death. Our FUS transgenic rats reproduced some phenotypes of ALS and FTLD. The establishment of these FUS transgenic rat lines will allow for more detailed studies of FUS-related diseases.
Animal use followed NIH guidelines. The animal use protocol was approved by the Institutional Animal Care and Use Committees (IACUC) at Thomas Jefferson University.
The open reading frame (ORF) of the normal human fus gene was PCR-amplified from a human cDNA pool (Invitrogen) and the mutation was introduced by site-directed mutagenesis (Stratagene). The normal and mutant human FUS ORF were inserted downstream of the TRE promoter as described previously [34]. Linearized transgenic DNA was purified from agar gel and injected into the pronuclei of fertilized eggs of Sprague-Dawley (SD) rats to produce transgenic founder rats [34], [35]. Transgenes were maintained on the SD genomic background and were identified by PCR analysis of rat's tail DNA.
Grip strength of the rat's fore and hind paws was measured twice a week (Columbus Instruments) and used for determining disease onset and progression. Disease onset was defined as an unrecoverable reduction in the grip strength of fore or hind paws. Disease end-stage was defined as paralysis in two or more legs or as a 30% reduction in body weight.
Spatial learning and memory tasks were examined with a Barnes Maze (Med Associates). Compared to Morris Water Maze or Radial Arm Maze, the Barnes Maze not only avoids dietary restriction and intense stress, but also gives comparable results on rodent's spatial learning and memory tasks [48], [49]. The Barnes Maze consists of a white, acrylic, circular disk (122 cm diameter) with 18 holes (9.5 cm diameter) spaced every 20° and a high stand (140 cm height) supporting the disk that is designed to discourage animals from jumping to the floor. Rats were given one training session and four test sessions for 5 consecutive days. During training or testing sessions, rats were placed in the same initial orientation inside a transparent cylinder (start box) that was located at the center of the maze disk and the rats remained in the start box for 1 minute such that a standard starting context was ensured. When a lamp above the maze was turned on to make the surface of the maze aversive, the start box was removed to allow the animal to escape the maze surface by locating and crawling through the correct hole under which a black safe box was located. When the animal entered the safe box, the light was turned off and the safe box was covered with a black sheet. The animal was allowed to stay in the safe box for 1 minute before it was placed back to its home cage. Before training, each rat was given 2 minutes to explore the maze and then placed inside the safe box for 1 minute for habituation. During training, each rat was guided to the safe box twice and then given two trials to locate the safe box by itself. During the test, rats were placed inside the start box for 1 minute to locate the fixed safe box. The number of incorrect hole pokes (error) and the latency to locate the safe box were recorded. An incorrect hole poke was indicated when an animal closely approached and visually inspected a wrong hole. Latency to locate the safe box was calculated from the time testing started to the time when the animal entered, or its four paws touched, the safe box. The maze was wiped clean with 70% ethanol and then with dry paper towel after each test to prevent animals following odor trails.
An antibody to human FUS was produced by immunizing rabbits with a synthetic peptide (Genemed): (N-terminal)-SYGQPQSGSYSQQPS. Antiserum was affinity-purified with a peptide-affinity column (Pierce).
Anesthetized rats were transcardially perfused with 4% paraformaldehyde (PFA) dissolved in 1X PBS buffer and tissues were dissected after perfusion. Tissues were cryopreserved in 40% sucrose and cut into sections on a Cryostat. Tissue sections of 12 µm were immunostained with the following antibodies: rabbit polyclonal antibody to human FUS (made in-house), chicken antibody to ubiquitin (Sigma), mouse monoclonal antibodies against Iba-1 (Wako Chemical) or GFAP (Millipore), and mouse monoclonal antibody against NeuN (Millipore). For histochemistry, immunostained sections were visualized with an ABC kit in combination with diaminobenzidine (Vector) and counterstained with hematoxylin to display nuclei. For immunofluorescent staining, tissue sections were incubated first with specific primary antibodies and then with secondary antibodies labeled with fluorescent dyes (Jackson Immunoresearch). Primary antibodies were diluted at 1∶1000 and secondary antibodies diluted at 1∶200. The primary antibodies were incubated overnight at 4°C and the secondary antibodies were incubated for 2 hours at room temperature. For detection of degenerating neurons, paraffin-embedded spinal cords were cut into transverse sections of 10 µm and stained using a protocol for Bielschowsky silver staining [34].
As described in a previous publication [34], neuromuscular junctions (NMJ) were visualized by immunofluorescent staining and confocal microscopy. PFA-fixed gastrocnemius muscles were cut into sections of 100 µm on a Cryostat. Muscle sections were incubated with α-bungarotoxin (Invitrogen) for 30 minutes at room temperature and subsequently immunostained with mouse monoclonal antibodies to neurofilament (Sigma) and synaptophysin (Millipore). Both primary and secondary antibodies were diluted at 1∶1000. The primary antibodies were incubated overnight at 4°C and the secondary antibodies were incubated for 2 hours at room temperature. NMJ images were captured with a Zeiss LSM510 META confocal system and the NMJ was reconstructed through z-stack projections from serial scanning every 1 µm.
As described previously [34], anesthetized rats were perfused with a mixture of 4% PFA plus 2% glutaraldehyde. Cervical spinal cords and L3 ventral and dorsal roots were dissected and post-fixed in the same fixative at 4°C overnight. Fixed tissues were embedded in Epon 812 (Electron Microscopic Sciences, PA) and cut into semithin and thin sections. Semithin sections (1 µm) were stained with 1% toluidine blue and visualized under a light microscope. Thin sections (80 nm) were stained with uranyl acetate and lead citrate and observed under a transmission electron microscope (Hitachi H7500-I).
Anesthetized rats were examined by EMG. Fibrillation and fasciculation potentials of gastrocnemius muscles were recorded with an EMG machine (CMS6600; COTEC Inc.) as previously described [34].
Motor neurons in the ventral horn of the L3 lumbar cord were stereologically counted as previously described [34]. Neurons larger than 25 µm in diameter were counted in the ventral horns on both sides. For estimation of neurons in the frontal cortex and dentate gyrus, one hemisphere of the brain was used for cell counting. The forebrain was cut into coronal sections of 30 µm between the apical rostral part of the brain and the first occurrence of hippocampus, and every 12th section (a total of 15 to 18 sections) was counted for neurons in the defined frontal cortex (Figure S3). The portion of the brain containing the dentate gyrus was cut into consecutive sections (20 µm) and every 12th section (a total of 16 to 21 sections) was counted for neurons in the dentate gyrus. Tissue sections were stained with Cresyl violet and mounted in sequential order (rostral-caudal). The number of targeted neurons was estimated using a fractionator-based unbiased stereology software program (Stereologer) run on a PC computer that was attached to a Nikon 80i microscope fitted with a motorized XYZ stage (Prior). At low magnification, the targeting area was outlined and a random sampling grid was created. At high magnification, an optical dissector probe in the designated area was randomly generated by the program. The presence of clearly definable neurons was noted according to defined inclusion and exclusion limits of the dissector. This process was repeated on all selected sections. The total number of defined neurons was calculated by the software according to the result of random counts as previously described [34].
The number of defined neurons in the defined region was statistically compared between groups of transgenic rats and comparison among experimental groups was performed by one-way ANOVA followed by Tukey's post-hoc test. The null hypothesis was rejected at the level of 0.05.
|
10.1371/journal.ppat.0030189 | Apicoplast Lipoic Acid Protein Ligase B Is Not Essential for Plasmodium falciparum | Lipoic acid (LA) is an essential cofactor of α-keto acid dehydrogenase complexes (KADHs) and the glycine cleavage system. In Plasmodium, LA is attached to the KADHs by organelle-specific lipoylation pathways. Biosynthesis of LA exclusively occurs in the apicoplast, comprising octanoyl-[acyl carrier protein]: protein N-octanoyltransferase (LipB) and LA synthase. Salvage of LA is mitochondrial and scavenged LA is ligated to the KADHs by LA protein ligase 1 (LplA1). Both pathways are entirely independent, suggesting that both are likely to be essential for parasite survival. However, disruption of the LipB gene did not negatively affect parasite growth despite a drastic loss of LA (>90%). Surprisingly, the sole, apicoplast-located pyruvate dehydrogenase still showed lipoylation, suggesting that an alternative lipoylation pathway exists in this organelle. We provide evidence that this residual lipoylation is attributable to the dual targeted, functional lipoate protein ligase 2 (LplA2). Localisation studies show that LplA2 is present in both mitochondrion and apicoplast suggesting redundancy between the lipoic acid protein ligases in the erythrocytic stages of P. falciparum.
| Plasmodium falciparum is the causative agent of severe malaria. The parasites possess two organelles that are integral to their metabolism—the mitochondrion and the apicoplast, a remnant plastid. Both organelles contain enzymes that depend on the attachment of the cofactor lipoic acid for their catalytic activity. These are the α-keto acid dehydrogenase complexes and the glycine cleavage system (GCS). The pyruvate dehydrogenase (PDH) is solely found in the apicoplast of the parasites whereas α-keto glutarate and branched chain α-keto acid dehydrogenase as well as the GCS are mitochondrial. Both organelles possess specific and independent mechanisms that guarantee the posttranslational lipoylation of these enzyme complexes. In this study we show that the apicoplast located lipoic acid protein ligase, octanoyl-[acyl carrier protein]: protein N-octanoyltransferase (LipB), is not essential for parasite survival by disrupting the LipB gene locus. Despite a drastic loss of total lipoic acid, the parasites progress through their intraerythrocytic development unperturbed although the apicoplast-located PDH shows a reduced level of lipoylation. This phenotype is attributable to the presence of the recently described lipoic acid protein ligase 2, LplA2, which we show to be dually targeted to mitochondrion and apicoplast.
| Lipoic acid (6,8-thioctic acid; LA) is an essential cofactor that is covalently attached to the transacylase subunit (E2-subunit) of α-keto acid dehydrogenase complexes (KADHs), namely pyruvate dehydrogenase (PDH), α-keto glutarate dehydrogenase (KGDH), and branched chain α-keto acid dehydrogenase (BCDH) as well as the H-protein of the glycine cleavage system (GCS) [1,2]. In eukaryotes, these multienzyme complexes are generally found in the mitochondrion. Only plants and plastid-containing organisms possess organelle-specific PDH with the plastid PDH providing substrates for fatty acid biosynthesis [3]. Therefore, mitochondrion and plastid require the enzymatic machineries for the posttranslational lipoylation of KADHs or H-protein [2–5].
LA is provided and ligated to the respective target proteins by two distinct pathways. The cofactor can be synthesised by almost all organisms using the LA biosynthesis pathway. This requires octanoyl-acyl carrier protein (ACP) as a substrate (a product of fatty acid biosynthesis) which is ligated to the apo-E2-subunits or the apo-H-protein by octanoyl-[acyl carrier protein]: protein N-octanoyltransferase (LipB) [6]. Subsequently, two sulphurs are introduced into position 6 and 8 of the protein-bound octanoic acid, a reaction that is catalysed by lipoic acid synthase (LipA) [7,8]. LA can also be acquired through the salvage pathway. In mammals free, salvaged LA is transferred to the E2-subunits of KADHs through two enzymatic steps but in bacteria, fungi, and apicomplexan parasites this reaction is catalysed by a single enzyme [6,9–12]. Scavenged LA in mammals is first activated through an ATP-dependent reaction catalysed by LA activating enzyme before the activated form of LA is then attached to the E2-subunits or the H-protein by LA transferase [9,10]. In contrast, bacterial-type LA protein ligases (LplA) catalyse the activation and transfer of LA in a single enzymatic step [6].
LA metabolism in the malaria parasite Plasmodium falciparum and the related apicomplexan parasite Toxoplasma gondii display an organelle-specific distribution of biosynthetic and salvage pathways [11–15]. LA biosynthesis is exclusively found in their plastid-like organelle, the apicoplast, whereas LA salvage is confined to their mitochondrion. It was shown that both organelles contain members of the KADHs, which require posttranslational lipoylation [16–18]. It is assumed that these multienzyme complexes play pivotal roles in the parasite's metabolism and it is thought that both LA biosynthesis and salvage are essential for parasite survival. This is further supported by the findings of Crawford and colleagues [13] showing that newly synthesised LA does not exit the apicoplast and by Allary and colleagues [14] who showed that radiolabelled LA is not utilised to lipoylate apicoplast PDH, but only leads to lipoylation of the mitochondrial E2-subunits. These studies therefore provide evidence that the organelles' lipoylation machineries act independently and inhibition of either one should have deleterious effects for the parasites. Indeed proof of this concept is supported by the lethal effect of the LA analogue 8-bromo-octanoic acid on intraerythrocytic stages of P. falciparum and also T. gondii [13,14]. In this study we have further tested this hypothesis by disrupting the Plasmodium apicoplast targeted LipB gene, which is part of the LA de novo biosynthesis pathway. Surprisingly, the gene disruption is not deleterious for the parasites suggesting that Plasmodium possesses alternative routes for LA ligation in this organelle. Our results provide evidence that a second LA protein ligase-like protein, LplA2 [13,14], can replace LipB function.
To verify that the protein bands recognised by an antibody directed against protein-bound LA (anti-LA), which was used in this study to detect lipoylated proteins in P. falciparum (Figure 1, lane 4), Western blots of parasite extracts were also analysed with antisera raised against P. falciparum H-protein, BCDH-E2 and KGDH-E2 (Figure 1, lanes 1–3). The sizes of the proteins detected by the anti-LA antibody correlated well with those protein bands detected by the antibodies directed against the three mitochondrial proteins. These results corroborate the previously published suggestion that the protein bands detected by the anti-LA antibody in fact represent the mitochondrial KADHs. The 75 kDa band that was detected by anti-LA was previously shown to correspond to the PDH-E2-subunit [14].
The gene encoding LipB was targeted using two different constructs (see Materials and Methods) cloned into the P. falciparum transfection plasmid pHH1. This plasmid confers single cross over recombination resulting in a disruption of the endogenous gene locus [19]. Both constructs lacked the last 100 amino acids including a catalytically essential, highly conserved cysteine residue at position 369 (Plasmodium LipB numbering) which was shown to form a catalytic dyad with a conserved lysine residue in position 307 (Plasmodium LipB numbering) in the Mycobacterium tuberculosum LipB protein [20]. Therefore it was assumed that the disruption of the endogenous LipB gene locus should result in the generation of a C-terminally truncated LipB protein unlikely to display any enzymatic activity.
Independent transfections of both constructs were performed and the genotype of the transfected parasite lines was analysed by Southern blotting to verify the presence of the transfected plasmid. All analysed parasite genotypes revealed that the LipB gene locus had been targeted and the independent transfectants were cloned by limiting dilution. Two independent clones from separate transfections with each construct were used for further analyses (Figure 2). LipBKO1–1 and LipBKO1–2 describe the 3D7 derived mutants and LipBKO2–1 and LipBKO2–2 describe the D10 derived mutants. Upon NdeI digestion of genomic DNA of all four clones, the endogenous gene of 2.8 kb (Figure 2, lanes 1 and 2) is replaced by two bands of approximately 3.7 kb and 4.9 kb (Figure 2, lanes 3–6) diagnostic for the disruption of the LipB gene locus.
Growth experiments performed according to Sanders et al. [21] showed that the LipB disruption with either construct, and regardless of parasite strain, resulted in a modestly increased growth rate of the LipB mutants compared to both parent lines (3D7 and D10) (Figure 3). Six days after the experiment was started with highly synchronised, ring-stage parasites, the parasitemia of the wild-type controls was determined to be between 1.8% and 2% whereas the LipB mutants consistently had a parasitemia between 4% and 5%. This increased growth rate could be explained in several ways; e.g., an accelerated cell cycle, the generation of a greater number of merozoites, or a more successful invasion rate by the mutant parasites.
The first point was addressed by following tightly synchronised parasites through the 48 h developmental cycle. Blood smears were taken every 8 h and the progression through the cell cycle of the different parasite lines was analysed microscopically (Figure 4). The experiment was started with tightly synchronised ring-stage parasites of all six parasite lines and this time point was set as 0 h. Eight hours later, between 50% and 80% of the LipB mutants had developed into trophozoites, whereas only 20% to 25% of either D10 or 3D7 Plasmodium had progressed to this developmental stage. Generally the 3D7 wild-type and the 3D7-based mutants showed a faster progression through their intraerythrocytic development than the D10 wild-type and mutant lines (Figure 4). Thirty two hours after the start of the experiment almost all of the LipB mutant parasites had re-infected fresh erythrocytes and had successfully progressed through their life cycle, whereas 3D7 needed another 8 h to reach this point and D10 needed even longer. In summary, the data clearly show that all four LipBKO mutants (the two selected 3D7-based KO1 clones and the two D10-based KO2 clones) progress through their intraerythrocytic cell cycle faster (about 4 to 8 h) than the two wild-type lines, which suggests that the disruption of the LipB gene locus in some way affects parasite cell cycle control. Determining parasite numbers at the beginning and the end of the parasites' life cycles revealed that the LipBKO schizonts generated, on average, the same number of successfully infective merozoites as wild-type parasites per life cycle.
The LA content of wild-type and the LipB null mutants was quantified by gas chromatography-mass spectrometry (GC-MS). This revealed that all of the LipB null mutants showed a drastic reduction in total LA content (Figure 5A; Table 1) compared to the wild-type parasite lines. Both P. falciparum 3D7 and D10 contain approximately 40 nmol/108 cells of LA whereas the LipB mutants only contain between 0.6 nmol/108 cells and 2.2 nmol/108 cells; a reduction of more than 90% of total LA. In addition to the total LA in the cell, it was possible to distinguish between oxidised and reduced protein-bound LA; the ratio of which was only marginally affected by the LipB disruption, varying between 1.2 and 4.5, with the oxidised form of LA being predominant (Table 1).
Apart from changes in LA levels, the LipB disruption lead to a 5-fold increase in myristate (C14:0) in the LipB mutants (from 1.8 nmoles/108 cells to 9 nmoles/108 cells). This is presumably the result of the increased availability of octanoyl-ACP, which is no longer required for LA biosynthesis and so can be further extended to C14 (Figure 5B). Thus, the decreased requirement for LA biosynthesis because of the lack of LipB function generates a surplus of longer chain fatty acids, which can be used for instance for protein modifications such as acylations and lipid biosynthesis. The major overall finding of this part of the study was that the LipB disruption leads to a drastic reduction of total LA in the parasites without affecting parasite viability.
From these data it also can be deduced that the level of lipoylation of PDH-E2 should be greatly reduced if not ablated if LipB is the only protein that transfers the LA cofactor to the multienzyme complex in the apicoplast. This was qualitatively analysed by Western blotting using anti-LA antibodies which shows that PDH-E2 lipoylation indeed decreases significantly but does not totally disappear (Figure 5C, compare lane 1 with lanes 2 and 3), suggesting an alternative mechanism that allows for partial PDH-E2 lipoylation in the absence of LipB activity. The E2-subunits of the mitochondrial KADH also show a slight decrease of lipoylation, suggesting that the knockout of the apicoplast LipB might also affect the modification of the mitochondrial enzyme complexes.
Apicoplast PDH is thought to be the source of acetyl-CoA for type II fatty acid biosynthesis in apicomplexan parasites [22] and the drastic reduction of LA in the parasites led to the conclusion that the activity of PDH-E2 might be negatively affected and as a consequence the provision of substrates for de novo fatty acid biosynthesis might be reduced. This hypothesis was tested by investigating the sensitivity of the mutants to triclosan, an inhibitor of FabI; one of the enzymes involved in fatty acid elongation in P. falciparum and previously validated as a drug target [23]. However, our data show that the LipB disruption does not have any effect on the susceptibility of mutant parasites towards triclosan, possibly suggesting that the residual lipoylation of PDH-E2 is sufficient to provide enough acetyl-CoA to sustain fatty acid biosynthesis at wild type level (Figure 6A).
Another hypothesis requiring investigation was the suggestion that LA might act as a principal antioxidant in the organelles of Plasmodium. In order to further substantiate this, the IC50 for two pro-oxidants were determined; wild-type and LipB null mutants showed no differential susceptibility towards tert-butylhydroperoxide and N-methylphenazonium methosulfate, respectively (Figure 6B and 6C).
It was shown previously that Plasmodium possess a second functional LplA-like protein, which compensated growth of a bacterial strain lacking both LipB and LplA, but which was unable to compensate growth of a LipB deficient bacterial line in a previous study [14]. It was suggested that LplA2 is not able to replace LipB function in the bacteria because of their distinct substrate specificities. In this study we expressed three different expression constructs in LipB and LipB/LplA deficient Escherichia coli lines, respectively [24,25]. The constructs used here were full length at their C-termini and included a C-terminal tag as opposed to the construct expressed in the previous study. The N-terminus of the three constructs differed because the deduced amino acid sequence has an N-terminal extension of 28 amino acids when compared to E. coli LplA which potentially encodes an N-terminal targeting sequence. It is possible that such N-terminal targeting sequences interfere with efficient expression or function of the heterologous parasite protein in the prokaryotic expression system and therefore it was decided to analyse three different expression constructs of LplA2 (as outlined in the Materials and Methods section) in the complementation assay. In our hands all three constructs complemented the growth of the LipB deficient bacteria (Figure 7A), suggesting that the protein can replace LipB function. These data are in contrast to those obtained by [14] but might be explained by differences in expression plasmids and constructs used in the two different studies. This does not exclude the possibility that LplA1 and LplA2 proteins have differential, albeit somewhat overlapping, substrate specificities or activity profiles through the developmental cycle of P. falciparum. The LplA2 constructs also compensated for the growth defect of TM136, a bacterial line deficient in LplA and LipB (Figure 7B) [25].
A clear prediction of LplA2′s localisation is not possible and therefore this was analysed by expressing a C-terminally green fluorescent protein (GFP)-tagged full-length LplA2 protein in the erythrocytic stages of P. falciparum. The results suggest that LplA2-GFP is targeted to two distinct organelles—one of which is clearly the mitochondrion as LplA2-GFP colocalises with Mitotracker (Figure 8). Given the close association of the second organelle to the mitochondrion we suggest that this is likely to be the apicoplast. This was corroborated by immunofluorescence studies using antibodies raised against LplA2 and apicoplast lipoamide dehydrogenase (aLipDH; aE3; P. J. McMillan and S. Müller, unpublished data). The results show that in some parasites both proteins clearly colocalise, supporting its localisation in the apicoplast (Figure 9). However, the localisation of LplA2 is not that straightforward. In some parasites we clearly observe colocalisation with either Mitotracker or, in the immunofluorescence study, with aLipDH, whereas in others the staining is present in both organelles. Analysing the distribution in 46 distinct parasites from the immunofluorescent experiment resulted in the following distribution: 19.6% of LplA2 colocalised with aLipDH and thus is apicoplast located; 53% were found in an organelle distinct to the apicoplast but likely to be the mitochondrion, and 28 % showed staining of both organelles. Similarly, the LplA2-GFP expressing parasites were analysed (50 parasites) and 15% of staining was likely to be apicoplast (distinct from the mitochondrion stained with Mitotracker), 68% were found to have LplA2-GFP in the mitochondrion and 18% showed staining in both organelles. These data indicate that LplA2 is dually targeted within the parasites albeit the precise mechanisms governing this distribution need to be analysed in future studies.
Overall, the functionality and localisation studies of LplA2 strongly support our hypothesis that LplA2 indeed compensates for the loss of LipB function, although it appears that its ability to utilise the octanoyl-ACP substrate provided by fatty acid biosynthesis might not be as efficient as it is by LipB given the extensive loss of protein-bound LA and the reduction of lipoylation of PDH-E2.
Lipoic acid metabolism in apicomplexan parasites is distributed between mitochondrion and apicoplast [11,12]. Both organelles lipoylate their KADHs independently, with de novo biosynthesis confined to the apicoplast and salvage solely found in the mitochondrion of the parasite [13–15].
Given this background it was surprising that the LipB gene, which encodes the principal LA protein ligase in the apicoplast, can be disrupted without negatively affecting P. falciparum growth. This was entirely unexpected because it was thought that LipB activity is absolutely required for the parasites to lipoylate PDH-E2, which in turn is necessary to provide acetyl-CoA for fatty acid biosynthesis operating in the organelle [22]. In fact, down regulation of ACP expression using a Tet-inducible conditional knockout system in T. gondii revealed that one major function of apicoplast located type II fatty acid biosynthesis is to provide octanoyl-ACP for the lipoylation of PDH-E2 in these related parasites [15]. These data indirectly suggest that salvage of LA and its subsequent ligation to the PDH-E2 is not an alternative for the apicoplast located LA de novo biosynthesis pathway in Toxoplasma. This is in agreement with data on bacterial LipB, which does not accept free LA as a substrate [6,26]. Thus, even if exogenous LA would be taken up into the apicoplast of Toxoplasma it would require the presence of a LplA-like protein to guarantee ligation of the cofactor to the apo-PDH-E2. Analyses of a gene identified in the ToxoDB database potentially encoding LplA2 (gene locus: 83.m01296) showed that the identity between the deduced amino acid sequences of the Toxoplasma and Plasmodium deduced amino acid sequences was modest, with less than 10% identity. This is in contrast to potential LplA2 encoding genes in other Plasmodium species but also other apicomplexans, such as Theileria, which contain highly conserved orthologues of P. falciparum LplA2 (similarities between 58% and 37%). In addition, highly conserved amino acid motifs necessary for substrate interaction and activation appear to be absent from the potential T. gondii LplA2 [27–29]. Therefore, it appears that the lipoylation pathways in Plasmodium and Toxoplasma differ to some extent.
The LA content of the LipB mutant parasites was analysed by GC-MS which revealed that the LipB mutants had a drastic reduction (∼90%) of LA compared to wild-type parasites. Concomitantly, the levels of myristate (C14:0) increased about 5-fold in the mutant parasites compared to wild-type parasites, which could potentially have implications such as the levels of protein lipidations or specific lipid species in the mutant parasite lines [30]. Despite the severe loss of LA, the LipB mutant parasites showed a faster growth phenotype which was primarily attributable to accelerated progression through the intraerythrocytic cell cycle. It is conceivable that the increased availability of endogenously generated myristate changes the acylation state of regulatory proteins and affects their activity and/or distribution, which potentially contributes to the observed phenotype. However, these speculations have to be further substantiated in future analyses of the LipB mutants. Another aspect is that the loss of LA has an impact on the parasites' capacity to defend themselves against oxidative stress considering that one of the most discussed roles of LA is its redox activity and its potential as antioxidant [31,32]. However, the LipB mutants appear to be unaffected in their susceptibility to exogenous and endogenous oxidants despite the significant loss of LA. This potentially could be explained by a compensatory upregulation of alternative antioxidants in these mutant parasites—a hypothesis that also needs further investigation.
The observed reduction of LA in the parasites also implies that PDH-E2 lipoylation should be greatly reduced; by using an anti-LA antibody this was confirmed by Western blotting. The reduced lipoylation appears to not negatively affect de novo fatty acid biosynthesis as shown by the fatty acid analyses data (reduction of LA biosynthesis leads to increased levels of longer chain fatty acids). In addition, mutant and parent parasite lines showed similar susceptibilities to the supposed FabI inhibitor triclosan [21]. This shows that the reduction of PDH-E2 lipoylation does not affect PDH activity as severely as originally believed and that the multienzyme complex still provides sufficient acetyl-CoA to sustain fatty acid biosynthesis. Previous studies on E. coli PDH showed that the loss of one or two of the three lipoyl-domains of the bacterial PDH-E2 subunit does not cause significant changes in PDH activity demonstrating that under-lipoylation does not necessarily yield a catalytically incompetent PDH complex [33].
The fact that PDH-E2 is at all lipoylated is surprising given that LipB was thought to be the principal LA protein ligase present in the apicoplast. However, we have shown in this study that an alternative pathway that allows lipoylation of PDH-E2 is provided by LplA2, a second LA protein ligase like protein, identified in the genome of several Plasmodium species [14] and a number of other apicomplexan parasites. However, Toxoplasma appears to lack a LplA2 orthologue—a gene potentially encoding LplA2 lacks amino acid motifs essential for a functional LplA protein. The functionality of Plasmodium LplA2 was corroborated in this study and it was shown that different expression constructs complement the growth defect of bacteria deficient in LipB and LipB/LplA, supporting that the protein compensates for both LplA and LipB in this bacterial expression system. This is in contrast to the findings of [14] who suggested that LplA2 can replace mitochondrial LplA1 only, and not LipB. However, the expression constructs used in the previous study differ considerably from the ones used in this study, which might explain these different results. Furthermore, it needs to be emphasised that LplA2 seems to be less efficient in using octanoyl-ACP as a substrate to lipoylate the PDH-E2 subunit as shown by the large loss of total LA and the under-lipoylation of PDH-E2. This clearly suggests that the substrate specificities of LipB and LplA2 might differ and studies to fully characterise LplA2 biochemically and its precise role for parasite survival are currently underway. The difference in substrate specificity between LipB and LplAs is, however, not very surprising and has been previously shown for the bacterial enzymes [6]. In fact, overexpression of LipB in E. coli render them insensitive towards selenolipoic acid, suggesting that LipB cannot use exogenously supplied LA or its derivatives as substrates [6,26].
The localisation of LplA2 cannot be reliably predicted and therefore the full-length protein C-terminally tagged with GFP was expressed in the erythrocytic stages of P. falciparum. The results that we obtained were intriguing because the GFP fluorescence was observed in both the mitochondrion and a closely associated organelle likely to be the apicoplast. In order to exclude that these results were attributable to the over expression of the LplA2-GFP-fusion protein we also performed immunofluorescence studies on wild-type parasites. The anti-LplA2 antibody used detected a protein that either colocalised with an apicoplast marker (aLipDH), was closely associated with the apicoplast marker, or was observed in both organelles. This implies that LplA2 is dually targeted to mitochondrion and apicoplast supporting the hypothesis that the loss of LipB functionality can be compensated by LplA2. Dual targeting has been shown previously in P. falciparum for the metalloprotease falcilysin [34] and potential mechanisms by which this is achieved or governed have been discussed by Ralph [35]. In organisms that contain plastid and mitochondrion dual targeting is not unusual, particularly for those proteins involved in biological processes that are found in both organelles [36,37]. Most proteins targeted to the two organelles are nuclear-encoded and they possess certain targeting signals that are not always predictable using bioinformatics approaches [38]. It has been suggested that not only the primary amino acid sequence of a protein is involved in the control of dual targeting but that also untranslated regions of the mRNA play a role in the process [37,39]. In apicomplexan parasites the posttranslational trafficking of apicoplast and mitochondrial proteins differs considerably from those in plants—mitochondrial proteins are delivered via the cytosol to their destination whereas the apicoplast targeting is through the secretory pathway [40]. Therefore it has to be assumed that specific mechanisms allow for dual targeting of apicomplexan proteins to both organelles. Recently, a study by Pino and colleagues showed that the nature of the signal peptide affects targeting of a number of unrelated proteins to both mitochondrion and apicoplast in T. gondii [41]. It is well possible that similar mechanisms occur in Plasmodium and the fact that lipoylation is an essential process in both organelles might be the reason for the dual targeting of LplA2 that we observed in this study.
Overall, this study shows that redundancies exist between LA protein ligases in the malaria parasite P. falciparum, which appear to be achievable through dual targeting of LplA2. The accelerated progression through the cell cycle during intraerythrocytic growth of LipB mutants implies that the lack of LA and LipB is affecting cell cycle control mechanisms. Future studies will elucidate the underlying reasons for the rapid progression through the intraerythrocytic cycle and analyse whether there are growth and developmental impairments during other life cycle stages of the LipB mutant parasites.
Albumax II and RPMI 1640 were obtained from Invitrogen Corporation, UK. Irgasan (triclosan), tert-butylhydroperoxide and N-methylphenazonium methosulfate were purchased from Sigma-Aldrich, UK. The ImmobilonTM Western Chemiluminescent HRP Substrate was obtained from Millipore, UK. Anti-LA rabbit polyclonal antibody was supplied by Calbiochem and the anti-rabbit IgG (H+L), HRP conjugate was from Promega. [α-32P]-ATP (Adenosine 5′-triphosphate [α-32P], EasyTides, specific activity: 3,000 Ci/mmol) was purchased from Perkin-Elmer. [8-3H]-hypoxanthine (specific activity: 10–30 Ci/mmol) was from GE Healthcare, UK. All restriction enzymes were obtained from New England Biolabs. WR99210 was generously provided by Dr Jacobus, Jacobus Pharmaceuticals, USA. The vector pASK-IBA3 was purchased from Institut für Bioanalytik, Germany. The LipB-deficient (KER 184) and LipB/LplA deficient (TM136) bacterial strains were a kind gift from Dr John Cronan (University of Illinois at Urbana-Champaign, USA). Plasmids pHH1and pCDH3/4, PfHSP86 5′-pDONR4/1, and PfCRT 5′-pDONR4/1 were kind gifts from Professor A. F. Cowman (The Walter and Eliza Hall Institute for Medical Research, Melbourne, Australia) and Professor G. I. McFadden (University of Melbourne, Australia), respectively.
The expression and knockout fragments of the P. falciparum LipB gene were amplified from P. falciparum 3D7 and D10 genomic DNA using Pfx Supermix (Invitrogen). The specific oligonucleotide primers 5′-GCGCAGATCTAATAAAATAAACCTGCTTGTAC-3′ (sense) and 5′-GCGCCTCGAG(TTA)TTTATCCTTATAAAAGATACC-3′ (antisense) with the BglII and XhoI restriction sites, respectively, in bold, and an artificial stop codon (in brackets) within the antisense oligonucleotide, were used to generate the 999 bp insert equivalent to nucleotides 4–1,002 of the PfLipB open reading frame for the PfLipBKO1 construct. The PCR product was subcloned into the TOPO-Blunt PCR cloning vector (Invitrogen) and its sequence was verified (The Sequencing Service, University of Dundee, UK, http://www.dnaseq.co.uk/) before it was cloned into the P. falciparum transfection plasmid pHH1 [19]. The second construct comprises nucleotides 304–1,002 of the open reading frame missing the potential bipartite targeting sequence. The insert for PfLipBKO2 was amplified using the sense primer 5′-GCGCAGATCTATTATGAAAAATAAAAATGAAGTACAAATATCAAATCATTTAG-3′ and the same antisense primer as for PfLipBKO1. The 699 bp product was subcloned into TOPO-Blunt as described above for sequence verification before being cloned into pHH1.
P. falciparum 3D7 (The Netherlands) and P. falciparum D10 (Papua New Guinea) were cultured according to Trager and Jensen [42] with modifications in human erythrocytes, RPMI 1640 containing 11 mM glucose, with the addition of 0.5% Albumax II. The parasites were maintained under an atmosphere of reduced oxygen (1% oxygen, 3% CO2, and 96% nitrogen) at 37 °C. Parasites were synchronised using sorbitol according to Lambros and Vanderberg [43].
Transfection of PfLipBKO1-pHH1, PfLipBKO2-pHH1, HSP86-LplA2-GFP-pCHDR, and CRT-LplA2-GFP-pCHDR into P. falciparum erythrocytic stages was performed as described previously [44,45]. WR99210 resistant parasites appeared between 40 and 60 days after transfection. Parasites were cloned by limiting dilution according to Kirkman et al. [46].
The effect of triclosan, N-methylphenazonium methosulfate, and tert-butylhydroperoxide on P. falciparum erythrocytic stages was determined by measuring the incorporation of [3H]-hypoxanthine in the presence of increasing drug concentrations (0.5 μM to 100 μM) according to [47].
Relative parasite growth rates were determined using the method of Sanders et al. [21]. Parasite cultures containing mainly ring stages were synchronised twice within 4 h using sorbitol [43]. Parasite density was determined and the culture was diluted to 0.5% parasitaemia, 5% haematocrit. Cultures were maintained under an atmosphere of reduced oxygen at 37 °C and medium was refreshed every 24 h. Cultures were diluted 5-fold at 48 h intervals and growth was monitored by Giemsa-stained thin blood smears every 24 h. For each determination of percentage parasitaemia the number of infected erythrocytes per 1,000 erythrocytes was recorded.
Cultured parasites were enriched for late-stage forms by using the VarioMACS separator and CS MACS columns (Miltenyi Biotec). The columns were equilibrated with MACS buffer (PBS supplemented with 0.5% (w/v) BSA, 2mM EDTA) for 5 min and rinsed with 60 ml of MACS buffer. The equivalent of 100 ml cultured P. falciparum was resuspended in MACS buffer and applied to the column. Following flow-through of the cell-suspension, the column was washed with 50 ml MACS buffer. The column was then removed from the magnetic field of the VarioMACS separator and the late-stage parasitised erythrocytes were eluted from the column using 30 ml MACS buffer. The suspension was centrifuged and the cells were resuspended in complete medium and returned to culture conditions for 30 min prior to harvesting by saponin lysis. The typical yield from this procedure was 1–2 × 108 isolated late-stage parasites.
The parasites were liberated from erythrocytes by saponin lysis [48] and genomic DNA was isolated using the QIAamp DNA Mini Kit (Qiagen). Protein extracts were prepared from saponin-isolated parasites by resuspending the pellets in lysis buffer (100 mM HEPES (pH 7.4), 5 mM MgCl2, 10 mM EDTA, 0.5% (v/v) TritonX-100, 5 μg/ml RNAse, 1 mM phenylmethylsulphonyl fluoride, 1 mM benzamidine, 2 μg/ml leupeptin, 10 μM E-64, 2 μM 1,10-phenanthroline, 4 μM pepstatin A) followed by three cycles of freeze/thawing and sonication in a sonicating water-bath (Fisherbrand). Protein concentrations were determined using the Bradford assay [49].
One μg of genomic DNA was digested with NdeI, separated on a 0.8% agarose gel, and blotted onto positively charged nylon membrane (GE Healthcare) using standard methods [50]. The blot was probed with the LipB coding sequence. Radioactive probes were made using the MegaPrime Labelling Kit from GE Biosciences and [α-32P]-ATP (Adenosine 5′-triphosphate [α-32P], EasyTides, Specific Activity: 3,000 Ci/mMole) from Perkin-Elmer following the manufacturer's recommendations. The membrane was prehybridised (0.5% (w/v) SDS, 5 × Denhardt's solution, 100 μg/ml salmon sperm DNA, 0.1% (w/v) sodium pyrophosphate) for 2 h at 60 °C before addition of the probe. Hybridisation was then allowed to proceed at 60 °C over night. Membranes were washed once in 6× SSC, 0.1% (w/v) SDS at 60 °C for 20 min, and then twice in 2× SSC, 0.1% (w/v) SDS at 60 °C for 10 min. Membranes were exposed to Kodak film for several days before development depending on the activity of the probe used.
To determine lipoylation of KADH-E2 subunits in P. falciparum, protein extracts of parasites were subjected to Western blotting. Briefly, 15 μg of each sample was separated on a 4%-12% SDS-PAGE (Invitrogen) and then blotted onto nitrocellulose (Schleicher and Schüll), using standard techniques [50]. The blot was incubated with a rabbit anti-LA antibody (Calbiochem) at a dilution of 1:500 and the secondary anti-rabbit IgG (H+L), HRP conjugate (Promega) at a dilution of 1:10,000 before being developed using the ImmobilonTM Western Chemiluminescent HRP Substrate (Millipore). Similarly, blots of wild-type parasite extracts were also probed with antibodies against BCDH-E2 (raised in rabbit), KGDH-E2 (raised in rat) and H-protein (raised in rabbit) of P. falciparum (all generated by Eurogentec, Belgium) at dilutions of 1:5,000, 1:100 and 1:2,000, respectively.
Samples for LA analyses were prepared from late trophozoites prepared using the MACS columns described above. The LA detection method was modified from that of Pratt and colleagues [51]. LA determinations were done in triplicate, along with a parallel control of standards, all containing an internal standard of heptadecanoic acid (10 nmol). Experimental details for the determination of LA will be published elsewhere (T. K. Smith, in preparation). Briefly, total LA was determined by acid hydrolysis of freeze-dried parasite pellets to release protein bound fatty acids. Reduction with NaBH4 and methylation with methyl iodide of the sulphydryl groups under basic conditions was followed by organic extraction of all fatty acids, including the methylated LA. The dried fatty acids were converted to fatty acid methyl esters (FAME) with diazomethane and stored dried at −20 °C until subjected to GC-MS. Quantification of oxidised LA was as above except no reduction or methylation of the sulphydryl groups is required. Free LA was extracted from a freeze-dried cell pellet with organic solvents which were checked for protein by analysis of SDS-PAGE and staining, prior to treatment with methanolic HCl and conversion to fatty acid methyl esters as above.
Analysis of the FAMEs was conducted on a Hewlett Packward 6890–5973 system equipped with a ZB-5 30M × 0.25 mm (I.D.) column. The electron impact ionization/quadrupole mass detector was programmed to monitor selected ions for all FAMEs m/z 123, heptadecanoic acid m/z 284, oxidised LA m/z 220, reduced and methylated m/z 250, with typical elution times of oxidised LA, 26.4 min; reduced and methylated LA, 27.2 min; and heptadeconic acid (C17:0), 28.5 min. Molar response factor for oxidised LA and methylated reduced LA versus heptadecanoic acid were determined to estimate the LA content. In addition, the total ion current (TIC) chromatograms were analysed for changes in other longer chain fatty acids.
The functionality of three expression constructs of LplA2, generated in pASK-IBA3, was analysed. The following anti-sense primer 5′-GCGCGCGGTCTCAGCGCTTAGAAAATATGTTGGTATATCGTAATACC-3′ was used to amplify all three constructs. In combination with the sense primer 5′-GCGCGCGGTCTCGAATGAGAATTATAAAGTGCCTGGATC-3′ a 1,152 bp fragment was amplified corresponding to the full length gene. The sense primer 5′-GCGCGCGGTCTCGAATGAAAAAAATAAACATTCTTTATTTTATTGATGTCAGC-3′ generated a truncated fragment from nucleotide 79–1,152 (S1 construct). The third fragment (S2) amplified using 5′-GCGCGCGGTCTCGAATGAATGAGTCCAAAGGAAACGAATGC-3′ corresponds to nucleotide 235–1,152 bp. All primers contained a BsaI restriction site (boldface) to allow directional cloning into pASK-IBA3. The constructs were amplified from P. falciparum 3D7 genomic DNA using Pfx Supermix (Invitrogen) and were initially cloned into TOPO-Blunt PCR cloning vector (Invitrogen) for sequence verification. Subsequently, they were subcloned into pASK-IBA3 and transformed into KER 184 and TM 136 [24,25] to assess whether they complement the growth defect of the bacterial lines when grown on minimal medium agar plates.
To analyse the localisation of LplA2, the full length LplA2 gene was amplified from P. falciparum 3D7 genomic DNA using the sense primer 5′-CACCATGAGAATTATAAAGTGCCTGG-3′ and antisense primer 5′-TAGAAAATATGTTGGTATATCGTAATACC-3′. The PCR product was cloned directionally into the plasmid pENTR/D-TOPO (Invitrogen) and the sequence was verified by sequencing. The cloning of the constructs was performed as described by van Dooren et al. [52]. The destination plasmid used was pCHDR-3/4 which contains the human dihydrofolate reductase (hDHFR) as a selectable marker. Two entry plasmids were used that differed in their promoter regions. The PfHsp86 5′-pDONR4/1 contains the P. falciparum heat shock protein 86 5′ UTR and the PfCRT 5′-pDONR4/1 possesses the P. falciparum chloroquine resistance transporter 5′UTR. The C-terminal GFP-tag was provided by GFP-pENTR2/3. Thus, four plasmids (either PfHspP86-pENTR4/1 or PfCRT-pENTR4/1, LplA2-pENTR/D-TOPO, GFP-pENTR2/3, and pCHDR-3/4) were incubated in the LR MultiSite cloning reaction according to manufactures guidelines (Invitrogen) which resulted in the generation of two LplA2 constructs (Hsp86-LplA2-GFP-pCHDR and CRT-LplA2-GFP-pCHDR). Constructs were transfected as described above and parasites resistant to WR99210 were analysed using an Axioskop-2 mot plus microscope (Zeiss) equipped with a Hamamatsu C4742–95 CCD camera.
Fixations of wild-type 3D7 parasites for subsequent immunofluorescence analyses were carried out according to [53]. The primary antibodies raised against LplA2 (in a rat, Eurogentec) and against apicoplast lipoamide dehydrogenase (aLipDH; aE3) (in a rabbit, Eurogentec) were diluted in 3% (w/v) BSA in PBS at 1:500 and 1:200, respectively. Secondary antibodies (anti-rat conjugated with Alexa fluor 488 and anti-rabbit conjugated with Alexa fluor 594, Molecular Probes) were applied at 1:500 dilution in 3% (w/v) BSA in PBS for 1 h at 4 °C. DAPI at 0.5 μg/ml (Sigma) was added to the secondary antibody for 1 min and was then washed off as before. The slides were mounted with 2.5% (v/v) DAPCO in 50% (v/v) glycerol (Sigma) and were analysed using an Axioskop-2 mot plus microscope (Zeiss) equipped with a Hamamatsu C4742–95 CCD camera.
The accession numbers and ID numbers of the genes (obtained from the NCBI-protein database) described in this study are as follows: Bos taurus lipoate-activating enzyme (BAB40420), Bos taurus lipoyltransferase (BAA24354), E. coli (K12) LplA (NP_418803), E. coli (K12) LipB (NP_415163), P. berghei LplA2 (XP_679932; CAH95194), P. chabaudi LplA2 (XP_745010; CAH79244), P. falciparum ACP (XP_001349595; AAC71866), P. falciparum apicoplast lipoamide dehydrogenase (XP_001349365; CAD51214), P. falciparum BCDH-E2 (XP_001351112; CAB38991), P. falciparum enoyl-ACP reductase (FabI) (XP_966137; CAG25389), P. falciparum H-protein (XP_001348010; AAN35923), P. falciparum KGDH-E2 (XP_001349947; CAD52355), P. falciparum LipA (XP_001350160; CAD52569), P. falciparum LipB (XP_001349288; CAD51137), P. falciparum LplA1 (XP_001349882; CAD52290), P. falciparum LplA2 (XP_001352107; CAD51918), P. falciparum PDH-E2 (XP_001347486; AAN35399), P. knowlesi LplA2 (gene ID PlasmoDB: PKH_072080), P. vivax LplA2 (PlasmoDB gene ID: Pv099590), P. yoelii LplA2 (XP_730272 ; EAA21837), Theileria parva strain Ankara LplA2 (XP_954802), T. gondii ACP (AAC63956; AAC63953), T. gondii LplA2 (ToxoDB gene ID 83.m01296). |
10.1371/journal.pgen.1000929 | Genetic Tests for Ecological and Allopatric Speciation in Anoles on an Island Archipelago | From Darwin's study of the Galapagos and Wallace's study of Indonesia, islands have played an important role in evolutionary investigations, and radiations within archipelagos are readily interpreted as supporting the conventional view of allopatric speciation. Even during the ongoing paradigm shift towards other modes of speciation, island radiations, such as the Lesser Antillean anoles, are thought to exemplify this process. Geological and molecular phylogenetic evidence show that, in this archipelago, Martinique anoles provide several examples of secondary contact of island species. Four precursor island species, with up to 8 mybp divergence, met when their islands coalesced to form the current island of Martinique. Moreover, adjacent anole populations also show marked adaptation to distinct habitat zonation, allowing both allopatric and ecological speciation to be tested in this system. We take advantage of this opportunity of replicated island coalescence and independent ecological adaptation to carry out an extensive population genetic study of hypervariable neutral nuclear markers to show that even after these very substantial periods of spatial isolation these putative allospecies show less reproductive isolation than conspecific populations in adjacent habitats in all three cases of subsequent island coalescence. The degree of genetic interchange shows that while there is always a significant genetic signature of past allopatry, and this may be quite strong if the selection regime allows, there is no case of complete allopatric speciation, in spite of the strong primae facie case for it. Importantly there is greater genetic isolation across the xeric/rainforest ecotone than is associated with any secondary contact. This rejects the development of reproductive isolation in allopatric divergence, but supports the potential for ecological speciation, even though full speciation has not been achieved in this case. It also explains the paucity of anole species in the Lesser Antilles compared to the Greater Antilles.
| Over the last 150 years, since Darwin's study of islands and his “Origin of Species,” island archipelagos have played a central role in the understanding of evolution and how species multiply (speciation). Islands epitomise the conventional view of geographic (allopatric) speciation, where genomes diverge in isolation until accumulated differences result in reproductive isolation and the capacity to coexist without interbreeding. Current-day Martinique in the Lesser Antilles is composed of several ancient islands that have only recently coalesced into a single entity. The molecular phylogeny and geology show that these ancient islands have had their own tree lizard (anole) species for a very long time, about six to eight million years. Now they have met, we can genetically test for reproductive isolation. However, when we use selectively neutral markers from the nuclear genome, on this naturally replicated system, we can see that these anoles are freely exchanging genes and not behaving as species. Indeed, there is more genetic isolation between adjacent populations of the same species from different habitats than between separate putative allospecies from the ancient islands. This rejects allopatric speciation in a case study from a system thought to exemplify it, and suggests the potential importance of ecological speciation.
| Speciation generates biodiversity and is therefore a key process in evolution and ecology, and the relative importance of factors contributing to speciation in sexually reproducing animals, such as genetic drift in spatial isolation, natural selection, sexual selection and mutation-order, remains an active area of research [1]–[8]. Since neo-Darwinism [9] the most conventional view of speciation in sexually reproducing animals has been by the accumulation of differences by genetic drift and selection in allopatry. While there has been growing paradigm shift towards models [10] and processes such as ecological speciation [1], [5], [7], [8] that are not dependent on allopatry, there have been few critical tests of allopatric speciation in systems which are regarded as exemplifying the process, such as island archipelagos [9], [11]–[17]. This is primarily because, from a contemporary perspective, genetic isolation cannot be assessed in spatially isolated populations. However, a historical perspective allows us to test the genetic isolation of anole species isolated for a very substantial time before their islands coalesced.
Anolis (small insectivorous lizards) is the most speciose amniote genus (circa 400 species) [18] and show little inter-specific hybridization [19]. Just two colonizations of the Caribbean islands have resulted in 150 species, so they may be thought of as exemplifying allopatric speciation in island archipelagos [11]–[14], [18], [20]. These anole radiations appear to have inhabited the Lesser Antilles since the origin of the younger island arc, or just before (i.e circa 8–9 mybp) with a southern and a northern series [21]. On what is currently recognized as the island of Martinique (southern series), the paraphyletic anole Anolis roquet has deep phylogeographic divisions, with Anolis extremus from Barbados nested within it [21]. Geology [22]–[23], molecular phylogeography and molecular clock analysis [21] reveals that four precursor islands of Martinique (Figure 1) are associated with four mtDNA lineages of ‘A. roquet’. The island ages, molecular clock and geographic distribution of the lineages link closely to suggest that the precursor islands of Martinique (together with Barbados) had separate anole allospecies for up to about 8mybp, before central uplifting joined the Martinique precursors to form a single island (with Barbados remaining independent) [21]. This gave three secondary contact zones in Martinique (Figure 2) between previously allopatric forms (south-central, SW-central, NW-central) that:- 1) are phylogenetically deeper than the species-level split between A. extremus (consistently regarded as a valid species [18]) and its sister clade within A. roquet [21]; 2) have diverged a substantial time ago (6–8 mybp) and have a level/time of phylogenetic divergence that is comparable to other Lesser Antillean anole species [11]–[12]; 3) may show distinct mtDNA lineages with almost no haplotype inter-digitation [21]; and 4) may show a prima facie case for parapatric bimodality in multivariate quantitative traits at some points of contact [24] (Figure S1). Hence, this is an appropriate test of allopatric speciation.
Martinique anoles also provide a test for ecological speciation, or isolation by adaptation [7]. The quantitative traits of Lesser Antillean anoles adapt by natural selection to environmental zonation, as shown by common garden and natural selection experiments [25]–[26], parallels among island species, and correlation studies that take phylogenetic history into account [27]. In Martinique, the montane rainforest and coastal xeric woodland are distinctly different habitats with pronounced differences in the environmental conditions across the ecotone between them. As with other Lesser Antillean anoles, the Martinique anole adapts to these conditions and their populations show marked habitat-related differences in quantitative traits such as morphology (shape, color, pattern and scalation) and dewlap hue [21],[24], resulting in distinct ecotypes. The ecotone between these coastal xeric and montane rainforest habitats provides a test for ecological speciation for comparison with secondary contact zones.
Hence, with the Martinique anole there is the potential for speciation to occur in accordance with both an allopatric model (where the different lineages on precursor islands speciate), and an ecological model (where the different ecotypes speciate). Preliminary analysis of a single transect suggested that under specific circumstances there may be greater restriction of genetic exchange between habitat types than previously allopatric forms in secondary contact [21]. However, this analysis examined just one of the three pairs of coalescing islands (northwest vs. central), under only one set of selection regimes (strong convergent selection for montane rainforest on both lineages where they met along that transect). Furthermore, this study was not replicated and no control was used, limiting the capacity to generalize, and raising several questions. Specifically, do the other pairs of coalescing precursor islands populations (southwest-central and south-central) show evidence of genetic isolation or not; is the pattern of inter-digitation of the mtDNA lineages and introgression of the neutral nDNA consistent along the length of each of the three secondary contact zones, or does it vary dependant on other factors; how do the selection regimes along the transect influence the extent of genetic isolation among previously allopatric forms; under what ecological conditions (extent and abruptness of habitat change) is there restricted genetic exchange among habitat types and how long does it take to develop?
To answer these questions we investigated the xeric/rainforest ecotone and all three cases of island coalescence, each with two to four replicate transects, together with a control transect (Figure 2, Table 1, Table 2). By measuring nuclear genetic structure, mtDNA lineage, quantitative traits and climate variation along these replicated transects, across both geological and habitat contact zones, we are able to critically test the role of these two factors, and their interaction, in the differentiation of island anoles. We show that, although there is always a signature of past allopatry in the nuclear genetic structure, and this can be quite strong dependent on the comparative selection regimes across the secondary contact zone, there is no complete allopatric speciation for any of the three allopatric pairs. Instead, if there is sufficient magnitude and abruptness of habitat change, then there is even greater differentiation across the ecotone, and this can develop over a brief period of time. Although the ecological speciation is not complete, it has reached what has been characterized as a “later stage” in the speciation continuum [7]. This supports a relatively important role for ecological speciation under the appropriate circumstances [1], [5], [7], [8].
Recent island-wide phylogenetic studies identified four main mtDNA lineages within A. roquet whose geographical limits correspond very closely to the geological junctions between precursor islands [21], [28], with the timing of divergence between these lineages compatible with the age of the different precursor islands. This supports the scenario illustrated in Figure 1B, which suggests that the individual lineages evolved in allopatry for about 6mybp (central-south) to 8 mybp (central southwest and central northwest) until the precursor islands merged to form present day Martinique. Here, we use a large sample per site, with sites along transects focussed on the contact zones. Estimating the frequency of mtDNA lineages at localities along these transects enables us to test for any inter-digitation of the lineages and the fit between the distribution of the lineages and the precursor islands at this fine scale.
With the exception of transect VIII (φ = 0.61), we observed a very close association between the precursor islands and the mtDNA lineages (0.71<φ<0.95) and little, or almost no (transects I,IV), inter-digitation, even at this fine spatial scale (Table 1, Figure 3, Figure 4, Figure 5). This absence of substantial inter-digitation, despite a relatively long period of contact (the precursor islands merged about 1 Mya [21]), implies the absence of extensive female-driven gene flow [29] between these previously allopatric lineages.
Climate is a strong determinant of habitat and can be objectively measured and quantified. The results (Table 2) show strong climatic variation along the transects (III, IV) that run from the xeric coast to the montane rainforest with a sharp transition (ecotone) between these habitats (Figure 3). Other transects generally run within habitat types and show more subtle climatic variation, i.e., are without abrupt changes in habitat of a high magnitude.
A wide-ranging multivariate profile of the quantitative traits (QTs) of individuals was taken (these include both spectrometric dewlap hues [21] and morphological traits such as colour pattern, body dimensions and scalation) to estimate the change in QTs along a transect in relation to habitat type, ecotone and lineage. Lesser Antillean anole quantitative traits (QTs) are generally tightly linked to the habitat and have generally been shown to adapt to environmental conditions and reflect selection regime rather than phylogeographic lineage [24]–[28]. Hence, as predicted, the large magnitude of habitat variation in transects III and IV is matched by a high magnitude of QT variation (16–17 within group standard deviations, Table 2) with highly divergent rainforest and xeric ecotypes (Figure 2 and Figure 3), and a close correlation between QTs and climate variation along the transect (r = 0.96 to 0.97, Table 2). The large magnitude and close association of the climate and QT variation indicates the potential importance of the ecotone in determining population structure. Elsewhere, where the magnitude of climatic variation along a transect is less (because they largely run within habitat types), such a very high correlation between climate and QTs is not predicted or observed. Even so, the correlation is only insignificant in one non-control transect (Table 2), once again suggesting the general importance of habitat type in determining quantitative traits.
The correlation between quantitative traits and lineage frequency is significant along four transects (Table 1), and is particularly high in transect I, where, on this spatial scale, there is no overlap between multivariate morphology of morphs either side on the lineage contact zone (Figure S1).
The allopatric model of speciation predicts that the four lineages that spent a substantial time in isolation on separate islands (divergence at circa 6–8my) should all be reproductively isolated entities. That is, there should be four species with very little (if any) gene exchange among them where they meet along all three contact zones (northwestern/central, southwestern/central and southern/central) on what is currently Martinique. The very close association (with only one exception) between the precursor islands and the mtDNA lineages along the transects does not contradict this (Table 1).
This prediction was tested by estimating the population structure along transects I–VIII using neutral, hypervariable, nuclear microsatellite markers, primarily analysed by Bayesian assignment, with support from AMOVA and standardised FST′ values. Principal component analysis (PCA) provides an independent perspective on the population affinities. The analysis of these markers along the replicated transects across these three zones clearly rejects the presence of reproductively isolated (or even partially isolated) species. This is the case for all three precursor island contacts: the central/northwest contact (Figure 3), the central/southwest contact (Figure 4), and the central/south contact (Figure 5) along the length of the contact for all replicates and various types of selection regimes (but see transect I below). The Bayesian assignment method detects two clusters in most transects (Table 3, Table S1), but the transition between the two clusters generally is not closely associated with the lineages and/or forms a smooth cline (Figure 3, Figure 4, Figure 5). The PCA (Figure S2) supports the Bayesian clusters in transects I–VIII as the pattern of relative frequency of the Bayesian clusters (where K = 2) is almost identical to the pattern of PC1 scores for each transect (r = 1.0 with one exception).
The association between nuclear genetic clusters and allopatric speciation model (lineage categories) is, with one exception, modest (0.32<φ<0.51) even if significant (Table 3) and φ does not approach unity (complete isolation). This pattern is supported by the AMOVA and standardized FST′ values. The AMOVA show sporadic significant structure associated with lineages (transects I, II, V, VII), but all the ΦCT values are substantially less than unity and too low for reproductive isolation (Table 4). Similarly, mean standardized genetic differentiation between pairs of populations on each side of lineage boundaries is low to moderate (0.072<FST′<0.166, Table 4). This suggests high levels of nuclear gene exchange between lineages (the equivalent unstandardized FST values are 0.014<Fst<0.043). Of particular interest are transects III and IV where the nuclear genetic structure associated with the northwest and central lineages can be compared directly with that associated with habitats (Table 3 and Table 4). Here the Bayesian clusters show substantially poorer fit to the allopatric speciation model (0.32<φ<0.37) than the habitat categories (0.62<φ<0.71). The AMOVA shows low (−0.00064<ΦCT<0.00158) and insignificant ΦCT for the lineage categories, but higher (0.03481<ΦCT<0.01665) and significant ΦCT for the habitat categories. The mean standardized genetic differentiation is also much lower between lineage categories (0.072<FST′<0.075) than habitat categories (0.137<FST′<0.213).
In general, while there may be a nuclear genetic signature of past allopatry for all four mtDNA lineages associated with precursor islands, there is no allopatric speciation. The partial exception to this general trend is transect I, where the central and northwest lineages meet on the northeast coast. Here, there is almost no inter-digitation of mtDNA lineage markers (φ = 0.95), and a sharp stepped cline in quantitative traits at the junction of the precursor islands (Figure 3B). There is some genetic isolation between the lineages as shown by the Bayesian assignment (φ = 0.68, Figure 3C), although neither AMOVA nor standardized FST′ values are exceptionally high (Table 4). If these lineages were equally isolated along their entire secondary contact zone there might have been a rather weak case for partial allopatric speciation and recognition of their status as separate species. However, they are not. Even along the adjacent transect (II) in the transitional forest, which is only 5km inland, the lineages show little genetic isolation (no Bayesian clusters, Table 3, Table S1, Figure 3C) and do not have distinct quantitative traits (Figure 3B). Further along this secondary contact zone in the montane rainforest (transects III, IV) the quantitative traits are identical either side of the secondary contact zone with little nuclear genetic isolation estimated from Bayesian assignment, AMOVA or standardized FST′. Direct experimental measures of selection in Lesser Antillean [25], [26], and other [18], [30] anoles, as well as other studies of adaptation [21], [24], [27], have shown strong selection intensity on anole quantitative traits, and the pattern of climate variation and QT variation along transects differs among transects I–IV. Hence, although the populations from transects I–IV may broadly share the same history (particularly adjacent transects I and II), they differ in the pattern and intensity of selection along the transect. The similarity of the environment either side of this secondary contact zone in the rainforest (transects III, IV, Figure 3B), and the remarkably parallel appearance of these northwestern and central lineages forms in the rainforest [28] (Figure 2, image 3) suggests strong convergent selection working on these populations. Along the coastal transect (I) there may be no such strong convergent selection, and indeed the environmental variables show a smooth cline along the transect so there may be some divergent selection. This suggests that the persistence of a strong genetic signal of past allopatry may be contingent on the pattern of selection regimes.
In conclusion, even though there has been a substantial period of allopatric divergence between northwest/central (8 mybp), southwest/central (8 mybp) and south/central (6 mybp) lineages, and only restricted inter-digitation of the mtDNA, there is no evidence of complete allopatric speciation even though there may be a significant signal of past allopatry. This is consistent across all three pairs of putative allospecies and between the replicates along the length of all three contact zones, irrespective of the pattern of selection regimes. Nevertheless, if the pattern of selection allows, a stronger signal of past allopatry may be retained. Overall, the results are compatible with divergence in allopatry followed by substantial introgression on secondary contact due to a lack of reproductive isolation.
The distinctly different habitats of the xeric coast and the montane rainforest, associated with strongly divergent quantitative traits, provide an opportunity to test for ecological speciation along transects III and IV (Table 2). Bayesian assignment indicates that both transects have restricted genetic exchanges across the xeric-montane ecotone, although this is stronger in transect IV (φ = 0.71) than III (φ = 0.62). The populations of Anolis in the area of transect III (Figure 2) were most likely severely impacted by the 1902 pyroclastic surge that destroyed St Pierre [31]. Although the reinstatement of the reduced gene exchange associated with the ecotone may have been facilitated by ecotypes colonizing the vacant area from adjacent populations of the same altitude, anoles can readily colonize adjacent areas of different attitudes. Consequently, perhaps to some extent, the signal of restricted genetic exchange may have to have developed in circa 100 years, which is likely to be much shorter than elsewhere along this ecotone, and may be too short even for ecological differentiation in these terrestrial amniotes [32].
The results of the AMOVA also support a reduction of gene exchange between habitats for the two transects. This test shows a significant structure when the sites are grouped according to their habitat, but not when they are grouped according to their lineage (Table 4, Table S1). Similarly, the mean standardized genetic divergence (FST′) is much higher between habitats than lineages (see above). Even if this is not full reproductive isolation, the restriction of gene exchange between the habitats is very substantial, and along transect IV it is greater than any in this study. Moreover, (with the above caveat regarding altitudinal restrictions on re-colonization) it may be capable of developing rapidly as transect III shows greater isolation than associated with allopatric divergence with the AMOVA and (with one exception) the goodness of fit (φ) statistics.
Nosil et al [7] recognise several stages in the continuum of ecological speciation: 1) population differentiation, 2) ecotype formation, 3) speciation and 4) post-speciation divergence. They suggest that increased genotypic clustering (as evidenced here) indicates a later stage of the speciation process, and the degree of genetic isolation here is as great, or greater, than that associated with their [7] example of the most reproductively isolated Pundamilia cichlid pairs. Moreover, the adjacent, and environmentally very comparable, island of Dominica also has distinct anole ecotypes. A study of microsatellite variation among anole populations on Dominica did not indicate genetic clustering of the ecotypes [27], so Martinique anoles appear to be at a later stage than the stage 2 of the Dominican ecotypes. Hence, although it is clear that this there is no full ecological speciation here, it appears that the Martinique anoles are between the ecotype (2) and speciation (3) stages in the ecological speciation continuum. It may be that the situation is in equilibrium, or is a stage in a progression towards greater isolation. Moreover, even if progression to greater isolation was possible, it could be prevented by persistent volcanic disturbance of the ecotone and/or its spatial discontinuity.
Both natural and sexual selection may play a role in this ecological pattern of gene exchange as predation pressure for crypsis [33] may interact with the need for conspecific communication. Substantial work on Lesser Antillean anole ecotypes, including natural selection experiments, indicates that a wide range of character systems, rather than just single characters, adapt these ecotypes to the specific biotope [24]–[27]. Hence, natural selection will be impacting many independent traits [7]. Moreover, sensory drive may be important [34] as these habitats have different light conditions which may impact on visual conspecific communication via secondary sexual traits, including dewlap hue. If assortative mating occurs, where a female preferentially chooses a male with the appropriate pattern and hue for that habitat, then this could result in reduced gene exchange among populations in different habitats.
This replicated population genetic study robustly and consistently suggests that, across a range of opportunities and conditions, there is pronounced introgression after allopatry and that even a very substantial amount of time in spatial isolation does not, on its own, necessarily allow for the development of reproductive isolation and speciation. This is all the more notable as fertile, natural inter-specific hybrids are extremely rare in this large, well-studied, genus [18], [19], and this is a radiation that is generally regarded as exemplifying allopatric speciation [11]–[14], [18], [20]. Even though the habitat forms are partially, rather than completely, reproductively isolated, they can show greater isolation than the putative allospecies, and it may be that this can develop rapidly. In addition, the extent of the genetic signature of past allopatry may be dependent on the pattern of selection regimes across the secondary contact. These observations have implications for animal speciation in general and speciation in anoles in particular. While one could choose to emphasize the lack of complete ecological speciation in this case, we believe these observations reveal the potential importance of ecological divergence as a contributory factor in speciation, including in situations where ecological divergence initiates speciation, but does not complete it [7], and where allopatry is important, but adaptation to environmental differences are also required, as recently suggested for speciation in birds [35]. Consequently, a role for ecology in speciation, including ecological speciation, or isolation by adaptation [1], [5], [7]–[8], [32], [36]–[38], may be of widespread relevance, and non-allopatric models [10] should not be excluded from consideration. These implications are particularly relevant to the most speciose amniote genus, Anolis, including the large Greater Antillean communities, where sympatric and parapatric speciation have been regarded as not being an important phenomena in anole evolutionary diversification [14], [18], [20]. Finally, it contributes to an explanation of why there are so few species of Anolis in the Lesser Antilles compared to the Greater Antilles [14]. At the stage of the allopatric model where species number on an island is increased by colonization from other islands [14], the colonizers interbreed with the species already on the island, because no reproductive isolation has developed while they are in allopatry. The genetic signal of this interbreeding is then lost because the number of overseas colonizers per unit time will be vanishingly small compare to the turnover in the large endemic population.
Replicate transects were taken across each precursor island junction (Figure 2); northwest lineage to central lineage transects I, II, III and IV, southwest lineage to central lineage transects V and VI, south to central lineage transects VII and VIII, with a control transect (IX) within the central lineage. The number of sites per transect was 8, 8, 7, 7, 9, 9, 8, 7 and 5 respectively for transects I to IX. At each site 48 naturally autotomized tail-tip biopsies were sampled for molecular analysis, while quantitative traits and dewlap hue were recorded from ten adult males. Where transects crossed the same lineages and were in broadly comparable habitats (eg, III+IV, V+VI, VII+VIII) samples were collected, and data was recorded and analysed in these transect pairs.
The lineages were first investigated using complete cytochrome b sequence from the mtDNA. PCR-RFLP analyses were then designed to efficiently assign numerous individuals to a specific lineage (northwest, southwest, south or central). The cyt b fragment used in the phylogeographic analysis was digested after amplification using the restriction enzyme SspI (New England Biolabs) for 3 hours at 37°. The digested products were run on a 2% agarose gel containing ethidium bromide. This enzyme distinguishes between the central lineage (uncut by this enzyme), the southern lineage (cut at position 598) and the clade comprising the southwestern and the northwestern lineages (cut at position 166). To further distinguish between southwest and northwest lineages, we digested the same fragment using the restriction enzyme DraI (New England Biolabs) that cuts the PCR products from the northwest lineage at position 227, while those from SW lineage were uncut by this enzyme.
The habitat type at each site was estimated from a multivariate climatic profile using nineteen climatic variables from Worldclim (http://www.worldclim.org/). These variables were annual mean temperature, mean diurnal range, isothermality, temperature seasonality, maximum temperature warmest month, minimum temperature coldest month, temperature annual range, mean temperature wettest quarter, mean temperature driest quarter, mean temperature warmest quarter, mean temperature coldest quarter, annual precipitation, precipitation wettest month, precipitation driest month, precipitation seasonality, precipitation wettest quarter, precipitation driest quarter, precipitation warmest quarter, and precipitation coldest quarter. Logarithm (natural) transformed data was subjected to principal component analysis. The component defining the climatic trend along the transect was plotted and the magnitude of climatic change in this trend can be taken as the range between maximum and minimum component scores. If there was an ecotone the cut-point between habitat types was defined as the midpoint between these maximum/minimum component scores.
A multivariate suite of 21 morphological characters (colour pattern, trunk hue, scalation, body dimensions) were recorded [21], [39]. The hue of the anterior and posterior dewlap was recorded using reflectance spectrometry [21] and the spectrum of each was divided into 6 independent hues following a multiple-group eigenvector procedure [21], [40]. The morphological and spectrometric characters were then subjected to canonical analysis with the CVs scaled so that the pooled within-group standard deviation was unity. Heteroscedasticity was a problem with transect III so, as an alternative, a principal component analysis was also run on normalized site means for this transect.
The samples were genotyped at nine nuclear microsatellite loci (AAE-P2F9, ABO-P4A9, AEX-P1H11, ALU-MS06, ARO-035, ARO-062, ARO-065, ARO-120, ARO-HJ2) [41]–[43] in a single multiplex using a Qiagen Multiplex PCR kit with the annealing temperature at 55°. PCR products were then analysed on an ABI 3130xl genetic analyser and the genotypes scored using Genemapper v4.0 (Applied Biosystems). Hardy-Weinberg equilibrium and linkage disequilibrium were tested for using Genepop v3.4 [44]. After Bonferoni correction, there were no consistent departures from Hardy-Weinberg equilibrium, or linkage disequilibrium. Only one locus in one population showed a significant departure from Hardy-Weinberg equilibrium (transect I, site 8 for locus ARO-HJ2), and there was only one significant association between loci ALU-MS06 and ARO-035 in one population (transect IV site 3).
The primary genetic structure along each transect was studied using Bayesian clustering performed by the program STRUCTURE v2.1 [45]. We defined the number of populations (K) from 1 to 9 and 10 independent runs were performed for each value of K using the admixture model, a burn-in of 100,000 steps followed by 400,000 post burn-in iterations. We determined the optimal number of populations using the maximum value of the posterior probability of the data [45]. We also used AMOVA, performed by Arlequin v3.11 [46], to test for genetic differentiation predicted by alternative speciation models. Within each transect populations were grouped by modal lineage, or, where appropriate, by habitat. For two transects (III, IV), where both types of speciation could have occurred, this allowed direct comparison of competing speciation hypotheses. Finally, the mean genetic differentiation among populations either side of a lineage, or habitat, boundary along a transect was estimated by calculating the mean standardized pairwise FST′ using RecodeData v0.1 [47] and FSTAT v2.9.3 [48]. To give an independent perspective on the population affinities revealed by the Bayesian clustering we performed principal component analysis (PCA) of transect site gene frequencies using PCAGEN [49]. For each transect the PC1 site scores were compared to Bayesian site frequencies (where K = 2) by correlation.
The relationship between lineage, genetic isolation, past allopatry, ecotone, climate, and adaptive quantitative traits was investigated at sites along a series of replicated transects (Figure 2) across the secondary contact zones (transects I to VII) and ecotone (transects III and IV). Transect IX did not cross any lineage boundary or ecotone and was used as a control transect.
|
10.1371/journal.pntd.0000614 | Bacillus thuringiensis Cry5B Protein Is Highly Efficacious as a Single-Dose Therapy against an Intestinal Roundworm Infection in Mice | Intestinal parasitic nematode diseases are one of the great diseases of our time. Intestinal roundworm parasites, including hookworms, whipworms, and Ascaris, infect well over 1 billion people and cause significant morbidity, especially in children and pregnant women. To date, there is only one drug, albendazole, with adequate efficacy against these parasites to be used in mass drug administration, although tribendimidine may emerge as a second. Given the hundreds of millions of people to be treated, the threat of parasite resistance, and the inadequacy of current treatments, new anthelmintics are urgently needed. Bacillus thuringiensis (Bt) crystal (Cry) proteins are the most common used biologically produced insecticides in the world and are considered non-toxic to vertebrates.
Here we study the ability of a nematicidal Cry protein, Cry5B, to effect a cure in mice of a chronic roundworm infection caused by the natural intestinal parasite, Heligmosomoides bakeri (formerly polygyrus). We show that Cry5B produced from either of two Bt strains can act as an anthelmintic in vivo when administered as a single dose, achieving a ∼98% reduction in parasite egg production and ∼70% reduction in worm burdens when delivered per os at ∼700 nmoles/kg (90–100 mg/kg). Furthermore, our data, combined with the findings of others, suggest that the relative efficacy of Cry5B is either comparable or superior to current anthelmintics. We also demonstrate that Cry5B is likely to be degraded quite rapidly in the stomach, suggesting that the actual dose reaching the parasites is very small.
This study indicates that Bt Cry proteins such as Cry5B have excellent anthelmintic properties in vivo and that proper formulation of the protein is likely to reveal a superior anthelmintic.
| Intestinal parasitic nematode diseases infect over one billion people and cause significant disease burden in children (growth and cognitive stunting, malnutrition), in pregnant women, and via their dampening of the immune system in infected individuals. In over thirty years, no new classes of anti-roundworm drugs (anthelmintics) for treating humans have been developed. Because of limitations of the current drugs and the threat of parasite resistance, new anthelmintics are needed. The soil bacterium Bacillus thuringiensis (Bt) produces crystal (Cry) proteins that specifically target and kill insects and nematodes and is used around the world as a safe insecticide. Here we test the effects of the Bt Cry protein Cry5B on a chronic, natural intestinal roundworm infection in mice, namely the helminth parasite Heligmosomoides bakeri. We find that a single dose of Cry5B can eliminate 70% of the parasites and can almost completely block the ability of the parasites to produce progeny. Comparisons of Cry5B's efficacy with known anthelmintics suggest its activity is as good as or perhaps even better than those currently used. Furthermore, this protein is rapidly digested by simulated stomach juices, suggesting that protecting it from these juices would reveal a superior anthelmintic.
| Neglected tropical diseases (NTDs) have a worldwide devastating impact on the lives of billions of people. Helminth infections comprise approximately 85% of the NTD burden [1]. The top three ailments on this list of NTDs are all caused by intestinal nematodes [2]. These infections consist of ascariasis (caused by Ascaris lumbricoides), trichuriasis (caused by Trichuris trichiura or whipworm), and hookworm disease (caused by Necator americanus and Ancylostoma duodenale). Approximately 807-1,221 million people are afflicted with ascariasis, 604–795 million with trichuriasis, and 576–740 million with hookworm infections [3]. The widespread and detrimental effects of parasitic worm infections on human growth, nutrition, cognition, school attendance and performance, earnings, and pregnancy have been well documented [2],[3]. These infections also contribute to increased severity/infectivity of HIV/AIDS, malaria, and tuberculosis due to compromised immune responses [3],[4]. Furthermore, parasitic nematode infections confound vaccination efficacy [5],[6]. Despite the high prevalence and destructive nature of these infections, there are few treatment options. Although four anthelmintics (levamisole/pyrantel and mebendazole/albendazole) are approved by the World Health Organization for use in humans, one, albendazole, is generally preferred in a single-dose regimen over the others since it is relatively more effective against hookworms and whipworms [7],[8]. However, resistance to albendazole may already be appearing [9],[10]. Furthermore, the reliance upon one compound for treating hundreds of millions of people will have devastating consequences if widespread resistance ever becomes a reality. Tribendimidine, developed by the Chinese Centers for Disease Control and Prevention, is emerging as a second anthelmintic with efficacy similar to albendazole, but is a member of the levamisole/pyrantel class to which resistance in human populations has been reported [11],[12],[13]. Furthermore, none of the compounds have been shown to be totally effective against all helminth infections [8]. Consequently, there is an urgent need for efficacious, safe, inexpensive, single-dose anthelmintics with new mechanisms of action.
This search for new anthelmintics has led to examination of Bacillus thuringiensis (Bt) crystal (Cry) proteins. These proteins are the most extensively used biologically-produced insecticides in the world [14]. Bt is a soil bacterium that produces crystal inclusions during sporulation. These inclusions contain Cry proteins that are highly toxic to some invertebrates but nontoxic to humans and other vertebrates [15]. The high efficacy against insects, absence of toxicity towards vertebrates, and low production cost of these proteins has led to their widespread use in pesticides and in transgenic crops [14]. So far, three Bt Cry proteins toxic to a broad range of free-living nematodes and the free-living form of at least one intestinal parasitic nematode have been discovered, including: Cry5B, Cry14A, and Cry21A [16]. Cry13A may also have anti-nematode activity [17]. To date, only one of these, Cry5B, has been shown to be therapeutic in vivo with activity against intestinal hookworm parasite (Ancylostoma ceylanicum) infections in hamsters when delivered daily, per os, over the course of three days [18].
These studies suggest that Cry proteins could provide therapy for intestinal nematode infections. However, it remains to be shown that Cry5B can effect a cure against more than A. ceylanicum infections in hamsters or that Cry proteins are efficacious as single-dose anthelmintics. Heligmosomoides bakeri (formerly known as Heligmosomoides polygyrus and Nematospiroides dubius) is one of the most widely studied rodent intestinal parasite nematodes [19],[20]. The nematode has a high infection rate and is the best model for chronic intestinal nematode infections in immunocompetent mice. H. bakeri has also played a key role in the history of anthelmintic development via its use in the discovery of ivermectin [21]. In addition, H. bakeri infections in mice are a naturally occurring infection, unlike A. ceylanicum infections in hamsters. Thus, curative experiments in H. bakeri are complementary to those in A. ceylanicum, yielding important information as to how Cry proteins may fare against a broad range of natural intestinal parasites in vivo. Herein, we report our investigations into single-dose Cry5B therapy against H. bakeri.
Female Swiss Webster white mice were purchased from Harlan Laboratories and were infected at approximately 6 weeks of age at an average weight of 25g. Mice were provided with food and water ad libitum.). This research was approved by the UCSD Institutional Animal Care and Use Committee (IACUC), protocol number S08140. The maintenance and care of experimental animals complied with the University of California's Animal Care Program's guidelines for the humane use of laboratory animals.
Crystal-deficient Bt strains HD1 and 4Q7 were transformed with a plasmid containing the Cry5B gene [23]. Bacillus thuringiensis subspecies kurstaki HD1-4D8 was ordered through the Bacillus Genetic Stock Center. Spore lysates (SLs; HD1 and 4Q7 Cry-deficient strains) and spore-crystal lysates (SCLs; HD1 and 4Q7 transformed with Cry5B plasmid) were prepared using standard methods and then stored at −80° [23]. Bioactivity of SCLs was confirmed against Caenorhabditis elegans by a mortality assay over 24 h at 25°C. SLs (Cry-minus) were confirmed to lack toxicity against C. elegans. On the day of use, SL and Cry5B SCL aliquots were thawed and centrifuged at 4,500 rpm for 15 minutes at 4°C and the supernatant was removed. The pellet was then resuspended in distilled water to a final concentration of 2.5 mg/mL, for the HD1 strain, and 2.25 mg/mL, for the 4Q7 strain (protein concentrations were determined by comparing Cry5B band intensities for four different aliquots of SCLs to known amounts of bovine serum albumin on Coomassie-stained 8% SDS polyacrylamide protein gels). The placebo SL control strains were concentrated to the same extent. The samples were kept on ice until gavage.
On day 0, mice were infected per os with a suspension of 200±10 H. bakeri L3 larvae in 0.1 mL of distilled water. Larvae were counted under the microscope, then drawn into a pipette tip and placed into separate glass test tubes until gavage with a blunt-ended syringe. On days 14, 16, 18, and 20 post-infection (P.I.), fecal samples were collected from the mice. Mice were placed individually in empty plastic cages for 1 h each morning, and the fecal pellets were collected into 50 mL centrifuge tubes. The number of eggs present was counted using the modified McMaster technique [22]. Briefly, feces collected from mice were weighed and resuspended in a 1 g:15 mL volume of water. The pellets were allowed to soak overnight before being broken up for 1 h via heavy vortexing. The eggs were counted using a 2-chamber McMaster slide, each chamber holding a 0.6 mL volume of a 1∶1 mixture of fecal slurry and saturated sucrose solution. The number of eggs per gram of feces was thus calculated from the following equation: number of eggs counted x (1/0.3 mL slurry) x (15 mL slurry/g feces). For each mouse and each time point, three different egg counts were made and then averaged.
Each mouse was treated per os on day 15 P.I. with 0.1 mL of relevant treatment (placebo or Cry protein) through a blunt-ended syringe. All mice were killed by exposure to CO2 on day 20 P.I. and the intestines were removed in their entirety. These were opened longitudinally with a pair of blunt-ended dissecting scissors and then placed into a 50 mL centrifuge tube with 10–20 mL of pre-warmed (37°C) PBS for approximately 1 h to allow worms to dislodge from the intestine. The solution and intestine were examined under a microscope, using fine tweezers when necessary for further extrication of worms from the intestine, for determination of final worm burden.
Tribendimidine was kindly provided by Dr. Shu-Hua Xiao at the Chinese Centers for Disease Control and Prevention. The drug was suspended in 20 mM citrate buffer pH 7.3 and delivered per os on day 15 P.I. in a total volume of 0.1 mL as per Cry5B experiments. For these curative experiments, the mice were infected with on average 150 L3 larvae (six/group except for placebo group, which only had five mice). Placebo control for these experiments was 0.1 mL of buffer only.
Simulated gastric fluid (SGF) was prepared freshly as described in the United States Pharmacopeia and stored at 4° until use [24]. Cry5B SLC was added to a 1 mL solution of SGF for a final concentration of 2.5 mg/mL and incubated at 37°C [25]. 50 µL aliquots of the digestion stock were removed at each time point as the digestion solution was agitated. Each aliquot was immediately quenched by neutralization with 15 µL of 0.2 M sodium carbonate per 50 µL of SGF [25]. Quenched samples were kept on ice until 2x SDS-PAGE loading buffer was added to each sample. Mixtures were then heated for 5 min in boiling water and stored at −20°C until analysis.
Data analysis of intestinal worm burdens and fecal egg counts was carried out and plotted using Prism 5 (GraphPad Software Inc., La Jolla, CA, U.S.A.). For worm burdens, average indicates the average worm burdens amongst all the mice in each treatment group. For fecal egg counts, average indicates the egg count per mouse averaged from all mice in the group at a given time point. Fecal egg count data was analyzed via pair-wise comparisons between groups and days through two-way analysis of variance (ANOVA) with repeated measures and Bonferroni post tests. Results were as follows: FTreatment = 70.69, degrees of freedom (df) = 1, P<0.0001; FTime = 5.241, df = 3, P = 0.003; FInteraction = 11.43, df = 1,3, P<0.0001. Worm burdens for Cry5B treatment versus placebo were compared using Mann-Whitney U test (one-tailed). Values are as follows: U = 7.5, P = 0.0007 for HD1 Cry5B versus placebo; U = 2.0, P = 0.0012 for 4Q7 Cry5B versus placebo. Worm burdens for tribendimidine experiment were compared using one-way ANOVA and Tukey's Multiple Comparison Test (F = 9.387, df = 3, 19, P = 0.0005).
To determine if Cry5B could provide therapy as a single-dose anthelmintic against H. bakeri, we treated H. bakeri-infected mice with Bt spore-crystal lysates expressing or not expressing (placebo control) Cry5B. When the bacterium Bt sporulates, it produces spores, large crystal protein-containing inclusions, and lysate produced when the mother cell that gives rise to the spore and crystal lyses upon completion of sporulation. Bt spore-crystal lysates (SCLs) from many Bt strains, including Bt kurstaki HD1 that targets caterpillars (Lepidoptera), have been extensively tested against mammals (including humans) and found to be non-pathogenic [15],[26],[27]. We transformed a crystal protein-minus HD1 strain with a Cry5B-expressing plasmid. Twenty mice were infected with H. bakeri larvae. Fifteen days post-infection (P.I.), we delivered into each mouse per os either a single 0.1 mL dose of Cry5B-containing HD1 SCLs (715 nmoles/kg or 100 mg/kg of Cry5B) or, as a placebo control, a single 0.1 mL dose of spore lysates (SLs, crystal-minus) from the parent, untransformed HD1 strain. Beginning the day before treatment (day 14 P.I. or day -1 treatment), and then continuing every other day (day 1, 3, 5 post-treatment), we collected fecal samples from each mouse to measure parasite progeny production (eggs/gram of feces). Five days after treatment (day 20 P.I.), the mice were euthanized and the total number of parasites present in the small intestine tallied.
With regards to progeny production, we found that on the day prior to treatment, the parasites in both groups of mice (placebo treated and Cry5B treated) were producing statistically indistinguishable amounts of eggs (Figure 1, Table 1). At days 1, 3, and 5 post-treatment, the placebo group showed no reduction in egg production, consistent with the hypothesis that the parent Bt strain alone has no effect on the parasites. In contrast, a rapid and remarkable reduction in egg production took place in the Cry5B-treated animals, resulting in a 95%, 99%, and 98% reduction on days 1, 3, and 5 post-treatment respectively (Figure 1, Table 1). With regards to parasite clearance, we found that the single dose of Cry5B achieved a remarkable therapeutic effect, clearing away 67% of the parasites relative to placebo control (Figure 2A, Table 2). Thus, a single dose of Cry5B has strong effects on parasite reproduction and the ability of parasites to maintain an infection.
The reduction in fecal egg count (>97%) was much larger than would be expected from the final mouse worm burden of the SCL-treated animals (67% cleared). There are at least two possible explanations for this—either the treatment was affecting the status of the worms so that any worms left behind were severely compromised in health or the treatment was preferentially eliminating female over male parasites. To distinguish between these possibilities, we made a note of the number of females present in placebo versus Cry5B treated controls during the counting of the worm burdens. In placebo treated mice we found that there were 35.3±5.9 (standard error of the mean, or sem) females while in the Cry5B treated mice there were 7.8±2.3 (sem) females per mouse intestine. Thus, there was a 78% reduction in the number of females present. This drop, although greater than that for males (51% reduction), does not seem sufficient to account for the observed >97% drop in egg production seen. These data suggest that the parasites that remained in the intestine were severely compromised in health.
We also determined if this capacity to clear an infection was dependent upon a particular Bt strain. We performed a similar curative experiment, measuring intestinal worm burdens after treatment, using the Bt strain 4Q7 (derived from Bt israelensis, which targets Diptera) either transformed with the Cry5B-expressing plasmid or untransformed. Fourteen mice were infected with H. bakeri larvae, and fifteen days P.I. a single dose of 0.1 mL Cry5B-containing 4Q7 SCLs or 0.1 mL 4Q7 SLs (crystal-minus) were delivered per os. The dose delivered per mouse was 644 nmoles/kg (90 mg/kg). We found a similar therapeutic effect as above—71% of the parasites were cleared relative to placebo control (Figure 2B, Table 2). We note that in this experiment the total number of parasites present in the small intestine in placebo control animals was greater than in the first experiment. The variability appears to be due to relative infectivity of different batches of L3 parasite larvae. These data demonstrate that, regardless of parent Bt strain and of the initial parasite load, a similar single dose of Cry5B is able to achieve comparable therapeutic effect.
These results are significant when the relative efficacy of Cry5B is compared to other standard anthelmintic treatments. Published reports, employing a treatment timeline against H. bakeri parasites that is similar to our own, show that levamisole (10 mg/kg or 49 µmoles/kg delivered on day 12 P.I.) effected a 90% reduction in worm burdens and ivermectin (5 mg/kg or 5.7 µmoles/kg) or pyrantel (50 mg/kg or 84 µmoles/kg) or piperazine (4000 mg/kg or 46 mmoles/kg) delivered on day 18 P.I. effected an 87%, 98%, and 34% reduction in worm burdens respectively ([28],[29]. Another study showed that 2.9 µmoles/kg of ivermectin delivered on day 10 P.I. effected ∼70% reduction in H. bakeri worm burdens [30]. We could not find comparable studies with H. bakeri and benzimidazoles, although we did find that mebendazole delivered for 7 consecutive days, starting day 9 P.I. at 22 mg/kg/dose or 75 µmoles/kg/dose, achieved an 84% reduction in worm burdens [31]. Benzimidazoles (including albendazole) in general seem to be less active against H. bakeri [32]. Therefore, our single dose of ∼700 nmoles/kg (which is the highest dose we can currently pipette with SCLs) that achieved ∼70% reduction in worm burdens is 70X, 4–8X, 120X and 65,000X lower than the doses of levamisole, ivermectin, pyrantel, and piperazine used in the above studies.
This comparison suggests that the efficacy of Cry proteins relative to known anthelmintics is excellent. To directly compare our results to a known anthelmintic using the same treatment conditions, we performed curative experiments using the newest human anthelmintic and the only one taken to human clinical trials in the past thirty years, tribendimidine. We performed dose-dependent curative assays with tribendimidine against H. bakeri infections, finding an estimate dose of ∼1 mg/kg or 2.2 µmoles/kg tribendimidine to give a curative effect similar to ∼700 nmoles/kg Cry5B (Figure 3, Table 3). Based on this comparison, Cry5B is at least as good as tribendimidine at curing H. bakeri infections and in fact appears to be ∼2–3 fold superior.
These data indicate that Cry5B is an excellent anthelmintic when delivered at a single dose. However, Cry proteins are thought to be digested rapidly in the mammalian digestive tract, most notably by the acidic stomach [33]. If so, then it is possible that the dose of Cry protein reaching the parasites might have been very small.
To determine how well Cry5B would survive the mammalian stomach, we incubated Cry5B HD1-derived SCLs in simulated gastric fluids. We find that Cry5B is almost completely digested in this environment within four minutes (Figure 4). These data suggest that very little Cry5B is actually reaching the parasites.
Our study demonstrates that the Bt Cry protein Cry5B is an excellent anthelmintic in vivo against a natural and chronic intestinal roundworm infection in mice, namely H. bakeri. Cry5B is able to achieve significant reductions in parasite egg production (∼98%) and intestinal worm burdens (∼70%) following a single dose delivered per os at ∼700 nmoles/kg. This therapeutic effect, on a mole-by-mole basis, is on par with or superior to those of other anthelmintics commonly used in human therapy. Although this level of efficacy may seem surprising at first glance, upon deeper reflection it is not. Cry proteins, although they only attack the gut cells of invertebrates, are pore-forming toxins (PFTs; [34]). PFTs are the single most common virulence factors made by pathogenic bacteria and are also used by our immune system to combat pathogens [35],[36]. PFTs are potent weapons and the consequences of their attack on the integrity of the plasma membrane are great. In combination with previous data showing that Cry5B is also able to cure A. ceylanicum infections in hamsters [18], we have now demonstrated in vivo anthelmintic activity of Cry5B against two very different parasitic nematodes (one a blood feeder, the other not) in two different mammalian hosts. Taken together, along with the fact that Cry5B is active against Nippostrongylus brasiliensis larvae, against Haemonchus contortus larvae in vitro, against a phylogenetically wide range of free-living nematodes, and against the plant-parasitic nematode Meloidogyne incognita [16],[17],[37], our data indicate that Cry5B has very broad anti-nematode activity and that Cry5B has superb potential in human anthelmintic therapy.
As a natural product, it is interesting to compare the efficacy of Cry5B to other natural product anthelmintics. No recently investigated biological treatments against H. bakeri demonstrate comparable in vivo efficacy using single-dose regimens. Many of these natural compounds, such as the extract of Embelia schimperi, nitazoxanide, santonin, and Myrsine Africana, showed only small reductions in intestinal worm burden as a single dose, with efficacies of 30%, 21%, 18%, and 10%, respectively [29],[38],[39]. A single dose of 500 mg/kg of Albizia anthelmintica, not only revealed low efficacy, with a total worm burden reduction of only 3–23%, but also displayed significant toxicity [28]. Even the macrolactam N-methylfluvirucin, delivered at a daily dose of 50 mg/kg over 3 days, effected only a 42% reduction in total worm burden [40]. While other compounds were more efficient, they required extremely high doses and/or multiple-day dosing regimens. These included a daily treatment of ethanol extract of Canthium manni (Rubiaceae) at 5600 mg/kg, which showed a 75% decrease in fecal egg count and 84% reduction in worm burden with 7 days of treatment [31]. A 600 mg/kg treatment with extract of stem bark of Sacoglottis gabonensis was extremely effective, but exceedingly toxic, with mice showing signs such as depression, drowsiness, unsteady gait and paralysis of the hind limbs, dyspnoea, coma and death apparent within 1–2 min following intraperitoneal injection [41]. Perhaps the most promising of other natural treatments is papaya latex. A single-dose administration of papaya latex at 8 g/kg achieved an efficacy of 84.5%, with fecal egg count reductions of 93.3% [42]. Mice treated daily over 7 days with 133 nmoles of papaya latex showed a decrease in fecal egg count of 87–97% and a 92% reduction of worm burden [43]. In general, few of the natural compounds tested above proved to be practical treatments due to dosing and toxicity issues.
It is clear that Cry5B has great promise as an effective, safe, and much-needed addition to anthelmintic therapy. The vertebrate and human safety profiles of Cry proteins are outstanding—Cry proteins as insecticides are used around the world on a large-scale in organic farming, in aerial spray campaigns, and in vector (mosquito, black fly) control programs and have even been approved for expression in transgenic foods such as corn, potatoes, and rice [14],[44]. Although Cry5B has not been studied in this regard, it is a member of the same family of three-domain Cry proteins expressed in transgenic crops and used in all these spray programs and thus is predicted to have the same safety profile. Indeed, extensive research from our laboratory has confirmed that the receptor Cry5B needs to bind to in order to intoxicate nematodes is an invertebrate-specific glycan (carbohydrate) [45].
It is interesting to note that, although Cry5B has comparable if not superior activity against H. bakeri on a mole-by-mole basis with other anthelmintics, it is likely that only tiny amounts of the protein being delivered per os in our experiments are reaching the parasites. In four minutes, virtually all Cry5B is degraded in simulated gastric fluids. These experiments suggest that a simple enteric coating to protect Cry proteins against the stomach while releasing it in the small or large intestines might greatly increase the efficacy of Cry proteins. These data thus emphasize the importance of formulation in the next stage in the evolution of Cry protein anthelmintic development and suggest that such a formulation has the potential to reveal an anthelmintic with therapeutic properties comparable or superior to those currently in use.
|
10.1371/journal.pmed.1002741 | Lifetime risk and multimorbidity of non-communicable diseases and disease-free life expectancy in the general population: A population-based cohort study | Non-communicable diseases (NCDs) are leading causes of premature disability and death worldwide. However, the lifetime risk of developing any NCD is unknown, as are the effects of shared common risk factors on this risk.
Between July 6, 1989, and January 1, 2012, we followed participants from the prospective Rotterdam Study aged 45 years and older who were free from NCDs at baseline for incident stroke, heart disease, diabetes, chronic respiratory disease, cancer, and neurodegenerative disease. We quantified occurrence/co-occurrence and remaining lifetime risk of any NCD in a competing risk framework. We additionally studied the lifetime risk of any NCD, age at onset, and overall life expectancy for strata of 3 shared risk factors at baseline: smoking, hypertension, and overweight. During 75,354 person-years of follow-up from a total of 9,061 participants (mean age 63.9 years, 60.1% women), 814 participants were diagnosed with stroke, 1,571 with heart disease, 625 with diabetes, 1,004 with chronic respiratory disease, 1,538 with cancer, and 1,065 with neurodegenerative disease. NCDs tended to co-occur substantially, with 1,563 participants (33.7% of those who developed any NCD) diagnosed with multiple diseases during follow-up. The lifetime risk of any NCD from the age of 45 years onwards was 94.0% (95% CI 92.9%–95.1%) for men and 92.8% (95% CI 91.8%–93.8%) for women. These risks remained high (>90.0%) even for those without the 3 risk factors of smoking, hypertension, and overweight. Absence of smoking, hypertension, and overweight was associated with a 9.0-year delay (95% CI 6.3–11.6) in the age at onset of any NCD. Furthermore, the overall life expectancy for participants without these risk factors was 6.0 years (95% CI 5.2–6.8) longer than for those with all 3 risk factors. Participants aged 45 years and older without the 3 risk factors of smoking, hypertension, and overweight at baseline spent 21.6% of their remaining lifetime with 1 or more NCDs, compared to 31.8% of their remaining life for participants with all of these risk factors at baseline. This difference corresponds to a 2-year compression of morbidity of NCDs. Limitations of this study include potential residual confounding, unmeasured changes in risk factor profiles during follow-up, and potentially limited generalisability to different healthcare settings and populations not of European descent.
Our study suggests that in this western European community, 9 out of 10 individuals aged 45 years and older develop an NCD during their remaining lifetime. Among those individuals who develop an NCD, at least a third are subsequently diagnosed with multiple NCDs. Absence of 3 common shared risk factors is associated with compression of morbidity of NCDs. These findings underscore the importance of avoidance of these common shared risk factors to reduce the premature morbidity and mortality attributable to NCDs.
| The burden and preventive potential of disease is typically estimated for each non-communicable disease (NCD) separately, yet NCDs often co-occur, which hampers reliable quantification of their overall burden as well as the potential to prevent NCDs jointly in the general population.
Smoking, hypertension, and overweight are 3 key risk factors that are shared by all NCDs, but their effects on the lifetime risk of developing any NCD, age at onset, and overall life expectancy with and without NCDs are uncertain.
Between 1989 and 2012, we continuously followed 9,061 community-dwelling individuals in the Dutch population-based Rotterdam Study for occurrence of NCDs (stroke, heart disease, diabetes, chronic respiratory disease, cancer, and neurodegenerative disease).
Nine out of 10 community-dwelling individuals aged 45 years and older will develop any NCD during their lifetime, with a third of them developing multiple NCDs during follow-up.
Individuals without the 3 risk factors of smoking, hypertension, and overweight develop their first NCD on average 9 years later than those with all 3 risk factors.
Absence of smoking, hypertension, and overweight is associated with a longer overall life expectancy of about 6 years, and a 2-year compression in lifetime spent with NCDs.
In this western European community, we showed that the burden of NCDs in the general population is extensive and their multimorbidity is common.
Absence of smoking, hypertension, and overweight is associated with a longer overall life expectancy, and most of this increase is due to an extension of disease-free life expectancy. This means that individuals without these risk factors not only live longer than individuals with these risk factors, but also spend less of their lifetime after the onset of symptomatic disease (which is referred to as compression of morbidity).
These findings underscore the potential to substantially reduce premature NCD morbidity and mortality in the general population through prevention of smoking, hypertension, and overweight.
| Non-communicable diseases (NCDs), including stroke, heart disease, diabetes, chronic respiratory disease, cancer, and neurodegenerative disease, are the most frequent causes of prolonged disability and premature death worldwide [1–3]. Major changes in lifestyle and medicine over the past decades have led to significant reductions in premature mortality from NCDs such as heart disease and cancer, especially in high-income countries [4,5], shifting the burden of disease in these countries from premature mortality to prolonged disability. In low- and middle-income countries, however, rates of premature mortality caused by NCDs are rapidly increasing, leading to severe socio-economic burdens in these societies [6].
The risks for most NCDs are highly modifiable, with the potential to halve lifetime risks through prevention of risk factor occurrence [7–10]. Avoidance of risk factors is referred to as primordial prevention, which is pivotal in reducing the growing burden of NCDs [11]. Primordial preventive efforts aim to eradicate risk factors before they occur, such as by maintaining a healthy weight in order to prevent overweight. In contrast, primary prevention is the reduction of risk factors that already exist, such as by efforts to lose weight in obese individuals. Population-based data on the multimorbidity of NCDs are needed to help in understanding the impact of risk factors on the lifetime risk and age at onset of NCDs. Three common shared risk factors—namely smoking, hypertension, and overweight—underlie most years spent with disability and the subsequent deaths caused by NCDs [12–14]. NCDs often co-occur, but few longitudinal data are available to inform us about patterns of multimorbidity in the general population [15,16]. Most studies that have assessed the burden of NCDs in the population have investigated each NCD separately [3,17,18], but the burden of multimorbidity—the coexistence of 2 or more chronic diseases—is now considered a global healthcare priority [16]. In view of continuing increases in life expectancies and improvements in healthcare systems worldwide, an increasing number of individuals are growing into old age, many of whom will survive their first NCD and become at risk for multimorbid age-related NCDs [19]. Thus, longitudinal data across the lifespan are required to better understand the clustering of NCDs. This understanding could help in shaping policy, public education, identification of individuals at increased risk of multimorbidity, and developing joint interventions to prevent multiple NCDs simultaneously. Mitigating shared risk factor burden is not only a cost-effective preventive strategy to curb the rapidly growing burden of NCDs [6], it is also the single most feasible way to meet one of the key Sustainable Development Goals: to reduce premature deaths from NCDs globally by a third by 2030 [20]. Long-term data on the occurrence of NCDs are useful to inform societies about the burden and multimorbidity of NCDs, and the potential to prevent multiple NCDs in the general population.
Lifetime risk and life expectancy are metrics that are relatively easy to interpret and can readily be used by relevant stakeholders. Importantly, these metrics can also capture the burden of NCDs and the potential for prevention in comparable detail to more complex metrics such as incidence rates and hazard ratios—thereby enabling policymakers, clinicians, and other stakeholders to expand current and future efforts aimed at preventing risk factors and reducing the growing burden of NCDs.
We used long-term data from a community-based, prospective cohort study to quantify the occurrence and multimorbidity of NCDs. We also calculated the lifetime risk of any NCD, accounting for NCD multimorbidity and the competing risk of death from other causes. Finally, we studied the association of shared risk factor burden with lifetime risk, age at onset of NCD, and life expectancy with and without NCDs.
The Rotterdam Study has medical ethics committee approval per the Population Study Act: Rotterdam Study, executed by the Ministry of Health, Welfare and Sport of the Netherlands. Written informed consent was obtained from all participants. For the current study, the analysis plan was drafted in March 2018 (S1 Analysis Plan).
This is a substudy from the Rotterdam Study, a prospective, population-based cohort study designed to assess the occurrence and determinants of age-related diseases in the general population [21]. In 1989 and 1990, all inhabitants aged 55 years and older from a well-defined suburb in the city of Rotterdam, the Netherlands, were invited to participate. This initial cohort comprised 7,983 participants. In 2000, 3,011 participants who had become 55 years of age, or had moved into the study district since the start of the study and were aged 55 years and older, were added to the cohort. In 2006, a further extension of the cohort was initiated in which 3,932 participants were included, aged 45 years and older. In total, the Rotterdam Study comprises 14,926 participants aged 45 years or over. The overall response rate across the 3 recruitment waves was 72% (14,926 out of 20,744 invitees).
To estimate lifetime risk of NCDs, we excluded 4,461 participants with a history of 1 or more NCDs at baseline (stroke n = 244, heart disease n = 915, diabetes n = 957, chronic respiratory disease n = 783, cancer n = 343, neurodegenerative disease n = 485, or a combination of these diseases n = 734). We further excluded participants who were incompletely screened at baseline for at least 1 of these diseases (n = 1,404), retaining 9,061 participants available for analyses.
Smoking status was assessed during an in-depth interview, conducted by trained research assistants who visited the participants at home 2 weeks before their scheduled visit to attend the research centre. Smoking behaviour was categorised into never, former and current. At the research centre, blood pressure was measured at the right upper arm after at least 5 minutes’ rest in a seated position. Hypertension was defined as a resting sitting blood pressure exceeding 140/90 mm Hg (mean of 2 measurements) and/or the use of blood-pressure-lowering medication. Drugs categorised by the World Health Organization Anatomical Therapeutic Chemical (ATC) classification as antihypertensives (c02), diuretics (c03), beta blockers (c07), calcium channel blockers (c08), and RAAS-modifying agents (c09) were considered as blood-pressure-lowering medication. A body mass index ≥ 25 kg/m2 was considered as overweight. Marital status (living with or without partner) and educational attainment were also assessed during the same home interview to assess smoking status. Educational attainment was categorised as primary education (‘primary’), lower/intermediate general education or lower vocational education (‘lower’), intermediate vocational education or higher general education (‘further’), or higher vocational education or university (‘higher’).
Baseline and follow-up ascertainment methods for stroke, heart disease (fatal or non-fatal coronary heart disease or heart failure), diabetes, chronic respiratory disease (chronic obstructive pulmonary disease or asthma), cancer (solid or haematological cancer), and neurodegenerative disease (dementia or parkinsonism) have previously been described in detail [9,22–26]. Disease-specific definitions, procedures, and data collection are summarised in S1 Methods. In brief, data on clinical outcomes are collected continuously through an automated follow-up system involving digital linkage of the study database to medical records maintained by general practitioners working in the research area. Trained research assistants affiliated with the study regularly check the medical records of each participant by hand for diagnoses of interest. Consultation notes, outpatient clinic reports, hospital discharge letters, electrocardiograms, pharmacy dispensing records, and imaging results are collected from general practitioner records and hospital records. Research physicians affiliated with the study independently review all data associated with events. Medical specialists also affiliated with the study review the potential cases, and adjudicate the final diagnosis in accordance with standardised diagnostic criteria. Further coverage of disease monitoring and subtyping is obtained through linkage to nationwide medical registries, the national cancer registry, and the Dutch pathology database. Information on vital status is obtained from the central registry of the municipality of the city of Rotterdam.
When individuals are followed for long time periods (e.g., from mid-life to occurrence of disease or death), preclusion of disease-specific outcomes of interest by death from other causes or by competing events may lead to overestimation of absolute risks in standard Kaplan–Meier analyses. To overcome the issue of such competing risks, we analysed the data taking into account the occurrence of competing events to compute remaining lifetime risks in left-truncated data, with age as the time scale [9,27,28]. Lifetime risk estimates reflect the competing-risk-adjusted cumulative incidences from that particular age onwards until the participant’s age at last follow-up. In this study, the maximum age was 106 years for men and 107 years for women.
First, we studied the occurrence and patterns of multimorbidity of NCDs during follow-up. We quantified the number of events for each NCD separately and visualised all observed combinations of NCD multimorbidity during follow-up with an intersection diagram [29].
Second, we calculated the combined cumulative incidences of these diseases from the age of 45 years to the participant’s age at last follow-up. The combined cumulative incidence equals the remaining lifetime risk of developing any NCD from the age of 45 years onwards. For these analyses, follow-up started at study entry (with the age of 45 years as minimum) and ended at the first date of diagnosis of any of the NCDs. This meant that we considered only the first occurring event of the 6 potential outcomes in this analysis in order to calculate the overall risk of developing any NCD, because when studying different risks of first manifestations of NCDs, the occurrence of 1 manifestation precludes consideration of any subsequent NCD event. For instance, participants who first experienced heart disease during follow-up were no longer considered as being at risk for stroke or any other NCD. This combined cumulative incidence of any NCD can be divided for each of the 6 diseases separately, which then represents the cumulative incidence of that specific disease occurring as the first manifestation of the 6 potential outcomes.
As a complementary analysis, we also calculated the disease-specific cumulative incidence, in which we considered only the disease of interest as the outcome, while disregarding the occurrence of the 5 other diseases. Thus, participants remained at risk of the 6 diseases irrespective of the occurrence of a first event. This meant that, for example, participants with heart disease during follow-up remained at risk for stroke in the analysis for stroke.
Third, we repeated these analyses stratified on the 3 risk factors at baseline, i.e., current smoking, hypertension, and overweight, to study whether these risk factors were related to overall lifetime risk. We also evaluated the effects of these risk factors on the age at onset of the first NCD and determined whether the type of disease as first manifestation differed across risk factor profiles.
Finally, we studied the effects of risk factors at baseline on life expectancy of participants with and without NCDs across absence and presence of risk factors using multistate lifetables [30]. This demographic tool combines all the life experiences of participants in 3 different health states: free of NCD, living with an NCD, and death. Transitions between these states could be from free of NCD to NCD (incident NCD), NCD to death (mortality among participants with NCD), and free of NCD to death (non-NCD mortality among participants without NCD). We considered only the first event into a state, and backflows were not allowed. For instance, for participants who developed multiple NCDs during follow-up, only their first NCD event (‘incident NCD’) is considered in these analyses. We adjusted for age, sex, birth year, marital status, and educational level. Detailed methodology for these calculations has been previously described, and is summarised in S1 Methods [30].
For the analyses on multimorbidity, study follow-up ended at date of death, loss to follow-up, or January 1, 2012, whichever came first. For the analyses on the lifetime risk of any NCD and on overall life expectancy, follow-up ended at the date of any incident NCD diagnosis (or, for the complementary disease-specific analysis, the date of incident NCD of interest), death, loss to follow-up, or January 1, 2012, whichever came first. Participants were considered lost to follow-up if they moved out of the Netherlands, if their medical records could not be accessed, or if participants withdrew their informed consent during follow-up. We used imputation procedures to impute missing data (<2.0%). This study is reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (S1 Checklist). All analyses were performed at the significance level of 0.05 (2-tailed) using SPSS Statistics version 24.0.0.1 (IBM, Armonk, NY) and R version 3.4.3.
Study population characteristics are presented in Table 1. Median age at baseline was 61.7 years (range 45–107 years), and 60.1% of the population were women. Study population characteristics stratified by study wave are presented in Table A in S1 Results. Compared to participants included in the first study wave, participants in the second and third wave were generally younger at the start of follow-up. They attained a higher educational level, and were more likely to smoke or be overweight at baseline. Participants in the third wave were less likely to have hypertension at baseline than participants in the 2 other recruitment waves. During 75,354 person-years of follow-up (99.3% of potential person-years observed), 6,617 events occurred among 9,061 participants: 814 participants were diagnosed with stroke, 1,571 with heart disease, 625 with diabetes, 1,004 with chronic respiratory disease, 1,538 with cancer, and 1,065 with neurodegenerative disease. In total, 2,941 participants died during follow-up, of whom 421 died free of these diseases.
A third (33.7%) of all participants who developed an NCD (n = 4,633) were diagnosed with multiple NCDs during follow-up (Fig 1). The 5 most frequent clusters of disease were heart disease and neurodegenerative disease (n = 170; 3.7%), cancer and heart disease (n = 138; 3.0%), neurodegenerative disease and stroke (n = 137; 3.0%), heart disease and chronic respiratory disease (n = 123; 2.7%), and cancer and chronic respiratory disease (n = 122; 2.6%). Most participants who developed an NCD during follow-up were diagnosed with a single (n = 3,070; 66.3%) or 2 (n = 1,201; 25.9%) NCDs, whereas 1 individual was diagnosed with all 6 NCDs of interest. Among participants who developed a single NCD, cancer (n = 888; 28.9%) and heart disease (n = 724; 23.6%) were the most common, followed by neurodegenerative disease (n = 455; 14.8%), chronic respiratory disease (n = 452; 14.7%), diabetes (n = 294; 9.6%), and stroke (n = 254; 8.3%).
To calculate the lifetime risk of developing any NCD, we only considered the first NCD event during follow-up. In all, 1,173 participants were diagnosed with cancer as their first NCD, 1,035 with heart disease, 833 with chronic respiratory disease, 711 with neurodegenerative disease, 468 with stroke, and 413 with diabetes. In Fig 2, the combined cumulative incidence for developing any of these diseases from the age of 45 years onwards is presented for women and men separately. The risk of developing an NCD increased steeply with age, ranging from 33.3% for men and 29.8% for women between age 45 and age 65 years, up to 87.3% and 77.6%, respectively, until the age of 85 years. While men were at higher risk of developing an NCD at a younger age, the risk for women rapidly caught up with advancing age. Consequently, the overall lifetime risk of developing any of these diseases was only slightly higher for men compared to women, with a 94.0% (95% confidence interval (CI) 92.9%–95.1%) lifetime risk for a 45-year-old man and a 92.8% (95% CI 91.8%–93.8%) risk for a 45-year-old woman (p-value for sex difference < 0.001). This lifetime risk of developing any NCD remained stable across the included study waves (Table B in S1 Results). Dividing the combined cumulative incidence across the various NCDs showed that the difference in overall lifetime risk for these diseases between men and women was driven by a marked difference in risk for the type of first NCD. While cancer was the most common first NCD for both men (25.9%) and women (24.4%, p-value for sex difference = 0.22), the lifetime risk of heart disease as first NCD was significantly higher in men (22.5%) than women (17.0%, p-value for sex difference < 0.001). Conversely, the risk of neurodegenerative disease as first NCD was higher in women (14.1%) compared to men (6.3%, p-value for sex difference < 0.001). The complementary analysis, in which participants remained at risk for the specific NCD of interest irrespective of the occurrence of other NCDs, is presented in Table 2.
When stratifying by risk factors, the overall lifetime risk of developing an NCD only slightly increased from 90.3% (95% CI 87.8%–92.9%) for those without the risk factors to 96.8% (95% CI 95.3%–98.2%) for participants with all risk factors (Fig 3). However, a large difference in the age at onset was observed (Fig 4): compared to participants who smoked, were hypertensive, and were overweight at baseline, participants without these risk factors were on average 9.0 years older (95% CI 6.3–12.6) when they were first diagnosed with an NCD. Similar trends were seen across the entire age range, for example, at age 55 years, cumulative incidence was 14.3% for those with all 3 risk factors, whereas this cumulative incidence was not reached until age 62.5 years (i.e., 7.5 years later) for those without the risk factors. Similarly, at age 75 years, cumulative incidence was 73.2% for those with all 3 risk factors, whereas this cumulative incidence was not reached until age 86 years (i.e., 11 years later) for those without the risk factors (Fig 4). These effects were accompanied by marked differences in the type of first NCD. For instance, participants without the 3 risk factors of smoking, hypertension, and overweight had a 16.8% lifetime risk of developing heart disease as a first manifestation, whereas this risk was 25.8% for those with all 3 risk factors. Participants without the risk factors remained at risk for neurodegenerative disease and cancer, while those with risk factors were particularly at risk for the development of heart disease, diabetes, and chronic respiratory disease. Compared to the joint effects of the 3 risk factors, differences in cumulative incidence and lifetime risk of developing any NCD became smaller when considering the effects of individual risk factors (Table C and Fig D in S1 Results). Similar patterns of risk factor effects were observed in disease-specific lifetime risk analyses (Table D and Fig E in S1 Results).
Participants aged 45 years without the 3 shared risk factors smoking, hypertension, and overweight lived on average 6.0 years (95% CI 5.2–6.8) longer than those with all 3 of these risk factors (Fig 5). Moreover, while participants without these risk factors on average lived longer than those with these 3 risk factors, they also spent a smaller proportion of their life from the age of 45 years with at least 1 NCD. For instance, participants with the 3 risk factors of smoking, hypertension, and overweight spent on average almost a third (31.8%) of their remaining life expectancy from the age of 45 years with at least 1 NCD compared to approximately a fifth (21.6%) for those without these common shared risk factors. Similar trends in risk factor effects were observed when considering participants who had at least 1 or 2 of these risk factors (Figs E and F in S1 Results). Associations of individual risk factors with life expectancy were smaller compared to the joint effects of smoking, hypertension, and overweight (Table E and Fig G in S1 Results). When studying the individual association of smoking, hypertension, or overweight with life expectancy, participants who did not smoke had the longest life expectancy without an NCD (24.5 years [95% CI 24.3–24.7] compared to 20.0 years [95% CI 19.7–20.3] for participants who smoked).
In this population-based cohort study, a substudy of the ongoing prospective Rotterdam Study, we assessed the lifetime risks of developing co-occurring NCDs, and quantified their multimorbidity. We show that 9 out of 10 individuals develop an NCD from the age of 45 years onwards. Among those individuals, at least a third are subsequently diagnosed with multiple NCDs. Importantly, absence of 3 common shared NCD risk factors—namely smoking, hypertension, and overweight—is associated with a 9-year delay in the first diagnosis of any NCD compared to those with these 3 risk factors. Furthermore, absence of these risk factors is associated with an extended life expectancy of 6 years. These findings highlight the potential to lower the proportion of a lifetime spent with disability and the number of premature deaths caused by NCDs through prevention of shared risk factors among community-dwelling individuals.
Several cohort studies have assessed the disease-specific lifetime risks of various NCDs, including stroke [31,32], heart disease [10,28], diabetes [9], chronic respiratory disease [8,33], cancer [7] and neurodegenerative disease [32]. However, data were lacking on the combination of all these major NCDs within a single population and had not been used to calculate the overall lifetime risk of NCD among community-dwelling individuals. The extent of NCD multimorbidity and the effects of common shared risk factors on the onset of NCDs were largely unknown. Here, we show that as many as 9 out of 10 individuals aged 45 years and older will develop 1 or more of these NCDs during their remaining lifespan. In fact, according to our analyses that did not take multimorbidity into account, each of the 6 NCDs assessed in this study posed a high risk for community-dwelling individuals. For instance, the disease-specific remaining lifetime risk ranged from 21% for stroke in men up to 48% for cancer in men from the age of 45 years onwards. These numbers highlight the frequent occurrence of these NCDs in this western European population and show that—in view of the chronically progressive nature of these diseases—they are likely to impose a substantial burden on societies similar to the one in our study.
Modelling studies have shown the contribution of risk factor management to reducing NCD-related years spent with disability and (premature) mortality [17,18]. These studies were either based on prevalence data or relied on several assumptions, including the use of prediction models to estimate disease occurrence at the country level, and the derivation of relative risks from systematic reviews to quantify the effects of exposures on disease risk. Consequently, such analyses are hampered by the competing risks of death and other NCDs, the substantial multimorbidity of NCDs, and interactions between underlying risk factors. Indeed, several risk factors are shared between NCDs: most of the risk factors are, for instance, not only implicated in heart disease, but are also strongly related to other NCDs such as stroke, as well as an increased risk of mortality. If the competing risks of death and precluding NCD events are not appropriately accounted for, the risk of developing any NCD will be overestimated. For example, in our study, considering the lifetime risk of each NCD separately, without taking into account the competing risk of multimorbidity, results in a summed lifetime risk of the NCDs together exceeding 100%. Additionally, risk factors can often aggravate each other’s detrimental effects. Therefore, the preventive effect of removing multiple risk factors is often larger than the sum of the effects of removing the individual risk factors. Indeed, in our study the absence of smoking, hypertension, and overweight was far more beneficial in lowering disease risk and extending life expectancy than could be expected based on the individual associations of smoking, hypertension, or overweight alone.
In this study, we were able to overcome the aforementioned challenges by using real-world data, which allowed us to study the association of shared risk factors with the lifetime risk of developing any NCD, and with overall life expectancy with and without NCDs. Absence of the 3 shared risk factors of smoking, hypertension, and overweight was associated with a substantial delay in the onset of any NCD, and was also associated with a longer life expectancy free from NCDs. Results from this study in a western European population also suggest that many individuals residing in comparable populations with similarly organised healthcare systems will be affected by an NCD at some point in their life, although at what age and which disease type will manifest first are strongly influenced by an individual’s underlying risk factor profile. We found substantial differences in first manifestation of NCD between men and women. Although men and women have a roughly similar lifetime risk of developing any NCD, men are more likely to develop NCDs at a younger age, and to develop heart disease, chronic respiratory disease, or diabetes as their first event. Women are more likely to develop their first NCD at an older age, and subsequently have a higher risk of stroke and neurodegenerative disease, as compared with men. These results extend prior findings on sex differences in lifetime risk and first manifestation of heart disease, stroke, and neurodegenerative disease [28,32].
Some limitations of this study must be acknowledged. First, findings from this community-based study of predominantly white individuals from western Europe have limited generalisability to capture the contemporary burden of NCDs in low- and middle-income countries. Nevertheless, given the expected increase in life expectancy in coming years in low- and middle-income countries, these results may be informative to help the societies of these countries prepare future resource allocation. Second, although the overall response rate in this study was high (72%), non-responders may have had higher than average risk factor burden and associated risk of NCDs, which may have led to some underestimation of results. Third, in the Dutch healthcare system, the entire population is entitled to primary care that is covered by their (obligatory) health insurance. In this primary care setting, a general practitioner provides primary prevention for NCDs, which may have affected the risk, age at onset, and type of NCD. Generalising the results of our study to healthcare systems that are organised differently, or to those that have limited availability of primary preventive healthcare, should therefore be done with caution. Fourth, risk factors were ascertained at baseline, which does not capture the possibility of individuals developing additional risk factors during follow-up or, conversely, transitioning from an adverse to a more optimal risk profile. Finally, individuals who do not smoke or have hypertension and are not overweight might also have other factors associated with a healthy lifestyle, such as a healthy diet and physical activity. Indeed, residual confounding may have influenced the results, which limits causal interpretation of findings from this observational study. Nonetheless, a major strength of this study is the long-term follow-up of multiple NCDs systematically assessed in a single contemporary population study. This allowed the quantification of NCD multimorbidity, the calculation of the lifetime risk of developing any NCD, and the calculation of life expectancy with and without NCD.
Over the past decades, smoking control and treatment of hypertension, along with improvements in treatment options, have led to a marked decline in premature deaths from heart disease despite clear trends in increased prevalence of overweight [34]. Our results indicated that the absence of smoking, hypertension, and overweight was associated with a 9-year delay in the age at onset of NCD, compared to individuals with these risk factors. This delay in the age at onset of NCD remained roughly similar across the entire age span, and subsequently led to marked differences in first manifestation of NCD. Individuals with the most adverse risk factor profiles had an approximately doubled risk of developing chronic respiratory disease, diabetes, or heart disease as first NCD. Conversely, healthy individuals with optimal risk factor levels were more likely to develop an NCD at older age, and as a consequence had a 5-fold increased risk of developing neurodegenerative disease as their first manifestation. Life expectancy analyses showed that individuals without the common shared risk factors at baseline lived longer without an NCD, yet lived for a shorter time after diagnosis of an NCD. Individuals without these 3 risk factors had a 6-year longer overall life expectancy than people with these risk factors, and they also spent much more of their remaining life expectancy from the age of 45 years free from NCD compared to individuals with these 3 risk factors. As such, individuals without the risk factors gained relatively more healthy life years than years of overall life expectancy, and as a consequence, the time between the onset of chronic illness or disability and the moment of death was compressed. This is referred to as compression of morbidity [35]. In summary, the findings of this study emphasize that efforts aimed at optimal prevention of common shared risk factor occurrence may benefit healthy aging at a population level.
In this cohort study of a western European community, 9 out of 10 individuals aged 45 years and older develop an NCD during their remaining lifetime. Among those individuals, at least a third are subsequently diagnosed with multiple NCDs. Absence of the 3 shared risk factors of smoking, hypertension, and overweight is associated with a 9-year delay in the age at onset of any NCD, and a significantly prolonged life expectancy. These findings highlight the potential to reduce premature disability and death caused by NCDs through primordial prevention of smoking, hypertension, and overweight.
|
10.1371/journal.pbio.1001258 | Genome-Wide Analysis of the World's Sheep Breeds Reveals High Levels of Historic Mixture and Strong Recent Selection | Through their domestication and subsequent selection, sheep have been adapted to thrive in a diverse range of environments. To characterise the genetic consequence of both domestication and selection, we genotyped 49,034 SNP in 2,819 animals from a diverse collection of 74 sheep breeds. We find the majority of sheep populations contain high SNP diversity and have retained an effective population size much higher than most cattle or dog breeds, suggesting domestication occurred from a broad genetic base. Extensive haplotype sharing and generally low divergence time between breeds reveal frequent genetic exchange has occurred during the development of modern breeds. A scan of the genome for selection signals revealed 31 regions containing genes for coat pigmentation, skeletal morphology, body size, growth, and reproduction. We demonstrate the strongest selection signal has occurred in response to breeding for the absence of horns. The high density map of genetic variability provides an in-depth view of the genetic history for this important livestock species.
| During the process of domestication, mankind recruited animals from the wild into a captive environment, changing their morphology, behaviour, and genetics. In the case of sheep, domestication and subsequent selection by their animal handlers over thousands of years has produced a spectrum of breeds specialised for the production of wool, milk, and meat. We sought to use this population history to search for the genes that directly underpin phenotypic variation. We collected DNA from 2,819 sheep, belonging to 74 breeds sampled from around the world, and assessed the genotype of each animal at nearly 50,000 locations across the genome. Our results show that sheep breeds have maintained high levels of genetic diversity, in contrast to other domestic animals such as dogs. We also show that particular regions of the genome contain strong evidence for accelerated change in response to artificial selection. The most prominent example was identified in response to breeding for the absence of horns, a trait now common across many modern breeds. Furthermore, we demonstrate that other genomic regions under selection in sheep contain genes controlling pigmentation, reproduction, and body size.
| Man's earliest agricultural systems were based on the captive management of sheep and goats. The transition from hunting to animal husbandry involved human control over the reproduction, diet, and protection of animals. The process of domestication was initiated approximately 11,000 years ago in the Fertile Crescent [1]. The impact was a profound redirection of human society, as domesticated livestock and plants increased the stability of human subsistence and fuelled population growth and expansion. Domestication also reshaped the morphology, behaviour, and genetics of the animals involved, with the first consequences likely to have included changes to coat pigmentation and horn morphology. Sheep were first reared for access to meat before human mediated specialisation for wool and milk commenced ca 4,000–5,000 years ago [2]. Phenotypic radiation under selection is ongoing, resulting in a spectrum of modern breeds adapted to a diverse range of environments and exhibiting the specialised production of meat, milk, and fine wool. The last few hundred years has seen the pace of genetic gain increase dramatically through the division of animals into breeds, the implementation of quantitative genetics methodology, and the use of artificial insemination to prioritise genetically superior rams.
Patterns of genetic variation have long proven insightful for the study of domestication, breed formation, population structure, and the consequences of selection. Variation within the mitochondrial genome has documented the global dispersal of two major haplogroups in modern sheep [3],[4]. Analysis of endogenous retroviruses suggests the development of breeds has occurred in multiple waves, where primitive breeds have been displaced by populations which display improved production traits [2]. Investigations into the genetic relationship between populations have primarily relied on a modest collections of autosomal microsatellites [5]–[7], Y chromosomal markers [8], or SNP [9]. To date, the majority of populations tested have been European-derived breeds. This prompted assembly of the global sheep diversity panel, which contains animals from 74 diverse breeds sampled from Asia, Africa, South-West Asia (the Middle East), the Caribbean, North and South America, Europe, and Australasia. Our goal in assembling this animal resource was 2-fold. Firstly, we sought to examine levels and gradients of genetic diversity linking global sheep populations to better understand the genetic composition and history of sheep. We therefore genotyped all of the animals in the global diversity panel using the ovine SNP50 Beadchip, an array consisting of approximately 50,000 evenly spaced SNP. We present the relationship between breeds in terms of divergence time, estimated from the extent of haplotype sharing. Secondly, we sought to characterise the genetic legacy that selection and adaptation have imparted on the sheep genome. By performing a genome-wide scan for the signatures of selection, 31 genomic regions were identified that contain genes for coat pigmentation, skeletal morphology, body size, growth, and reproduction. By combining the collection of a global sample of ovine breeds with the ability to interrogate 50,000 genetic loci, the results provide unprecedented insight into the phylogeographic structure of sheep populations and the results of centuries of breeding practices.
Analysis of genetic variation was performed for 2,819 animals in the global sheep diversity panel. Breeds were sampled from each continent across the species range (Figure 1), including six breeds from both Africa and America, seven from South-West Asia (the Middle East), eight from Asia, and the rest from northern, north-western, central, and southern and south-western Europe (Table S1 lists the breed and their geographic origin). All animals were genotyped using the ovine SNP50 Beadchip, an array consisting of SNP derived from three separate sequencing experiments (Roche 454, Illumina GA and Sanger sequencing; Table S2). A series of quality control filters were applied to identify 49,034 SNP used in subsequent analysis (Table S3). Levels of SNP polymorphism were generally high, with greater than 90% of loci displaying polymorphism within the majority of breeds (Table S4). The distribution of minor allele frequency (MAF) differed between population groups chosen to reflect the geographic origin of breed development. African and Asian breeds had an excess of low MAF SNP (<0.1) compared to European-derived populations. This partly reflects ascertainment bias in SNP discovery, as the same analysis conducted using SNP discovered without use of African or Asian sheep (454 SNP; Figure S1, Table S2) shows a more pronounced excess compared with SNP discovered using a broad genetic base (Illumina GA SNP). To examine diversity on a global scale, we calculated observed heterozygosity (He) within breeds and between regions (Table S4). Allele frequency-dependent diversity estimates such as He are sensitive to ascertainment bias, prompting the removal of SNP in high LD, which acts to counter the effect of the bias and generate meaningful comparisons between populations [10]. Applied here, breed rankings based on He were generally stable following LD-based pruning and when calculated using SNP sets ascertained using different methods (Figure S2). Following LD-based correction, animals from Southern and Mediterranean Europe displayed the highest heterozygosity (Figure S2). This likely reflects the first migrations of Neolithic communities and their animals, following the Mediterranean as a sea route into Europe [11]–[13]. Relative levels of genetic diversity are expected to decrease with increasing distance from the domestication centre. For sheep, breed heterozygosity revealed only a weak association with increasing physical distance (Figure 1B, r = −0.40). This appears much less pronounced in sheep compared with human migration out of Africa [14]. One likely explanation is the widespread use of Merino sires across Europe that commenced after the Middle Ages. The result is extensive haplotype sharing between Merinos and other breeds (Figure 1C). Generally high SNP diversity in sheep was accompanied by many breeds displaying high current effective population size (Ne, Table S4). Compared with domestic cattle where the majority of breeds have a current Ne of 150 or less [15], estimates here revealed 25 breeds have Ne exceeding 500 and only two sheep populations showed evidence of a comparatively narrow genetic base (Ne<150).
Global patterns of genetic structure were inferred by principal components analysis (PCA, Figure 2). The analysis ignores breed membership but revealed clear structure as animals from the same breed clustered together. As demonstrated in human and other livestock species such as cattle [15]–[17], the combination of PC1, PC2, and PC3 separated individuals according to their geographic origin. The largest PC (2.98% of total variation) positioned European sheep apart from African, Asian, and South-West Asian animals. The second PC (1.44%) separated European-derived animals from those developed in Africa and Asia animals. PC3 (1.19%) identified admixed populations such as the African Dorper and breeds developed in South America and the Caribbean were positioned away from other clusters. It also resolved two primitive and geographically isolated Scottish breeds (Soay and Boreray) as outliers from all other animals [2],[6],[9]. PC4 (1.09%) separated British Dorset types (DSH, APD, and ASU) from other European derived breeds and PC7 identified the Valais breeds as genetically distinct. Additional PCs reveal the divergence of single or a few related breeds (refer to the heatmap in Figure 2). To explore in detail the relatedness between European animals, analysis was performed separately for Mediterranean and northern-derived breeds (Figure S3). Even closely related populations such as Irish and Australia Suffolk had non-overlapping clusters, confirming the dataset provides an extremely high resolution view of population divergence. This power of resolution results from the large number of markers used, as a pilot study using only 1,315 SNP failed to distinguish closely related European-derived breeds [9]. Model-based clustering partitioned the genome of each animal into a predefined number of components (K) [18]. For unsupervised clustering assuming two ancestral populations (K = 2), a clear division was observed between Northern European and Asian breeds (Figure S4), corresponding to PC1. Clusters were reproducible up to K = 9 and grouped individuals according to their geographic origin in the same way as for PCA (Figure 2). The 20 largest PCs accounted for only 16% of the total variation (Figure S5), consistent with reports suggesting sheep have a weak population structure [3],[9]. To evaluate if this was accompanied by high levels of haplotype sharing between breeds, the extent of LD was characterised by the signed r statistic between SNP pairs at different lengths (e.g., [19]). For SNP pairs separated by 10 kb or less, a high degree of conservation of LD phase was observed between all breeds (Figure S7). Given that LD at short haplotype lengths reflects population history many generations ago [20],[21], this also supports a common ancestral origin of all domestic breeds of sheep. The result is in contrast to cattle, where two distinct groups emerge from a similar analysis, even at haplotype lengths of 0–10 kb, reflecting the Bos taurus taurus and Bos taurus indicus sub-species and their separate domestication events [15]. To determine if our LD-based estimates of haplotype sharing and effective population size were influenced by strong admixture, simulation was performed using a mutation drift model [22] and populations designed to mimic HapMap sheep breeds. This revealed admixture did affect inferred Ne, however the impact was minimal outside of the period in which the admixture took place (Figure S6).
The relationship between breeds was examined using two distance metrics. Firstly, the divergence time separating all breed pairs was estimated from LD and haplotype sharing using the methods of (Figures S7, S8, S9, S10) [19]. Divergence time (in generations) revealed a strong correspondence with known population history for recently separated breed pairs. For example, breeds established within the last 100 years (e.g., Poll Dorset and Poll Merino) had the shortest divergence time (<80 generations). Breeds with longer history, such as American Rambouillet, had divergence from Merino estimated at 160–240, which matches with their export from Spain to America starting in the late 1800s. The deepest divergence was estimated at only 800 generations, which appears to be an underestimate likely reflecting the influence of admixture. Divergence times between all breeds were explored as a NeighborNet graph that had branches of approximately equal length, suggesting the approach is robust to differences in genetic drift and effective population size between populations (Figure 3). NeighborNet graphs allow for reticulation as a consequence of relatedness and mixed breed origin, and the topology of the graph reproduced both the geographic groups and relationships obtained by PCA. Reticulations were observed toward the extremity of the graph for breed pairs that clustered together in PCA (e.g., Dorset Horn and Australia Poll Dorset). Conversely breeds identified as outliers by PCA such as the Soay had branches that originated from the centre of the graph. The second distance metric, Reynold's distance, relies on allele frequency differences, and branch lengths were highly variable (Figure 4). To test for the impact of ascertainment bias in SNP selection, we compared graphs generated using different SNP sets. In each case, the graphs had highly similar topology, which argues against a major influence of bias during SNP discovery (Figure S11). Short branches were observed for Spanish, Italian, and Iranian breeds with a high heterozygosity, while long branches were found for isolated populations containing small effective population size. Omitting the crossbred populations resulted in a remarkable demarcation of the geographic clusters. The topology of the graph suggests a major migration route along an axis that runs from South-West Asia to the Mediterranean region and via central Europe to Britain and the Nordic regions. Testing of additional breeds will be required to assess if migration was strongly influenced by a Danubian colonisation route.
Animal husbandry and directed mating have been used to successfully adapt sheep to a diverse range of environments and to the specialised production. Selection is predicted to alter allele frequencies within the target population for both functional mutation(s) and their neighbouring SNP. Global FST was calculated, which measures differentiation within each breed versus all other breeds and detects both positive and balancing selection. The genome-wide distribution of global FST for 49,034 SNP revealed the highest selection signal was detected on Chromosome 10 (Figure 5). The highest ranked SNP (OAR10_29511510; FST = 0.682) was located at Mb position 29.54 near the Relaxin/insulin-like family peptide receptor 2 (RXFP2), which was recently linked with the absence of horns (poll) in sheep [23] and displayed strong evidence for selection in cattle [24]. This prompted calculation of pairwise FST between breeds defined as either being horned or polled. This recapitulated a single strong and striking selection signal at RXFP2 (Figure 6). Importantly, the FST signal was absent when polled breeds or horned were compared with each other.
A total of 31 genomic regions contained the top 0.1% of markers ranked using global FST (47 SNP, Table 1). This implicated 17.85 Mb of sequence containing 181 genes as being under selection. The exact target of selection was difficult to identify as six genes, on average, were present within each genomic region. Gene ontology (GO) terms associated with the 181 genes were evaluated for evidence of functional enrichment against a background set of 11,098 genes physically tagged by the ovine SNP50 Beadchip (Table S5). This revealed enrichment for GO terms associated with regulation of bone remodelling (p = 5.5×10−5) and bone resorption (p = 4.0×10−5). Given it is unlikely all 181 genes have undergone selection but each contributed to the GO analysis, caution is required during interpretation. Nonetheless, the content of the differentiated regions strongly suggests enrichment for genes under selection given their roles in pigmentation, body size, reproduction, animal production, and domestication. Selection for specialised coat pigmentation represents breed-defining characteristics across domestic animals including sheep. Selection signals were detected spanning KIT, ASIP, and MITF (regions 8, 19, and 26 on OAR 6, 13, and 19, respectively, Table 1). KIT and MITF interact during melanocyte development and account for pigmentation phenotypes in pigs and cattle [25],[26], while duplication of ASIP in sheep controls a series of alleles for black and white coat colour [27]. Global FST peaks spanned NPR2, HMGA2, and BMP2, which are each involved in skeletal morphology and body size (regions 1, 5, and 18 on OAR 1, 5, and 18, respectively, Table 1). HMGA2 is of particular interest as it was recently shown to be under selection in dogs with divergent stature [28],[29]. Positive selection was detected surrounding two genes known to regulate growth and reproduction (PRLR on OAR6l; TSHR on OAR 7; Table 1). Prolactin receptor (PRLP) is a key regulator of mammalian reproduction that is critical for the onset of lactation and is associated with milk traits in dairy cattle [30]. In addition, a very strong selection sweep surrounds the thyroid stimulating hormone receptor (TSHR) in chicken, which given its pivotal role in metabolic regulation and the control of reproduction, was postulated to be a domestication gene [31]. Finally, an FST peak on Chromosome 6 spanned the FGF5 gene, recently shown to contain mutations in dog responsible for variation in hair type [32]. Each putative gene target for selection is recorded in Table 1, however this does not include examples where the 31 regions intersect with previous findings arising from QTL that have not been resolved to identify individual genes. One example is Mb position 6.8–7.2 on OAR 25, which contains QTL for wool production and quality in a number of breeds [33],[34]. The location of all 31 regions were compared to selection signals identified within the cattle genome [15],[24],[35]–[40]. Eleven of the 31 genomic regions identified here appear to be under selection in cattle, suggesting genes such as KIT, FGF5, MITF, and RXFP2 are targets for selection across multiple mammalian lineages (Table S6).
To search for selection observed across multiple breeds, the number of populations that displayed divergence was plotted across the genome. This revealed peaks where selection was shared across breeds, and troughs where signals were absent or unique to only a small number of breeds. Four regions were detected with positive selection shared across 30 or more breeds, while five different regions were observed with shared balancing selection. The strongest balancing selection signal was observed for the MHC region on sheep Chromosome 20 (Figure 7), a result previously observed in other species including cattle [37]. Conversely, some selection signals were breed specific. The global sheep diversity panel contained three geographically separate samples of the Texel, a meat sheep known for its growth and muscling (Table S1). When Texels were grouped and compared against all other animals, a strong peak was detected on Chromosome 2 (Figure 7). The peak spans GDF8, a gene known to carry a mutation in Texel responsible for muscle hypertrophy [41].
Access to patterns of SNP diversity within a global sample of domestic sheep was used to examine the population history of a species amongst the first to be domesticated by man. Our analysis revealed this domestication process must have involved a genetically broad sampling of wild stock. Approximately 75% of modern sheep breeds have retained an effective population size in excess of 300, higher than cattle and much higher than most breeds of dog. This suggests a highly heterogeneous pre-domestication population was recruited, and the genetic bottleneck which took place was not as severe during the development of sheep as for some other animal domesticates. It is also possible that cross-breeding with wild populations persisted following the initial domestication events to generate the diversity observed. Surveys of ovine mtDNA variability support a broad genetic base during domestication, with at least five lineages identified within modern breeds that diverged well before domestication approximately 11,000 years ago [4],[11],[42],[43]. Three aspects of the SNP diversity documented in this study indicated high levels of gene flow have occurred between populations following domestication. First, a high degree of conservation in LD phase and haplotype sharing across short chromosomal distances was detected amongst almost all breeds independent of geographic origin. Secondly, we did not detect a strong association between genetic diversity and physical distance from the domestication centre, and thirdly, the proportion of variation explained by principal component analysis suggests a weak global population structure. High gene flow and introgression between breeds has been postulated previously, based on the phylogeographic distribution of mtDNA lineages [3],[4]. In addition, human-mediated transportation of sheep is well documented including the export of wool sheep from Italy during the Roman period and use of British sires on the European continent from the early Middle Ages onwards [44]–[46]. What remained unclear until now, however, was the extent of admixture that accompanied these sheep transportations and the high diversity this has left within many breeds.
Inspection of a much larger number of SNP than in previous studies [9] allowed PCA and model-based clustering to successfully detect a clear phylogeographic pattern within the breeds genotyped. At a global scale, clear genetic divisions were detected separating European, Asian, and Africa sheep. This division likely reflects variation between the populations that participated in the earliest migrations outwards from the domestication centre. At the breed level, isolated populations were identified as outliers in PCA with low Ne (e.g., Soay, Wiltshire Horn, and Macarthur Merino). Conversely, sheep from the Americas (Brazil and the Caribbean) had high Ne and clustered separately from European, African, or Asian populations. Decomposing the genome into two or more components (K<2; Figure S4) revealed a genetic origin for Caribbean breeds in common with African animals mixed with those of Mediterranean Europe. Similar results have been observed for New World Creole cattle [24]. This likely reflects the transportation of animals during the migration of enslaved West Africans bought to the Caribbean as slave labourers starting in the 1500s and the introduction of sheep by European colonialists.
The observed patterns of genetic variation used to make inferences about population history can be explained by neutral fluctuations and the action of genetic drift. Not all loci tested in this experiment, however, appeared neutral as clear evidence was obtained for accelerated divergence in response to selection. A genome-wide scan for differentiation using global FST revealed 31 chromosomal regions with evidence for selection. It is important to recognise that genome scans such as this, even when conducted using a meaningfully large number of loci and animals, have several limitations. Foremost amongst these is that the identification of SNP displaying outlier behaviour is not, in itself, proof that selection has taken place. Where convincing signals are detected, it can be difficult to clearly identify the target of selection within a region, and even more difficult to establish the link between selection and its morphological consequence. In this study the strongest selection signal was identified immediately adjacent to RXFP2, a gene involved in reduced bone mass and sexual maturation [47],[48]. Strong evidence supports that RXFP2 was targeted by breeding for the removal of horns, likely to be one of the oldest morphological modifications that accompanied domestication [49],[50]. The gene underpins QTL for horn morphology [23] and the selection signal was reconstituted only when comparing horned with polled populations. Taken together the results represent a rare example where selection has been detected and demonstrated to have occurred in response to a clearly identified human-mediated breeding objective. Given the long-standing nature of the selection, it was surprising it gave rise to the strongest selection signal. Our interpretation is that this reflects the widespread frequency of polled animals across a large number of breeds, as this assists in generating extreme FST when calculated across all breeds. Conversely, strong selection at a locus that is private to only one or two breeds is not reflected using the global FST metric. Selection surrounding Myostatin in Texel illustrates this clearly, as a strong signal is revealed when Texels are compared with all other breeds, but it is absent from the 31 regions identified using the full dataset (Figure 7 and Table 1). Analysis was performed to search for selection signatures common to more than one domesticated species. It seems reasonable to expect common signals may exist, given some breeding goals are constant across livestock species. One example is man's desire to breed animals that display consistent pigmentation type within breeds. It follows that key pigmentation genes may show evidence for selection in more than one species, and indeed that is what was detected here for genes such as MITF and KIT (Table S6).
In summary, the phenotypic variability and population history of domestic animals make them an appealing model to study the consequences of selection. This promise is being realised through the recent availability of meaningfully large collections of SNP. Applied here, patterns of diversity were examined to systematically identify genomic regions in sheep that have undergone accelerated change in response to selection. Identification of the adaptive alleles within each genomic region remains a challenge. If resolved, the outcome will be knowledge describing the functional variants that characterise differences between breeds. The analysis of genomic polymorphism conducted here carries practical consequences. With the division of animals into breeds during the last few hundred years, animal breeding has witnessed a dramatic change. Most recently, the identification of superior rams and their disproportionate genetic contribution via artificial insemination has lifted the pace of genetic gain for production traits. The population-level consequence is a dramatic reduction in effective population size, which is best illustrated for cattle where the sharp decline in Ne already threatens breed viability [15]. The finding here that the majority of breeds have retained a high genetic diversity and effective population size implies that selection response for wool, meat, adaptation, and welfare traits may be expected to continue.
The number of animals per population and geographic origin of breed development is given in Table S1. Individuals were collected from multiple flocks to capture a representative sample of within-breed genetic diversity. Beadchip array manufacture and genotyping was performed by Illumina (San Diego, CA) before raw signal intensities were converted into genotype calls using the Genome Studio software. SNP that failed any of the five following criteria were removed: (1) markers with <0.99 call rate; (2) markers identified during clustering as having atypical X-clustering, evidence for a nearby polymorphism, compression, intensity values only, or evidence of a deletion; (3) SNP with minor allele frequency equal to zero; (4) SNP with discordant genotypes identified by comparison of 10 animals genotyped independently at Illumina (San Diego, CA) and GeneSeek (Lincoln, NE); and (5) SNP showing Mendelian inconsistencies within 44 trios (dam, sire, and offspring) and the International Mapping Flock [51]. A total of 5,207 were removed (Table S3), leaving 49,034 SNP. Genotypes are available formatted for analysis in PLINK [52] from the ISGC website [53].
Five metrics were used to estimate levels of within-breed genetic diversity (Table S4). The proportion of polymorphic SNP (Pn) gives the fraction of total SNP that displayed both alleles within each population. Expected heterozygosity (He) and the inbreeding coefficient (F) were estimated using PLINK [52], while allelic richness (Ar) and private allele richness (pAr) were estimated by ADZE [54].
Analysis of allele frequency distributions, plotted separately for SNP identified by Roche 454 and Illumina GA sequencing, indicated the presence of ascertainment bias (Figure S1). To determine its effect on estimates of genetic relatedness between populations, Reynold's distance was calculated between breeds using five different subsets of SNP (Figure S11). The SNP sets were as (i) all 49,034 SNP, (ii) 33,115 SNP identified using Roche 454, (iii) 15,427 SNP identified using Illumina GA, and (iv) 22,678 SNP identified by application of LD pruning using PLINK–indep (50 5 0.05). This calculated LD between SNP in windows containing 50 markers before removing one SNP from each pair where LD exceeded 0.05 and (v) 20,279 SNP polymorphic in non-domestic sheep that were SNP pruned using LD as described for (iv). The resulting five NeighborNet trees were almost identical, indicating ascertainment bias did not have a large impact on the interpretations based on genetic distance. The removal of SNP in high LD has been shown to counter the effect of ascertainment bias and generate meaningful comparisons between populations [10]. LD-based pruning as described above preferentially reduced mean SNP heterozygosity within European populations used heavily during SNP discovery.
In order to understand the relationship within and between breeds across each major geographic group, Principal Components Analysis (PCA) was performed using EigenStrat [55]. Initial PCA using all 2,819 animals revealed six breeds containing in excess of 100 animals skewed the clustering. This prompted a reduction in the number of animals used, where 1,612 animals were randomly selected to ensure 26 or fewer animals were included per breed (Figures 2 and S3). To ensure uncorrected LD did not distort the PCA [55], SNP pruning was used to identify two SNP sets. First, all 49,034 markers were subjected to LD-based pruning (>0.05) using PLINK to identify 22,678 SNP. Secondly, 32,847 SNP that retain polymorphism within wild feral sheep were subjected to the same LD-based SNP pruning (>0.05) to identify 20,279 SNP. The PCA results obtained did not differ significantly dependent on the SNP set used. Model-based clustering was performed using the admixture model, correlated allele frequencies, and 15,000 burnin and 35,000 simulation cycles in STRUCTURE version 2.3 [18]. Convergence was checked using two runs for each value of K (number of subpopulations). For supervised clustering, prior population information was introduced from six meta-populations consisting of regional pool of breeds considered to represent ancestral populations. The same meta-populations were used for updating the allele frequencies during the simulations. NeighborNet graphs were constructed from a matrix of Reynolds' distances using Splitstree [56].
To estimate historic effective population size for each breed, the degree of linkage disequilibrium (LD) was calculated as r2 between all SNP pairs where MAF for each SNP in the pair was >0.10. r2 values were grouped into bins based on the distance between SNP from the physical map. Nt was then calculated as (1−r2)/(4cr2), where c is the distance between the SNPs in Morgans (we assumed 100 Mb = 1 Morgan) and Nt is the effective population size t generations ago, where t = 1/2c. The most recent estimate of effective size was taken as Nt when c = 1 Mb. We performed simulations to assess the sensitivity of the estimates of effective population size over generations based on LD, in populations with and without admixture events (Figure S6). A mutation-drift model was used in the simulations following [22]. The population consisted of individuals made up of a chromosome segment 50 Mb long with 6,901 SNP. A population of individuals was simulated with an initially very large population size 10,000 generations ago, declining to a small effective population size in recent generations. In the final 420 generations, the population was split into two “breeds.” In the non-admixed population, there was complete divergence between the breeds for the 420 generations. In the first admixed population, there was an admixture event, with crossing between the breeds (matings chosen at random across the two breeds) 220 generations ago. The admixing lasted 20 generations, after which the breeds diverged for a further 200 generations, with no more admixture events. LD (r2) was calculated between all marker papers and Ne estimated at different times in the past as described for the real data. Five replicate simulations were performed for each scenario. The extent of haplotype sharing among populations was characterised with the r statistic, where r is a signed r2 [19]. A high correlation between r values for all locus pairs separated by the same physical distance among two breeds requires that the same haplotypes are found within both breeds. This means the sign of the r statistic is preserved across breeds only if the phase relationship among alleles is the same in both populations (leading to a high value for r if this is the case). The correlation of r between breeds was calculated for SNP separated by <10 kb, 10–25 kb, 25–50 kb, 50–100 kb, and 100–250 kb (Figures S7, S8, S9). There will be some error in calculating the correlation of r between two breeds due to finite sampling of haplotypes within a breed (e.g., limited sample size). To determine the extent of this error, we calculated the correlation of the r values at these different lengths of haplotypes for the Merino and Industry Merino samples, which are samples from the same breed. This gave a correlation between the r values for each bin size of 0.6. All correlations of r values for all breed comparisons were then divided by 0.6 to correct for sampling. Only corrected values are presented. As detailed in [19] the change in correlation of r between two breeds with increasing marker distance can be used to estimate generations since divergence from a common ancestral population. From [19], the expectation for r after T generations of divergence is E(rT) = e−2cT. The natural logarithm of the expected correlation of r then follows a linear decrease as a function of distance with slope −2T, and this was used to calculate divergence time between all breeds (Figure S10).
Global FST was calculated as described by [57]. Raw values were ranked and used to identify regions under position selection. Centred on the top SNP (0.1%), neighbouring markers were included until consecutive markers were encountered ranking outside of the top 5%. The second marker was excluded and the Mb position of each region was determined using sheep genome assembly version 1. SNP-specific FST values were smoothed using a local variable bandwidth estimator as described in [35] and plotted as a line in Figures 6 and 7. To identify genomic regions with shared selection signals across breeds, raw FST within each population was smoothed into 500 genomic divisions (98 SNP per region). The number of breeds with smoothed FST in excess of one standard deviation of the mean was plotted for values at each tail of the distribution. Analysis was performed to identify gene ontology (GO) terms that were significantly overrepresented in 181 genes residing within the 31 regions under selection (Table 1). The terms associated with the 181 genes were compared against a background set of 11,098 genes. Each of the 11,098 genes contain a SNP present on the SNP50 Beadchip, or a SNP within 2.5 Kb. Comparison of the two gene lists (target and background) was performed using the software GOrilla, which implements a hypergeometric distribution and mHG p value approach to determine significance [58].
|
10.1371/journal.ppat.1006678 | Single-cell analysis identifies cellular markers of the HIV permissive cell | Cellular permissiveness to HIV infection is highly heterogeneous across individuals. Heterogeneity is also found across CD4+ T cells from the same individual, where only a fraction of cells gets infected. To explore the basis of permissiveness, we performed single-cell RNA-seq analysis of non-infected CD4+ T cells from high and low permissive individuals. Transcriptional heterogeneity translated in a continuum of cell states, driven by T-cell receptor-mediated cell activation and was strongly linked to permissiveness. Proteins expressed at the cell surface and displaying the highest correlation with T cell activation were tested as biomarkers of cellular permissiveness to HIV. FACS sorting using antibodies against several biomarkers of permissiveness led to an increase of HIV cellular infection rates. Top candidate biomarkers included CD25, a canonical activation marker. The combination of CD25 high expression with other candidate biomarkers led to the identification of CD298, CD63 and CD317 as the best biomarkers for permissiveness. CD25highCD298highCD63highCD317high cell population showed an enrichment of HIV-infection of up to 28 fold as compared to the unsorted cell population. The purified hyper-permissive cell subpopulation was characterized by a downregulation of interferon-induced genes and several known restriction factors. Single-cell RNA-seq analysis coupled with functional characterization of cell biomarkers provides signatures of the “HIV-permissive cell”.
| CD4+ T cells are the main target of human immunodeficiency virus (HIV) infection. However, CD4+ T cells are not equally permissive to infection, varying between individuals and across cells isolated from the same individual. We explored cellular heterogeneity by analyzing the transcriptome profile of CD4+ T cells at single-cell level. Results identified T-cell receptor-mediated activation as the major determinant of CD4+ T cell heterogeneity. We identified cell surface proteins that highly correlated with T cell activation and HIV permissiveness. Activated CD4+ T cells expressing CD25, CD298, CD63 and CD317 were highly enriched for HIV permissiveness. The single-cell analysis approach used in this study allowed for the identification of the most HIV-permissive cell.
| In vivo and in vitro data indicates that only a small fraction of the CD4+ T cell population is successfully infected by HIV. Cellular permissiveness to HIV infection differs between cell lines originating from different tissues, between T cell lines [1], between primary CD4+ T cells isolated from different HIV-negative blood donors [2], as well as between primary CD4+ T cells from the same donor [3]. HIV permissiveness in primary CD4+ T cells has been linked to: (i) activation, i.e. proliferating activated CD4+ T lymphocytes are more susceptible to HIV infection compared to resting CD4+ T cells [4–8]; (ii) specific CD4+ T cells subsets, such as effector memory cells [8, 9] or specifically, CCR4+CCR6+ and CXCR3+CCR6+ CD4+ T cells [10]; and (iii) expression of specific HIV dependent/restriction cellular factors that can modulate HIV replication [3]. Differences in permissiveness in human cell lines was recently associated with transcriptional and functional defects in key components of innate immunity [1]. Nevertheless, the specific phenotypic and functional characteristics of primary CD4+ T cells that are permissive to HIV infection remain elusive.
Single-cell technology is a valuable tool to study cellular heterogeneity in different biological settings, including virology [11–13]. Single-cell sequencing permits to explore whether inter-individual differences in successful infection can result from within-individual heterogeneity across individual cells. Cellular heterogeneity within an individual can arise from the presence of different subsets of CD4+ T cells [14, 15], differences in response to TCR activation [16], or other determinants such as the expression of innate immunity genes [17–20]. This suggests a model where a highly permissive individual possesses CD4+ T cells that are enriched in a given cell lineage, activate more rapidly or differently, express less antiviral factors, or a combination of factors.
Heterogeneity at the cellular level can result from differences in cell fate, i.e. permanent and irreversible commitment to a lineage, or in cell state, resulting from transient and reversible processes. Cell fate heterogeneity can lead to a mixture of cells of different types. Cell state differences arise from the intermediate stages of differentiation or activation of a cell type, or from stochasticity of gene expression, cell cycle, pulsating expression, circadian rhythms or level of reactivity to stimuli [21]. There is also increasing attention to the possibility of monoallelic expression, determining the use of different haplotypes in a given moment [22, 23]. Microfluidic control of cell capture and preparation of RNA-seq libraries from single cells allows the study of transcriptional heterogeneity in a reliable way. This methodology was successfully used to identify subpopulations and identify markers from many cell types [24–26].
The aim of this study was to use a non a priori and single-cell approach to identify molecular features that characterize the CD4+ T cell population most permissive to HIV infection. We hypothesized that permissiveness could be a feature of a specific cell lineage(s) (cell fate), or a feature of cells in a particular state. To this end, we performed comparative single-cell RNA-seq analyses and large-scale immunoprofiling on non-infected CD4+ T cells from uninfected individuals with extreme phenotypes of susceptibility to in vitro HIV infection. We showed that cellular activation state (degree of response to TCR stimulation) was the main factor of transcriptional heterogeneity at single cell level, which in turn translated in varying degrees of HIV permissiveness. Importantly, the permissive cell was identified before infection, thus revealing the biological basis of baseline CD4+ T cell susceptibility to HIV. Moreover, this single-cell based approach allowed the identification of specific biomarkers that partition cell populations into high and low permissive subsets. The combination of multiple candidate biomarkers further selected for highly permissive cells, thus defining the “HIV-permissive cell”.
To explore the basis of permissiveness in primary CD4+ T cells at a single-cell level we took advantage of a previous study where a panel of donors was characterized for their heterogeneous permissiveness to HIV infection [3]. Cells from 18 donors were TCR-activated in presence of IL-2 for three days and evaluated for their cell growth and permissiveness using VSV-G-pseudotyped eGFP expressing HIV-based vector (named HIV-GFP, S1 Fig). Two donors, “42” and “123”, were selected as displaying different levels of permissiveness (high and low permissive with ~40% and ~10% GFP+ infected cells, respectively), while showing similar growth capacity (Fig 1A and S1 Fig).
To investigate cell heterogeneity without a priori, uninfected CD4+ T cells from the selected donors were used for single-cell RNA-seq analysis using Fluidigm C1 technology. This technology is based on a microfluidic device that uses pre-defined plates based on cell size to separate single cells. To isolate activated CD4+ T cells, we used the medium size chip that can accommodate cells ranging from 10 to 17 μm (± 2 μm). Although the use of pre-specified size might introduce a bias, it more likely minimizes the difference between the two donors. Population RNA sequencing was performed in parallel on bulk cell preparations. The transcriptomes of 85 and 81 individual cells from the high and from the low permissive donor, respectively, were profiled. An average of 25 million reads per cell was obtained. Of these, an average of 6.6 million (standard deviation 2.7 million) paired reads (fragments) per cell was uniquely mapped. Average gene expression levels across individual cells from the same donor showed high correlation with the expression levels assessed on the equivalent bulk cell samples sequenced by population RNA-seq (r = 0.78 and 0.77 for the high and low permissive donor, respectively). Such high correlations served as a quality control of the overall single-cell sequencing procedure. The pairwise comparisons among all individual cells from the same individual showed a distribution of Spearman rank correlations of gene expression levels with median of 0.72 and 0.58 for the high and low permissive donor, respectively, indicating that transcriptional profiles of individual cells from the low permissive donor were more heterogeneous than those of the high permissive donor (Wilcoxon-rank sum test p-value < 2.2e-16; S2 Fig).
We then explored transcriptional heterogeneity by Sincell software for the statistical assessment of cell-state hierarchies developed in our laboratory [27]. The analysis showed a continuum of transcriptional cell state for both donors, without branching patterns (S3 Fig). This continuum was also observed in the Principal Component Analysis (PCA) of the single-cell libraries from the donors, together with RNA-seq libraries from reference bulk cell samples (resting CD4+ T cells and activated CD4+ T cells at 8h, 24h and 72h after TCR stimulation) (Fig 1B). The two main principal axes–PC1 and PC2- explained 18.5% and 5.4% of the total variance, respectively. PC1 presumably gathered heterogeneity related to the experimental protocol (single-cell versus bulk sequencing). PC2 ordered bulk cell samples most likely according to activation state, following time after TCR-mediated activation (resting, 8h, 24h and 72h), and supported by Pearson correlation with the prototypical activation markers CD25 (r = 0.562 with PC2, while r = 0.089 with PC1). Equivalent PCA analysis of the single-cell transcriptomes without considering the bulk samples confirmed this result (S4 Fig), with the main principal axis of single individual cells (PC1, S4A Fig) highly correlating with PC2 axis of bulk population (S4B Fig, Pearson r = 0.9994). The distribution of individual cells along PC2 showed that most of the cells from the high permissive donor had a transcriptional state closer to a full activation state (Fig 1B, black dots). In contrast, individual cells from the low permissive donor showed higher heterogeneity along this axis (Fig 1B, red dots) with transcriptional states spreading between resting and 72h-activated bulk cell samples. These data are consistent with activation bringing the cells toward a permissive phenotype. Cells from the high permissive donor are more homogenous towards the permissive phenotype, while cells from the low permissive donor are more heterogeneous, with fewer cells that display the permissive phenotype. These data highlight the importance of single-cell approaches to investigate heterogeneity that are otherwise masked by bulk population analyses, as shown by the proximity of the two donor populations at 72h. We also used a supervised approach to evaluate the presence of different cell lineages by examining the expression of 63 genes broadly used to classify CD4+ T cell subpopulations (S5A Fig and S1 Table), including helper (Th1, Th2, Th17), regulatory (T-reg), and memory (Effector and Central memory) CD4+ T cells. Clustering of cells according to these markers did not show a clear patterning. Similar analysis was also performed with 1503 innate immunity genes [19] that led to no obvious clustering of cellular sub-populations (S5B Fig).
In summary, the unsupervised whole transcriptome PCA analysis and the supervised clustering using prototypical CD4+ T cell markers and innate immunity genes identified T cell activation as the major driver of cell heterogeneity and failed to identify the presence of distinct lineages of T cell subtypes in the samples. This opened the door to hypothesize that the cellular state, rather than cell fate and lineage, is the most significant contributor to permissiveness.
To identify protein biomarkers that would help sort and characterize the most permissive cells, we explored the hypothesis of T cell activation having a major role in HIV permissiveness. For this, we focused on genes coding for cell-surface proteins and whose RNA expression levels across single cells correlated with cell activation (PC2 axis, Fig 1B), and tested whether their protein levels at single-cell level associated with permissiveness to HIV (Fig 2). FACS analysis was used to investigate at single-cell resolution the expression level of surface proteins in activated CD4+ T cells and assessed how they associated with successful HIV infection as determined by GFP. This was accomplished with a large-scale surface molecule antibodies screening (Fig 2A, upper panel). Forty-eight hours after activation, CD4+ T cells from the original high permissive donor (donor #42) were infected with HIV-GFP and 24h later, a panel of 332 PE-conjugated antibodies (LEGENDScreen Human Cell Screening, BioLegend) was used to investigate by FACS the co-occurrence of cell surface proteins and successful infection (Fig 2A).
We assessed the Spearman rank correlation between each surface protein expression and GFP levels in each individual cell (S2 Table). These correlation values reflect the association of the marker with the cell permissiveness to productive HIV infection. We then determined the level of association of these genes to activation using single-cell RNA-seq data (PC2 in Fig 1B and S2 Table). Fig 2B shows that the candidate markers with the highest association with HIV permissiveness are also the ones with the highest association with activation state. Moreover, these results show that the main source of transcriptional heterogeneity across single-cells (as captured by the activation state in the second principal component from Fig 1B) drives HIV susceptibility as measured by the correlation of protein marker levels with GFP (adjusted R-squared: 0.20; p-value = 9.7e-16).
From this correlation analysis we selected 14 candidate markers: CD63, CD317, CD25, CD58, CD74, CD48, CD71, CD28, CD26, CD2, TIM3, CD298, CD253 and CD3; whose expression highly correlated with both T cell activation and HIV permissiveness (correlation ≥ 0.3 in both analyses and correlation ≥ 0.5 at least in one of the analyses, Fig 2B). The list includes, as expected, CD25 (IL-2 receptor), a canonical activation marker [28] associated with susceptibility to HIV [29–31]. The correlation between marker expression and HIV permissiveness was further validated in CD4+ T cells from four independent blood donors (Fig 2C). With the exception of CD253, expression of each candidate marker was correlated with permissive status of cells.
To determine if the selected candidate proteins were predictive markers for HIV permissiveness, it was important to exclude that the association observed between the candidate markers and HIV permissiveness was not due to HIV infection per se. In fact, viral infection modulates expression of many cellular genes [2]. Moreover, marker expression (i.e. Mean Fluorescence Intensity (MFI) assessed by FACS) is often shifted or increased in the GFP+ population as compared to the GFP- population (S6 Fig).
Therefore, activated CD4+ T cells were first sorted according to the expression of each candidate biomarker and then infected with HIV-GFP (Fig 3A). Sorting was performed based on MFI expression (high vs low) rather than the proportion of cells expressing or not the marker (+ versus -) (S7 Fig), as some of the selected markers were always present in CD4+ T cells while their MFI could change over time (CD48, CD2, CD298 and CD3; S8 Fig). As depicted in Fig 3B, marker-high populations were always more permissive than marker-low populations for all tested candidate biomarkers, except for CD74 which performed poorly for sorting and which was thus discarded from further experiments. Sorting based on CD137 protein expression was used as control and data showed a similar permissiveness to HIV infection between CD137high and CD137low populations.
As CD25 is a strong activation marker, it might explain the whole phenotype of permissiveness [31]. In order to investigate the possible involvement of additional biomarkers, we first assessed how CD25 co-expressed with the other 11 biomarkers (S9 Fig) and whether these markers were able to further enrich for permissive cells in CD4+ T cells positive for CD25. To this end, TCR-stimulated CD4+ T cells were first sorted for CD25high regardless of the presence or the absence of other markers (MRK), yielding a purified CD25high CD4+ T cell population. This population was then further sorted for each of the other 11 markers to obtain cells CD25highMRKlow and CD25highMRKhigh. The different populations were then infected with HIV-GFP and analyzed by FACS, 48h post-infection (Fig 4A).
As expected, sorting by CD25high enriched for cells permissive to HIV as compared to CD25low and unsorted populations (2.4 fold and 1.6 fold, respectively). The CD25high population can be further separated according to high and low expression level of additional markers, displaying further increased and decreased permissiveness, respectively. Indeed, use of additional markers in sorting resulted in CD25highMRKhigh populations that increased the permissiveness to HIV as compared to CD25highMRKlow (ranging from 4.3 fold to 1.5 fold), confirming that all 11 markers contribute in addition to CD25 to further enrich for cells permissive to HIV (Fig 4B). Further evaluation showed that the increase in permissiveness to HIV observed in CD25highMRKhigh populations as compared to CD25high was also statistically significant for 8 out of the 11 candidates (S3 Table and S10 Fig).
Three markers, CD298, CD63 and CD317, with 2.2, 2.3 and 2.0 fold increase of permissive cells compared to unsorted cells, and 1.4, 1.5 and 1.3 fold compared to CD25high cells, respectively, were selected for further analyses. Expression of these markers in different CD4+ T cells subsets, i.e. naïve versus memory cells 48h post-TCR stimulation, showed similar profiles, suggesting that the association of these markers with HIV permissiveness remains valid across cell lineages (S11 Fig).
We tested the additive contribution of the 4 biomarkers, CD25, CD298, CD63 and CD317, to identify highly permissive cells. While the 4 markers generally co-express two by two (S12A Fig), only ~20% of activated CD4+ T cells highly co-express the 4 markers simultaneously (S12B Fig). These results led us to proceed with the sorting of the CD25highCD298highCD63highCD317high population, where TCR-activated CD4+ T cells were sorted successively for high expression of CD25, CD298, CD63 and CD317 cells and then infected with HIV-GFP (EF1-GFP). As shown in Fig 5A, the progressive addition of markers increased in a statistically significant manner the amount of permissive cells with an increase of 28 fold from the unsorted population to the four marker combination (S4 Table and S13 Fig). The role of these markers to HIV permissiveness was confirmed with a full-length, replication-competent CXCR4-tropic virus expressing GFP (NLENG1), entering the cells via CXCR4-mediated entry rather than VSV-G (Fig 5B). Activated CD4+ T cells, from different donors, infected with NLENG1 showed a statistically significant increase of permissiveness to HIV infection up to 5.4 fold with the four markers (Fig 5B, S4 Table and S13 Fig). Taken together, these results indicate that the initial selected markers, CD25, CD298, CD63 and CD317, whose expression was associated with activation at single-cell level both at mRNA and protein levels, identify highly permissive cells. These cells are characterized by an intracellular environment supportive of HIV infection, from entry to protein expression. Due to technical limitations and increasing rarity of specific cell subpopulations, our analyses concentrated on 4 selected markers. However, the use of additional candidate markers should further improve the selection of the highest permissive cell (S14 Fig).
The availability of novel markers allowed the sorting of a unique cell subpopulation for characterization of the transcriptome. RNA was extracted from the sorted subpopulations according to MRK expression and used for RNA-Seq. Principal Component Analysis showed separation and ordering of cell subpopulations according to their permissiveness to HIV infection, i.e. from less to more permissive: CD25low, CD25highMRK3low (CD25highCD298lowCD63lowCD317low), CD25high and CD25highMRK3high (CD25highCD298highCD63highCD317high) (Fig 6A).
Differential expression analyses among sorted CD25high and CD25highMRK3high subpopulations identified 96 genes (fold change higher or lower than 2 and adjusted p-value of < 0.001; DESeq2 test [32]), that could be involved in the observed differences in HIV permissiveness. 95 genes were differentially downregulated in CD25highMRK3high as compared to CD25high (Fig 6B and S5 Table). Functional enrichment analysis established that downregulated genes were in the type I interferon pathway (GO:0060337, False Discovery Rate 4.17e-06) and the defense response to virus (GO.0051607, FDR = 1.32e-04) (S6 Table). Genes involved included important effectors such as: IFIT2, IFIT3, IRF7, ISG15, ISG20, MX1, RSAD2, XAF1, IFI44L, and IL23A. In sharp contrast, the 95 genes appeared as highly expressed in both CD25low and CD25highMRK3low subpopulations. These observations further stress the relationship between activation state and innate immunity defense, in line with recent work [1]. Although not among the 96 genes differentially expressed between sorted CD25high and CD25highMRK3high populations (fold change higher or lower than 2 and adjusted p-value of < 0.001), the expression profile of the 5 prototypical HIV restriction factors still clustered the subpopulations similarly, i.e. according to the permissiveness phenotype (S15 Fig).
CD4+ T cells are the main target of HIV infection and the major cellular reservoirs of HIV in vivo. Understanding the heterogeneity of these cells in terms of permissiveness to HIV is crucial for the characterization of HIV infection and pathogenesis. In this study, we exploited the natural heterogeneity within and across individuals to explore differences between permissive and non-permissive cells by single-cell RNA-seq. This approach allowed the characterization of response to TCR activation as the main driver of heterogeneity. Moreover, coupling single-cell RNA-seq analysis with a high-throughput antibody screen for cell surface protein expression led to the identification of biomarkers for the HIV permissive cell.
The analysis of single-cell RNA-seq allowed assessing the dynamic nature of TCR activation. We observed a continuum of individual cells transiting intermediary activation states that recapitulated transcriptional changes occurring from resting CD4+ T cells to late times of activation. We then identified sets of genes reflecting such heterogeneity. Supervised analyses of RNA-seq levels of cell surface markers prototypical of CD4+ T cell subpopulations as well as of innate immunity genes did not make apparent the existence of distinct lineages of cells. Taken together, our results indicate that the activation state was the main component of CD4+ T cell transcriptional heterogeneity.
Activation state of CD4+ T cells has been extensively linked to HIV permissiveness in population, bulk samples. It is well known that quiescent cells are refractory to HIV infection [5, 33]. Here, we have further associated activation and permissiveness at single cell resolution characterizing both phenotypes at transcriptional and proteomic levels. In addition, the use of a high-throughput antibody screen assessing expression of more than 300 cell surface proteins allowed investigating protein expression with HIV infection success. This combined analysis showed that the more a gene and the encoded protein correlate with activation, the more they correlate with permissiveness.
The top candidate markers that correlated with both activation and permissiveness were further validated as predictors for HIV permissiveness. Expression of cell surface proteins allowed for sorting live cells from activated CD4+ T-cell populations from different donors. Individual markers identified cell subpopulations with greater susceptibility to HIV infection than the general activated cell population. The experimental approach examined the role of CD25 expression for HIV permissiveness and identified additional surface markers, which combined with CD25 led to the identification of a unique population of CD4+ T cells that is highly permissive to HIV. In addition to CD25, the best predictive biomarkers for permissiveness were CD298, CD63 and CD317.
CD63 is a tetraspanin associated with the membranes of intracellular vesicles. CD63 can also be expressed at cell surface, possibly upon T cell activation, thereby promoting sustained T cell activation and cell proliferation [34]. CD63 has been described to participate in HIV infection in early and late steps of the HIV life cycle, both in macrophages and CD4+ T cells [35–38]. CD317, also known as BST-2 or tetherin, is a lipid raft associated protein that restricts HIV infection by retaining nascent virions at the cell surface and preventing their release [39]. CD298 is the beta3 subunit of the Na,K ATPase, known to maintain the electrochemical gradients of sodium and potassium across the plasma membrane [40]. CD298 has been associated as regulator of T cell activation independently of the ATPase alpha subunit [41]. Interestingly, CD298 was shown to be an antagonist of BST-2 by binding and inducing BST-2 degradation and therefore facilitating HIV replication [42].
The transcriptional profile of CD25highCD298highCD63highCD317high expressing cells revealed a specific signature as compared to the CD25high cell population, characterized by the down-regulation of 96 genes, most of which are involved in innate immunity. These results are consistent with previous model suggesting that activated CD4+ T cells are more permissive to HIV infection than resting CD4+ T cells in part due to reduced innate immune responses [1]. The specific signature of this cell population highlights the potential of single-cell RNA-seq analysis as a general pipeline for biomarker identification.
In conclusion, single-cell RNA-seq allowed the investigation of CD4+ T-cell heterogeneity prior to infection and led to the identification of markers for permissiveness to HIV cellular infection. Their combined use serves to identify a cellular state that defines the “HIV permissive cell”. A similar approach should be used to identify the biomarkers of the resting cell predicting its activation potential, and thus its permissive phenotype. Finally, the relevance of these highly permissive cells in vivo, i.e. will these cells more likely be infected and then become latent and contribute to the reservoir or will they more likely die and be depleted, remains to be elucidated.
All blood donors have provided written informed consent. All samples were anonymized.
Peripheral Blood Mononuclear Cells (PBMCs) from different healthy blood donors were purified by Ficoll gradient separation, using Leucosep tubes (Greiner), followed by CD4+ T cell isolation using negative selection and magnetic separation, according to manufacturer’s instructions (Stemcell Technologies). Total CD4+ T cells were isolated using EasySep Human CD4+ T Cell Enrichment Kit, naïve cells were isolated using EasySep Human Naïve CD4+ T Cell Enrichment Kit and memory CD4+ T cells were isolated using Human Memory CD4+ T Cell Enrichment Kit. Cell enrichment yielded a purity of CD4+T cell populations of up to 96%.
Non-activated CD4+ T cells were maintained in RPMI-1640 culture medium (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS). T-cell receptor (TCR)-mediated activation of CD4+ T cells was performed using CD3/CD28 co-stimulation in presence of IL-2 (100 U/ml, R&D Systems). For this, anti-CD3 antibodies (4 μg) were plated in 1 ml PBS per well of a 6-well plate and incubated overnight at 4°C. Wells were washed once with PBS and cells were added in each well at 2x106 cells/ml in OpTmizer CTS T-Cell Expansion serum-free medium (Gibco), supplemented with 1 μg/ml anti-CD28 antibodies (BD Biosciences).
For the HIV-based lentiviral vector EF1-GFP/VSV-G (named HIV-GFP) production, 2.5 million of HEK293T cells (from [43]) per 10 cm dishes were transfected with 15 μg of pWPI-EF1α-GFP [44], 10 μg of psPAX2 packaging plasmid (gift from Didier Trono (Addgene plasmid # 12260)) and 5 μg pMD2G coding for the VSV-G envelope [45], using calcium phosphate transfection kit (Life Technologies). HIV-GFP particles were then harvested 48h after transfection, filtered through 0.45 μm filters and concentrated by filtration on Centricon units (Centricon Plus-70/100K, Millipore). Replication-competent CXCR4-tropic NLENG1 viruses were produced by transfecting HEK293T cells with pNLENG1-IRES (gift from David N. Levy, [46]) using jetPEI (Polyplus transfection), following manufacturer’s instructions and collected 48h later. Of note, NLENG1 derives from pNL4-3 and contains gfp as a fluorescent reporter gene (eGFP-IRES-nef). Viral titers were measured by HIV-1 p24 Enzyme-linked immunosorbent assay (ELISA) kit (Innogenetics).
Primary CD4+ T cells (100,000 cells) were infected by spinoculation (1500 x g, 25°C, 1h30) in the presence of 4 μg/ml polybrene with 300 ng of HIV-GFP or 25 ng NLENG1. After spinoculation cells were washed and maintained for 48h or 72h in culture with OpTmizer CTS T-Cell Expansion serum-free medium in the presence of 100U/ml of IL-2.
CD4+ T cells from donor #42 (from a previous study [3]) were TCR-activated for 48h as described above and infected with HIV-GFP. After 24h of infection, CD4+ T cells were incubated with the LEGENDScreen Human Cell Screening PE-conjugated Antibodies (BioLegend), according to manufacturer’s recommendations. Briefly, 100,000 cells were stained with each provided PE-antibody for 30 min at 4°C, according to the manufacturer’s protocol, washed with PBS-BSA (1%), fixed with 200 μl of CellFIX (BD Biosciences) and analyzed by FACS Gallios (Beckman Coulter). Marker expression and GFP reporter expression data were analyzed using FlowJo software (Tree Star).
CD4+ T cells were collected at 0h, 24h, 48h, and 72h after TCR activation and analyzed by FACS for longitudinal expression assays. Co-expression analysis of multiple biomarkers was performed at 48h post-activation only. Briefly 100,000 cells were used for each staining, washed twice with PBS-BSA, and incubated at 4°C for 30 min with the different surface antibodies (S7 Table). Cells were then washed twice with PBS-BSA, fixed with CellFIX and analyzed by flow cytometry using FACS BD Accuri C6 (BD Biosciences). Data was analyzed using FlowJo software.
Staining for cell sorting was performed on a cell suspension of 10–40 million CD4+ T cells, in 250–500 μl of RoboSep buffer (Stemcell Technologies). After cell washing, cells were incubated with the different antibodies (S8 and S9 Tables) at 4°C for 30 min, washed and resuspended in Robosep Buffer (1–2 ml) prior to sorting on MoFlo Astrios (Beckman Coulter Life Sciences; Flow Sorting Facility, University of Lausanne).
Cells from the two donors (#42 and #123, from a previous study [3]) were TCR-activated as previously described, and maintained in 2 ml OpTmizer expansion medium with 5% FBS and100 U/ml IL-2 for 72h. Dead cells were removed by centrifugation on a 3 ml Percoll gradient at 800 x g for 10 minutes. Cells were washed, counted and resuspended in PBS. Cells were either used for bulk RNA-Seq or loaded to the Fluidigm C1 platform for single-cell capture and single-cell RNA-seq library generation according to the manufacturer’s protocol. Briefly, a cell suspension of approximately 300,000 cells/ml was introduced into the medium size chip (10–17 μm plate), suitable to capture cells of 10–17 ± 2 μm, and thus able to capture activated cells (usually ranging from 10 to 13 μm). After cell separation and capture, empty or debris-occupied wells were identified by microscope visualisation and discarded from subsequent analysis. cDNA libraries were then produced directly and automatically on the chip with Clontech SMARTer Ultra Low RNA kit for Illumina using manufacturer-provided protocols. Illumina libraries were constructed in 96-well plates using the Illumina Nextera XT DNA Sample Preparation kit according to a protocol supplied by Fluidigm and sequenced on HiSeq2500 machine (Illumina), with 50 bp paired-end, 14 libraries multiplexed per lane.
RNA from bulk samples was extracted using Illustra RNAspin kit (GE Healthcare). mRNA-Seq library preparation was performed with TruSeq RNA sample prep kit (Illumina) starting with capture of polyA-containing transcripts, followed by cluster generation (TruSeq single-end cluster generation kit, Illumina) and high-throughput sequencing on Illumina HiSeq2500 (Genomics Technology Facility, University of Lausanne). 100 cycles single-end sequencing was performed for all bulk samples except for the high permissive/low permissive donors that used 50 cycles paired-end sequencing. Additional bulk samples from resting and activated CD4+ T cells used in the PCA analysis were from published work [47].
Sequenced reads obtained were cleaned before alignment using cutadapt v0.9.5 [48] to remove the adapter if present at the 3' end of the read with an overlap between the read and the adapter equal or higher than 13 bases, and prinseq v0.20 to remove: i) low quality nucleotides at the 3′ or 5′ end of the reads (PHRED score<6); ii) reads with mean PHRED score lower than 20; iii) poly-A/T tail with a minimum length of 13 either at the 5’-end or the 3’-end; iv) poly-N tail with a minimum length of 2 either at the 5'-end or at the 3’-end. Cleaned reads were aligned to the human reference genome with STAR aligner v2.3.0 [49] using the Ensembl gene GRCh37 release 70 annotation file. RUM aligner [50] was used for the first 8 bulk libraries indicated in S10 Table. The number of reads per gene was quantified with HTSeq-count v.0.6.1 [51], with parameters mode = union and type = exon. An average library size of 59,862,736 and 6,618,513 uniquely mapped reads in the bulk and single-cell libraries was obtained respectively. For downstream analysis, log-transformation of gene expression values was performed as the log10 of the number of library size-normalized reads per kilobase of exonic sequence. A pseudo-count of 1 was added previous to the log10 transformation to avoid NA’s: log10(RPKM*59862736/1000000+1). The index of bulk and single-cell RNA-seq libraries, the per-gene raw read-counts matrix and the described log-transformed matrix are provided in S10, S11 and S12 Tables. One of the single-cell libraries from the high resistant donor (“poolT12_2”) appeared as an outlier in initial exploratory analyses and was discarded from downstream analyses. Analysis of transcriptional heterogeneity across single cells was performed using Sincell Bioconductor package [27]. Differential expression analyses used DESeq2 bioconductor package [32]. Functional enrichment analysis was performed using STRING with default parameters [52]. An R script with the code necessary to reproduce all bioinformatics results and figures is provided as S1 File. Versions of R-packages used are detailed in at the end of S1 File.
Statistical analyses were performed using the Prism software (v6.0aGraphPad). All comparisons of HIV permissiveness in non-overlapping subpopulations (single populations or single and multiple sorting populations) were performed using paired t-test.
To assess the significance of the use of additional markers to CD25, thus comparing nested subpopulations, we fitted for each selected candidate marker a linear regression model on the fold-increase observed when adding to the CD25high subpopulation (taken as a basal reference level) the second marker high, while accounting for the Donor effects in the regression model. To make the model robust to departures from normality of the residuals distribution, we used 3 alternative transformations of the dependent variable: absolute rank, relative rank within donor and log-transformation of the fold increase.
|
10.1371/journal.pbio.2006250 | A Notch-mediated, temporal asymmetry in BMP pathway activation promotes photoreceptor subtype diversification | Neural progenitors produce neurons whose identities can vary as a function of the time that specification occurs. Here, we describe the heterochronic specification of two photoreceptor (PhR) subtypes in the zebrafish pineal gland. We find that accelerating PhR specification by impairing Notch signaling favors the early fate at the expense of the later fate. Using in vivo lineage tracing, we show that most pineal PhRs are born from a fate-restricted progenitor. Furthermore, sister cells derived from the division of PhR-restricted progenitors activate the bone morphogenetic protein (BMP) signaling pathway at different times after division, and this heterochrony requires Notch activity. Finally, we demonstrate that PhR identity is established as a function of when the BMP pathway is activated. We propose a novel model in which division of a progenitor with restricted potential generates sister cells with distinct identities via a temporal asymmetry in the activation of a signaling pathway.
| A major goal in the field of developmental neurobiology is to identify the mechanisms that underly the diversification of the subtypes of neurons that are needed for the function of the nervous system. When investigating these mechanisms, time is an often-overlooked variable. Here, we show that in the zebrafish pineal gland—a neuroendocrine organ containing mostly photoreceptors (PhRs) and projection neurons—different classes of PhRs appear in a temporal sequence. In this simple system, the decision to adopt a PhR fate is driven by the activation of the bone morphogenetic protein (BMP) signaling pathway. Following the final cell division of a PhR progenitor, the sister cells normally activate the BMP pathway at different times. When Notch signaling activity is abrogated, activation of the BMP pathway occurs earlier and synchronously, which in turn favors the development of early PhR fates at the expense of later fates. We propose a model in which preventing sister cells from activating a signaling pathway in a synchronous fashion after their final division allows diversification of cell fates.
| The development of a functional nervous system requires the production of an amazing diversity of cell types. The precise identity of each neuron is acquired through a complex process referred to as neuronal subtype specification. Although different molecular mechanisms have been reported to control the specification of neuronal subtype identity, the activity of signaling pathways is at the heart of this process. Bone morphogenetic proteins (BMPs) have been widely linked with neuronal specification, such as in the mouse retina, where they promote the expression of M-opsin at the expense of S-opsin in photoreceptors (PhRs) [1,2]. Cell–cell communication involving the Notch pathway has also been implicated in neural specification [3]. In numerous cases, neuronal subtype specification is influenced by the concomitant activity of several signaling pathways, but the mechanisms underlying how these signals collaborate to establish distinct neural subtypes are only now beginning to be uncovered [4–7]. For instance, progenitors in the p2 domain of the spinal cord have the choice between the v2a and the v2b interneuron fate. In this system, BMP and Notch cooperate to promote the v2b fate [8–11], with Notch acting to promote activation of the BMP pathway in the future v2b cell [5]. These two pathways are also involved in neural subtype specification in the zebrafish pineal gland but with the roles being reversed [6]; BMP operates first to promote responsiveness to Notch signaling during the choice between PhR and projection neuron (PN) fates.
Neuronal subtype specification can also be temporally guided, with different neuronal fates being produced over time from a common pool of progenitors. Indeed, it has been shown that neuronal progenitors in the vertebrate retina and spinal cord, as well as the mammalian cortex and olfactory bulb, generate distinct subtypes of neurons depending on when they are produced. In one model, the sequential production of distinct neuronal subtypes is the result of the evolution in the competence of neuronal progenitors [12]. In invertebrates, feed-forward cascades of temporal transcription factors have been described that control the evolution of competence within neural progenitors (see [13] for a review). Although this mechanism has apparently been conserved in vertebrates, so far only three factors have been identified that promote early or late fates; whereas forkhead box G1 (Foxg1) suppresses and IKAROS family zinc finger 1 (Ikzf1) promotes early cortical fates [14,15], Ikzf1 promotes early fates and castor zinc finger 1 (Casz1) promotes late fates in the retina [16,17]. In addition to transcription factors, it is expected that signaling pathways also contribute to temporally guided mechanisms of fate specification. For instance, neurospheres generated from cortical progenitors undergo temporal transitions that do not occur when cell–cell contact between progenitors is prevented [18]. In a different lineage, the signaling molecule transforming growth factor β2 (TGFβ2) operates as a temporal switch that promotes a late-born identity at the expense of an earlier one [19].
Another mechanism that has been proposed to influence neuronal fates is asymmetric division, which allows for the production of two different fates in sister cells derived from a common progenitor and largely involves asymmetric segregation of fate determinants during division. Asymmetric segregation of Notch pathway components has been extensively described in Drosophila neuronal lineages [3,20]. In vertebrates, however, asymmetric segregation of Notch interactors has been implicated in the decision to remain a progenitor or become a neuron [21,22] but not between adopting distinct neuronal fates. The decision to become a v2a or a v2b interneuron is by far the best-described vertebrate case of a Notch-dependent binary fate decision, but whether asymmetric segregation of Notch interactors plays a role in this specific instance is unclear. Indeed, although the v2a and v2b fates are produced from a common progenitor cell, the division producing these two neurons does not seem to occur at a specific angle, suggesting that a process of asymmetric segregation of fate determinants is unlikely [23]. Finally, the question of how asymmetric division is integrated with the activity of signaling pathways other than Notch and the evolution of competence in progenitors over time has yet to be thoroughly addressed.
The zebrafish pineal is a neuroendocrine organ containing two main populations of neurons: PhRs and PNs. We have previously shown that BMP and Notch cooperate during the acquisition of a generic PhR identity [6,24]. Here, we explore the mechanism underlying the specification of different PhR subtypes in the pineal gland. Using the expression of different opsin genes, we have identified three distinct subpopulations of pineal PhR. Two PhR subpopulations, which express exorhodopsin (exorh) or the parietopsin (PT), are specified sequentially, with exorh+ cells appearing earlier than PT+ cells. Reduction of Notch activity accelerates PhR production and concomitantly shifts the fate of the PhR produced from late PT+ to early exorh+ identity. Gain of BMP activity, on the other hand, promotes ectopic PhR whose subtype identity depends on the time when the activation of BMP is triggered. Using time-lapse confocal microscopy, we show that PhR-generating progenitors predominantly produce sister cells with a different timing of BMP activation. In contrast, in a context in which Notch activity is reduced, BMP activation occurs either more frequently before the final division or with more symmetric timing in sister cells. Our results suggest a model in which division of a PhR-restricted progenitor generates two sister cells that activate the BMP signaling pathway at different times, resulting in the acquisition of either an “early” or “late” PhR subtype identity.
Opsins are G-protein-coupled receptors that enable cells to sense a specific spectrum of wavelengths and intensities of light [25]. The expression of several opsins has been reported in the zebrafish pineal gland. For instance, the expression of exorh and red cone opsin (red) has been described in the developing and adult pineal gland using in situ hybridization [24,26], and the expression of PT, a green-sensitive photopigment that belongs to the so-called non-visual opsins, has been described using reverse transcription PCR (RT-PCR) [27,28]. To address whether these opsins are expressed in overlapping or restricted PhR populations, we mapped their expression relative to each other using double in situ hybridization. We found that their expression is largely restricted to distinct subpopulations of cells (Fig 1A, 1B, 1C, 1D, 1E, 1F, 1G, 1H and 1I), with only a few embryos showing one or two cells coexpressing exorh and PT or exorh and red (see quantification in Table 1); no coexpression was observed between PT and red. In parallel, we performed in situ hybridization against the endogenous opsins coupled with immunostaining for green fluorescent protein (GFP)/cyan fluorescent protein (CFP) in a previously described Tg(exorh:EGFP)ja1 transgene [29] or a transgenic reporter for the PT gene, Tg(-2.2parietopsin:CFP), that we established for this study. In the Tg(exorh:EGFP)ja1 background, 92.3% of the GFP+ cells express the endogenous exorh gene (S1A, S1B and S1C Fig); in addition, extensive overlap was also observed between CFP and the endogenous PT gene (S1P, S1Q and S1R Fig) in Tg(-2.2parietopsin:CFP) transgenic embryos, suggesting that the two transgenes recapitulate endogenous expression. Finally, no overlap was detected between the two transgenes (Fig 1J, 1K and 1L, Table 1). Cross comparisons between the expression of the endogenous opsins and transgenes revealed similar levels of coexpression of exorh:EGFP with PT or red (S1D, S1E, S1F, S1G, S1H and S1I Fig) and PT:CFP with exorh or red (S1J, S1K, S1L, S1M, S1N and S1O Fig) as between the endogenous genes (Table 1). Since cells coexpressing exorh and PT or exorh and red are rare (<2 cells) and not found in all embryos, our interpretation is that these cells represent inappropriate specification events rather than hybrid cell fates.
Two PhR subtypes have already been described in the embryonic pineal gland, as defined by the expression of rhodopsin (rhod) and Arrestin 3a (Arr3a) [30]; these populations were called rod and cone PhRs in reference to the morphology of cells expressing these genes in the retina. We found that rhod mRNA expression is largely restricted to exorh:EGFP+ cells in the Tg(exorh:EGFP)ja1 transgene (S2A, S2B and S2C Fig). Similarly, we used a Tg2PAC(opn1lw1:GFP,cxxc1:RFP) transgenic line [31], in which the red+ population of the pineal gland is labeled with red fluorescent protein (RFP), to establish that the red+ fate we describe corresponds to the Arr3a+ population described previously (S2D, S2E and S2F Fig; [30]). Given that the combined number of exorh, PT, and red+ cells accounts for the total number of PhR and that at 48–54 hours post fertilization (hpf) the average pineal gland contains 20 exo+ cells, 15 red+ cells, and 6 PT+ cells, we conclude that there are three subpopulations of PhR in the pineal gland, which at these stages corresponds to a composition of 48% exorh+, 36.6% red+, and 14.6% PT+ cells.
As a previous study has provided insights into the specification of the red+/arr3a+ PhR fate [30], we chose to look more closely at the exorh+ and PT+ populations. These subpopulations do not appear to occupy a specific region of the pineal gland except that the PT+ population seems to be more peripheral (Fig 1D, 1E, 1F, 1G, 1H, 1I, 1J, 1K and 1L); PT+ cells are found in the center of the pineal gland at early stages (Fig 2D), suggesting that this peripheral pattern is reached secondarily. We noted, however, that whereas exorh expression can already be detected in the pineal at 24 hpf (Fig 2A), PT expression was not detected at 24 hpf and was detected only in 2 out of 10 embryos at 26 hpf (Fig 2B and 2D). Further quantification between 24 and 30 hpf confirmed that the exorh+ and PT+ pineal PhR populations are specified in a temporal sequence (Fig 2F).
Analysis of the expression dynamics of exorh and PT suggests that these two PhR subtypes are specified at different stages during development, leading us to test the hypothesis that timing might play a role in their specification. Using a transgenic line, Tg(hsp70l:dnXla.Rbpj-MYC)vu21, in which a dominant negative form of the Notch effector recombination signal binding protein for immunoglobulin kappa J region (Rbpj) is overexpressed upon heat shock, we found that higher numbers of pineal cells express the pan-PhR marker anaat2:GFP [24] at 26, 36, and 42 hpf (but not at 48 hpf) relative to control siblings when heat shock was performed at 14 hpf (Fig 3A, 3B and 3G), suggesting that PhR production is accelerated upon reduction of Notch activity; this relatively late heat shock elicits only a limited effect on the PN population (S3A Fig) compared to situations in which Notch signaling has been impaired from earlier stages [24]. We next addressed whether premature PhR production modifies the subtype of PhR produced and found an increase in the number of exorh+ cells and a reciprocal decrease in the number of PT+ cells (Fig 3C, 3D, 3E, 3F and 3H). Similar results were obtained using N-[N-(3,5-difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester (DAPT; the gamma-secretase inhibitor) (S3B and S3C Fig), a pharmacological inhibitor of Notch activity. Taken together, these data suggest that advancing PhR production favors early PhR fate at the expense of late PhR fate in the pineal gland.
To begin to address how a reduction of Notch activity could promote early production of PhR and the concomitant shift to early PhR fates, we analyzed the lineage relationships between PhR and PN in wild-type and Notch-impaired contexts. For this, we performed time-lapse confocal analyses on Tg(aanat2:GFP)y8 embryos from 15 hpf. Embryos were injected with synthetic mRNA encoding a Histone2B:RFP fusion protein at the one-cell stage to label all nuclei and permit backtracking aanat2:GFP+ cells identified at the end of the time-lapse acquisitions, and embryos were labeled for HuC/D expression at the end of the time-lapse series to identify cells that had adopted a PN fate. In an initial set of experiments, embryos were imaged for a duration of 20 hours, and based on the number of anaat2:GFP+ and HuC/D+ cells at the end of the movie, we estimate that the embryos reached 33–35 hpf at the end of the acquisition. Our results show that PhRs were born either from divisions generating 2 PhRs or a PhR and a cell with no defined identity (which we will refer to as "ø cell") (Fig 4; S1 Movie, S4 Fig, S2 Movie); ø cells do not express the PN marker HuC/D and do not divide a second time during the time-lapse series (S4 Fig, S2 Movie). Similarly, we observed that PNs were born from divisions resulting in 2 PNs or a PN and an ø cell (S5 Fig). Given that PhR–PN divisions were never detected, we hypothesized that most pineal neurons originate from fate-restricted progenitors at their last division and that in the cases of the PhR–ø divisions, the ø cell represents a future PhR that has not yet differentiated. To confirm this, we imaged embryos until they reach a stage close to 46 hpf based on the number of anaat2:GFP+ and HuC/D+ cells. Here, we found an increase from 67.8% to 92.3% of PhR–PhR divisions compared to the shorter acquisitions, supporting the notion that ø sisters of PhRs that have not yet acquired their PhR identity at 33–35 hpf will ultimately become PhRs (Fig 4I and 4J); again, no PhR–PN divisions were detected. Our results thus suggest that the vast majority of PhRs are born from fate-restricted progenitors at their last division.
To address whether impairing Notch signaling affects the lineage relationships described in wild-type embryos, we next performed similar lineage experiments in Tg(hsp70l:dnXla.Rbpj-MYC)vu21 embryos heat shocked at 14 hpf. We found that lineage outcomes were comparable to those in wild-type siblings, with PhR being generated from a PhR fate-restricted progenitor; the percentage of divisions generating two aanat2:GFP+ PhR was 75% in Notch-impaired embryos compared to 67.8% in the wild-type siblings (Fig 4I). We conclude that reducing Notch activity from 14 hpf onwards does not modify the lineage relationships between PhR and PN.
Activation of the BMP pathway is both necessary and sufficient to promote a PhR fate in the pineal [6]. A mechanism to explain how reducing Notch activity affects the timing of PhR determination could, therefore, involve the premature activation of the BMP pathway. To test this, we took advantage of the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgenic line, in which a Bmp-responsive element (BRE) from the id1 locus has been placed upstream of GFP to create a transgene that behaves as a faithful reporter of BMP pathway activation, including in the pineal gland [32]. We counted the number of BMPRE-AAV.Mlp:EGFP+ (Bre+) cells in wild-type or Tg(hsp70l:dnXla.Rbpj-MYC)vu21 embryos at 22 hpf after heat shock at 14 hpf. Under these conditions, we observed an increase in BMPRE-AAV.Mlp:EGFP+ cells when the Notch pathway was inhibited (Fig 5A, 5B and 5G). A similar result was obtained using a second independent transgenic line Tg(BMPRE-AAV.Mlp:d2EGFP)mw30 (S6 Fig; [32]).
Activation of the BMP pathway results in the phosphorylation of Smad1/5/8, which can be detected in the pineal anlagen from around 15 hpf and peaks at 18–20 hpf [6]. We tested the possibility that Notch might act to prevent premature phosphorylation of Smad in PhR progenitors. For this, we quantified the number of phosphorylated Smad1/5/8+ (P-Smad1/5/8+) cells in the pineal of wild-type and Tg(hsp70l:dnXla.Rbpj-MYC)vu21 embryos using the Tg(-1.6flh:GAP-EGFP)u711 transgene to delineate the pineal domain. We did not observe premature Smad phosphorylation in PhR progenitors in Tg(hsp70l:dnXla.Rbpj-MYC)vu21 embryos heat shocked at 14 hpf (Fig 5C, 5D, 5E, 5F and 5H). Altogether, these results suggest that Notch activity affects BMP signaling in the pineal downstream of Smad1/5/8 phosphorylation and upstream of the BMP reporter Tg(BMPRE-AAV.Mlp:EGFP)mw29.
To explore the dynamic nature of the effect of impairing Notch, we performed time-lapse analysis using the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgene in a manner similar to that described above in the Tg(aanat2:GFP)y8 background. As before, we observed two types of divisions (Fig 6A): those generating two BMPRE-AAV.Mlp:EGFP+ cells (2 Bre+: 80.4%; n = 46 divisions) and those generating one BMPRE-AAV.Mlp:EGFP+ cell and one BMPRE-AAV.Mlp:EGFP- cell at the end of the acquisitions (Bre+/ø; 19.6%; n = 46 divisions). In contrast to the situation found for Tg(aanat2:GFP)y8, however, expression of the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgene was observed either before or after the final division. To analyze the dynamics of BMP pathway activation in pineal PhR lineages, we compared a set of temporal variables for divisions generating two BMPRE-AAV.Mlp:EGFP+ cells between wild-type and Notch-impaired conditions (Fig 6B). To generate the variables for comparison, we first set the time of final division as zero. Next, we defined the time at which the first daughter cell expresses the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgene as t1 and the time of activation in the second daughter cell as t2. Finally, the asynchrony in BMP pathway activation between the two daughter cells was calculated (Δt = t2 − t1). In a wild-type context, we observed two different types of divisions generating a pair of BMPRE-AAV.Mlp:EGFP+ cells: those in which the expression of the transgene occurs before division (t1 = t2 = 0) and asymmetric divisions (Δt > 0), with the latter being far more numerous (Fig 6C, 6D, 6E, 6F, 6G, 6H, 6I, 6J, 6K, 6L and 6M; S3 Movie). In Tg(hsp70l:dnXla.Rbpj-MYC)vu21 embryos heat shocked at 14 hpf, the final outcome of divisions was globally similar to that observed in a wild-type context, with 81% of divisions generating two BMPRE-AAV.Mlp:EGFP+ cells and 19% divisions generating one BMPRE-AAV.Mlp:EGFP+ and one BMPRE-AAV.Mlp:EGFP- cell at the end of the acquisition (n = 31 divisions, Fig 6A). Unlike the wild-type situation, however, impairing Notch activity led to more symmetric divisions. In particular, we observed divisions in which the expression of the transgene occurs synchronously after division (Δt = 0, t1 > 0), a case never observed in wild-type controls. Furthermore, more cases of expression of the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgene before division were detected in Notch-impaired than in wild-type embryos (Fig 6C, 6N, 6O, 6P, 6Q, 6R, 6S, 6T, 6U, 6V and 6W; S4 Movie).
These results suggest that pineal progenitors respond earlier to BMP signaling when Notch signaling is compromised, and the effect of Notch in this context appears to occur downstream of Smad phosphorylation.
Reducing Notch pathway activity simultaneously accelerates the response to BMP signaling and favors early PhR subtype identity. To address a potential causal link between these two observations, we tested whether the timing of BMP activity controls the subtype of PhRs formed. Forced activation of the BMP pathway in the Tg(hsp70:bmp2b)fr13 transgenic line induces ectopic PhR production [6]. We quantified the populations of each PhR subtype obtained after induction of the Tg(hsp70:bmp2b)fr13 transgene at stages between 10 and 21 hpf (Fig 7A). We detected an increase of the exorh+ fate when the heat shock was performed relatively early (up to 18 hpf, Fig 7B, 7C, 7D and 7H). In contrast, only upon relative late heat shock (16–21 hpf) were the numbers of PT+ PhR increased (Fig 7E, 7G and 7I); activating BMP signaling at 10 hpf reduced the size of the PT population (Fig 7E, 7F and 7I). Together, these results suggest that activating BMP signaling at different stages produces different fates. We propose that the inhibition of the PT+ fate upon early BMP activation reflects the premature determination of exorh+ PhRs, with the consequence that the PhR progenitor pool is depleted at the time when the PT fate should be specified.
In this paper, we have used specification of PhR subtype identity in the zebrafish pineal gland to address how signaling pathways underlie temporally regulated neuronal fate choice. We show that most pineal PhRs are born from a fate-restricted progenitor and that exorh+ PhRs are specified before ones expressing PT+. We show that reducing Notch pathway activity has two concomitant effects on PhR production: it accelerates PhR production and favors the early PhR fate at the expense of a later one. Finally, our results indicate that ectopic activation of BMP signaling induces PhR with an identity that depends on the timing of BMP activation. Based on these findings, we propose a model of temporal specification, which is outlined in Fig 8, and discuss its mechanistic implications.
Our analysis of BMPRE-AAV.Mlp:EGFP+ cells in a wild-type context suggests that in nearly 90% of cases, the divisions that generate these cells are asymmetric in the sense that the activation of BMP activity does not occur simultaneously in sister cells. The timing of BMP pathway activation in these divisions becomes more symmetric in a context in which Notch activity is reduced. Indeed, we observe that reduction of Notch activity increases the percentage of cells that activate BMP prior to division, suggesting that there is an inhibition of BMP activity by Notch at the level of the PhR progenitor. In this regard, the approximately 11% of divisions generating sister cells that simultaneously begin expressing GFP after mitosis in Tg(hsp70l:dnXla.Rbpj-MYC) embryos could represent cells in which gfp is transcribed already in the progenitor but GFP protein is not yet detected. We propose that an important prerequisite for temporally asymmetric divisions is the prevention of a premature fate decision within the progenitor. In invertebrates, sister cells in neuronal lineages often communicate via Notch to acquire their fates [3,20]. At this point, it is unclear whether Notch communication is occurring preferentially between sister cells in the pineal gland rather than between neighbors of unrelated lineage.
Results presented here suggest an intricate interplay between Notch and BMP activities while also expanding the roles for BMP signaling during development of the pineal gland. The pathway is first required to define the dorsoventral position of the pineal anlage [33]. In a second step, it is required for the proliferation of pineal progenitors [6]. These two BMP activities do not seem to be controlled by Notch. BMP and Notch are then required to specify a generic PhR identity and, subsequently, specific PhR subtypes. The pleiotropic effects of BMP signaling on pineal development considerably complicate the identification of BMP targets relevant for each of these processes. Indeed, although classical BMP targets such as id1 and msxb/c/e are expressed in the pineal gland [32,34], it is unclear whether they underlie specific roles for BMP signaling.
Our results have led us to propose the following model (Fig 8). In a wild-type context, Smad1/5/8 is phosphorylated in fate-restricted PhR progenitors but at a level that does not translate into the activation of BMP target genes, because of inhibition exerted by Notch activity. Notch pathway activity thus restrains the progenitor from adopting a PhR fate. After division, this inhibition is progressively released, although not synchronously, and this allows for the production of exorh+ and PT+ PhR, whose fates are dependent on the timing at which they activate the BMP pathway. How PhR precursors evolve from generating only exorh+ cells early to also generating PT+ fates later remains an open question. Future studies concerning how different sets of BMP target genes evolve over time should shed light on this.
We have previously shown that during the PN/PhR fate decision, BMP is required for proper activation of Notch pathway targets, which in turn inhibit the PN fate in the future PhR [6]. The results presented here suggest a previously unanticipated complexity in Notch/BMP cross talk. Indeed, how can we reconcile that BMP signaling is required to activate Notch targets during the PN/PhR fate choice but that Notch pathway activation is required to prevent premature BMP activation during PhR subtype specification? One hypothesis relies on the existence of complexes containing the intracellular domain of Notch (NICD)/Rbpj and P-Smad1/5/8. Our analysis of P-Smad1/5/8 in wild-type and Notch-impaired embryos suggests that Notch inhibits the response to BMP downstream of Smad phosphorylation. The published BMP response element used in this study does not contain Rbpj binding sites [32]. This suggests that the effect of Rbpj on the Tg(BMPRE-AAV.Mlp:EGFP)mw29 transgene is either indirect or involves trapping of P-Smad1/5/8 species in inactive complexes. It has been shown that NICD coprecipitates with Smad1 in the presence of the coactivators P300 and the p300/CBP associated factor (P/CAF) [35]. These interactions have been proposed to reinforce the activation of Notch target genes in neural cells. Similarly, complexes containing NICD and Smad1/5/8 have been detected in cerebrovascular endothelial cells, where they are proposed to activate transcription of N-cadherin via an Rbpj binding site [36]. We propose that NICD/Rbpj/P-Smad1/5/8 complexes in the pineal participate both in the transcription of Notch target genes, as previously suggested [6], but also prevent the activation of BMP targets through a squelching mechanism. In this model, simultaneous activation of Notch and BMP receptors would lead to the formation of NICD/Rbpj/P-Smad1/5/8 complexes, but these complexes would first go to Notch targets, perhaps because of the higher affinity of Rbpj for its target sequences in the genome. The presence of Smad1/5/8 in these complexes would help transactivation of Notch targets, as previously described [35,36]. In a second step, either owing to a reduction in the level of ligands available to bind the Notch receptor or to a Notch-inhibiting signal, Notch activation would progressively decrease while activation of the BMP receptor would be maintained, and this would permit the activation of BMP target genes.
Finally, whereas, to our knowledge, our study provides the first evidence that temporal control of BMP activity is crucial for specification of postmitotic neurons, the importance of a proper timing of BMP activity for specification of progenitor pools has been previously demonstrated in the dorsal spinal cord. In this case, rather than the timing of onset of BMP activity, duration of the exposure to BMP ligands seems the most important variable [37]. It is at present unclear whether the duration of exposure to BMP activity also plays a role in the specification of pineal PhRs.
Examples of vertebrate neuronal lineages in which loss of Notch activity promotes early fates at the expense of late ones are already known [38–43]. In contrast, the mechanisms behind these effects of Notch are unclear. Notch could simply act to slow down neurogenesis and thus indirectly prevent temporal transitions. Alternatively, the pathway could play a more active role in triggering such transitions by promoting switches in the expression of temporal transcription factors in a manner analogous to what has been proposed for TGFβ2 in the hindbrain [19] or by regulating the competence of progenitors to respond to specific signaling pathways whose activity is interpreted differently over time, as is the case for pineal PhR subtype specification. Neurons in the left and right habenular nuclei of zebrafish develop with a temporal asymmetry in identity [38]. In this system, Notch has been proposed to promote identity through a general effect on the timing of neurogenesis [38]. Wnt activity was also recently shown to be necessary for the acquisition of right-sided neuronal phenotypes in the zebrafish habenulae [44]. Thus, an alternative hypothesis would be that Notch acts more directly to limit the competence to respond to Wnt activity. Along the same line, as Notch has been shown to facilitate Sonic hedgehog signaling during the specification of neural fates in the ventral spinal cord [4], it would be interesting to address whether a similar cross talk operates during the temporal neural to glial fate switch occurring within some of these ventrally specified progenitors, as such a switch has been shown to depend on a late burst of Sonic hedgehog activity [45].
Numerous diseases that affect retinal PhRs and lead to blindness have been described. One promising area of research aimed at treating these conditions involves generating PhRs in vitro with the longer-term aim of developing cell replacement strategies [46]. A recent study on the induction of retina from human induced pluripotent stem cells (iPSCs) in culture suggests that inhibition of Notch activity accelerates the production of PhRs [47]. These observations led the author to envisage using pharmacological inhibitors of Notch activity to accelerate the production of these cells. A question that has not yet been answered in this system, however, is whether modifications in timing are concomitant with fate changes in the types of PhRs that are induced, as we describe in the present study. We propose that the zebrafish pineal gland provides a powerful model for understanding molecular mechanisms driving neuronal subtype specification and for addressing specific questions concerning the establishment of distinct PhR identities.
All animals were handled in the CBI fish facility, which is certified by the French Ministry of Agriculture (approval number A3155510). The project was approved by the French Ministry of Teaching and Research (agreement number APAFIS#3653–2016011512005922), in accordance with the guidelines from the European directive on the protection of animals used for scientific purposes (2010/63/UE), French Decret 2013–118.
Embryos were reared at 28.5°C and staged according to standard protocols [48]. The Tg(exorh:EGFP)ja1 [29], Tg(hsp70l:dnXla.Rbpj-MYC)vu21 [49], Tg(aanat2:GFP)y8 [50], Tg(BMPRE-AAV.Mlp:EGFP)mw29 and Tg(BMPRE-AAV.Mlp:d2EGFP)mw30 [32], Tg(-1.6flh:GAP-EGFP)u711 [51], and Tg(hsp70:bmp2b)fr13 [52] have been described previously.
Conditions of heat shock were as follows: Tg(hsp70:bmp2b)fr13 30 minutes at 37°C and Tg(hsp70l:dnXla.Rbpj-MYC)vu21 30 minutes at 39.5°C.
Genotyping of Tg(hsp70l:dnXla.Rbpj-MYC)vu21 was performed either using immunohistochemistry against the Myc epitope tag or via a nested PCR with the following couples of oligos:
5′-GCCACTTTTGTCCCTGATGC-3′
5′-CTTTTTACATGTGGACTGCC-3′
and then,
5′-CCTTCCAGGTTCAGCTGCTG-3′
5′-CGGGCATTTACTTTATGTTGC-3′.
Genotyping of Tg(hsp70:bmp2b)fr13 was performed as described in [6].
To generate an in vivo marker of PT-expressing PhRs in the pineal gland, we amplified a 2.2-kb fragment of PT regulatory sequences immediately upstream of the ATG by PCR using the following oligos:
5′-CGACCTCGAGGTAGGCCTACATTAAGCGAT-3′
5′-GCGCGGATCCGATGATTCGGAATGATCTTC-3′.
The resulting fragment was subcloned into a pBS-I-SceI backbone upstream of the coding region of CFP. To generate the Tg(-2.2parietopsin:CFP) transgenic line, this construct was coinjected with I-SceI meganuclease into freshly fertilized embryos following previously described protocols [53]. The presence of successfully inserted transgenes was assessed using PCR with the following oligonucleotides:
5′-GGACACGCTGAACTTGTGG-3′
5′-GGTACTTGTTCAGATGGCTG-3′.
For experiments in which numbers of cells were assessed in fixed material, we assessed statistical significance using either a t test, a Mann Whitney test, or a Kruskal-Wallis test with Dunn’s post hoc comparisons (Fig 7). Statistical tests and number of embryos used are stated in each figure and/or figure legend.
|
10.1371/journal.pmed.1002185 | Lifestyle Advice Combined with Personalized Estimates of Genetic or Phenotypic Risk of Type 2 Diabetes, and Objectively Measured Physical Activity: A Randomized Controlled Trial | Information about genetic and phenotypic risk of type 2 diabetes is now widely available and is being incorporated into disease prevention programs. Whether such information motivates behavior change or has adverse effects is uncertain. We examined the effect of communicating an estimate of genetic or phenotypic risk of type 2 diabetes in a parallel group, open, randomized controlled trial.
We recruited 569 healthy middle-aged adults from the Fenland Study, an ongoing population-based, observational study in the east of England (Cambridgeshire, UK). We used a computer-generated random list to assign participants in blocks of six to receive either standard lifestyle advice alone (control group, n = 190) or in combination with a genetic (n = 189) or a phenotypic (n = 190) risk estimate for type 2 diabetes (intervention groups). After 8 wk, we measured the primary outcome, objectively measured physical activity (kJ/kg/day), and also measured several secondary outcomes (including self-reported diet, self-reported weight, worry, anxiety, and perceived risk). The study was powered to detect a between-group difference of 4.1 kJ/kg/d at follow-up. 557 (98%) participants completed the trial. There were no significant intervention effects on physical activity (difference in adjusted mean change from baseline: genetic risk group versus control group 0.85 kJ/kg/d (95% CI −2.07 to 3.77, p = 0.57); phenotypic risk group versus control group 1.32 (95% CI −1.61 to 4.25, p = 0.38); and genetic risk group versus phenotypic risk group −0.47 (95% CI −3.40 to 2.46, p = 0.75). No significant differences in self-reported diet, self-reported weight, worry, and anxiety were observed between trial groups. Estimates of perceived risk were significantly more accurate among those who received risk information than among those who did not. Key limitations include the recruitment of a sample that may not be representative of the UK population, use of self-reported secondary outcome measures, and a short follow-up period.
In this study, we did not observe short-term changes in behavior associated with the communication of an estimate of genetic or phenotypic risk of type 2 diabetes. We also did not observe changes in worry or anxiety in the study population. Additional research is needed to investigate the conditions under which risk information might enhance preventive strategies. (Current Controlled Trials ISRCTN09650496; Date applied: April 4, 2011; Date assigned: June 10, 2011).
The trial is registered with Current Controlled Trials, ISRCTN09650496.
| Despite questions regarding their clinical validity and utility, genetic tests aimed at predicting risk of type 2 diabetes are now widely available.
Some researchers and direct-to-consumer genetic testing companies are optimistic that personalized genetic information will encourage people at risk to adopt healthier behavior than standard advice about lifestyle, but scientific evidence is needed.
The researchers calculated the risk of developing type 2 diabetes for 569 healthy middle-aged adults.
The risk estimates were either based on a participant’s genetic makeup, as judged by presence of genetic information known to be associated with type 2 diabetes, or on phenotypic characteristics, such as sex, age, body mass index, etc.
Participants were randomly assigned to three groups to receive standard lifestyle advice, alone or in combination with their genetic risk estimate or phenotypic risk estimate. After 8 wk, participants’ physical activity, self-reported diet and weight, anxiety, worry, and beliefs about their risk were measured.
The researchers found that receipt of a genetic or phenotypic risk estimate did not affect participants’ physical activity or other relevant behaviors, in comparison with those receiving standard lifestyle advice. However, perception of risk became more accurate.
In this study, provision of personalized information about genetic or phenotypic risk of type 2 diabetes did not affect behavior when compared with standard lifestyle advice.
Provision of personalized information about risk of type 2 diabetes did not seem to cause anxiety.
More research is needed to understand the circumstances in which genetic risk information might motivate behavior change.
| The prevalence of type 2 diabetes is increasing worldwide, and primary prevention of the disease is a global priority [1]. Evidence from randomized controlled trials shows that positive changes in health behavior can significantly reduce the incidence of type 2 diabetes among those considered high-risk [2,3]. However, translating these findings into preventive strategies has proven difficult, as it requires motivation of individuals to adopt and maintain changes in physical activity and diet [4].
Risk of type 2 diabetes is also influenced by genetics. Despite questions regarding the clinical validity and utility of recently developed predictive genetic tests [5,6], some researchers and direct-to-consumer genetic testing companies are optimistic that the provision of genetic risk information for type 2 diabetes will motivate behavior change more than widely available phenotypic risk information [7–9]. This hypothesis has support from health behavior theory [10,11]. However, there is concern about the potential negative psychological impact of widely available genetic risk information including fatalism, anxiety, and false reassurance [12–14]. Furthermore, there is evidence that risk perceptions and communication of biomarker information have limited influence on behavior [15,16].
There are few clinical studies concerning the impact of genetic risk information [17]. The majority have been nonrandomized or underpowered. Furthermore, interpretation has been limited by risk of bias and additional differences between study groups than merely the provision of DNA-based disease risk information. Two recent trials report no behavioral impact of information about the genetic risk of type 2 diabetes [18,19]. However, neither included precise measures of behavior. In addition, Grant et al. recruited a small sample of individuals willing to participate in an intensive diabetes prevention program [19], and Voils et al. did not include a control group receiving no phenotypic risk information [18].
We examined the effect of communicating genetic or phenotypic risk of type 2 diabetes in combination with standard lifestyle advice on objectively-measured physical activity, self-reported diet, self-reported weight, anxiety, and several cognitive and emotional theory-based antecedents of behavior change in a sample of healthy middle-aged adults.
Details regarding the trial methods have been reported previously [20]. We obtained ethical approval from the Cambridgeshire 1 Research Ethics Committee (No. 10/H0304/78). Each participant provided written informed consent. The trial is registered with Current Controlled Trials (ISRCTN09650496; Date applied: April 4, 2011; Date assigned: June 10, 2011).
We recruited participants from the Fenland Study, an ongoing population-based observational study investigating the influence of lifestyle and genetic factors on the development of diabetes, obesity, and related metabolic disorders [21]. Individuals born between 1950 and 1975 registered with participating general practices in Cambridgeshire, UK were invited to take part. General practitioners excluded those with a diagnosis of diabetes, a terminal illness with a prognosis of less than one year, a psychotic illness, being pregnant or lactating, or being unable to walk unaided. Fenland Study participants undergo a health assessment, and blood samples are collected for the genotyping of single nucleotide polymorphisms (SNPs) associated with type 2 diabetes. At the end of the assessment, participants are fitted with a combined heart rate monitor and accelerometer (Actiheart) [22], which they are instructed to wear continuously for six days and nights to measure physical activity.
We sent invitations to take part in the Diabetes Risk Communication Trial (DRCT) to Fenland Study participants who 1) had agreed to be contacted regarding future studies, 2) had sufficient data to calculate their genetic and phenotypic risk of type 2 diabetes, 3) wore the combined heart rate monitor and accelerometer for three or more full days without experiencing a severe skin reaction, and 4) provided at least 36 h of complete physical activity data. Upon response, we excluded those who reported being diagnosed with diabetes or actively participating in another study.
We randomly allocated eligible participants who completed a baseline questionnaire to one of three groups that received either standard lifestyle advice alone (control group) or in combination with a genetic or a phenotypic risk estimate for type 2 diabetes (intervention groups). A statistician without knowledge of participant characteristics created a computer-generated list comprised of blocks of six that contained two of each of the three study groups per block in a random order. This was incorporated into an automated randomization computer program. Allocation was concealed from the researchers and participants until the interventions were assigned. Researchers assessing the primary outcome remained blinded to the allocation of participants throughout the study.
All participants received standard written lifestyle advice, which included a brief description of type 2 diabetes and an explanation of the risk factors, symptoms, diagnosis, treatment, and consequences of the disease. We informed participants that the disease is preventable and encouraged them to maintain a healthy weight and to adhere to United Kingdom governmental guidelines for physical activity and diet [23,24].
The interventions were designed to incorporate evidence regarding the most effective methods for communicating disease risk estimates [25]. As it remains unclear whether an individual’s understanding of risk is more accurate after the provision of a numerical risk estimate or a verbal risk estimate [26,27], both the genetic and phenotypic risk estimates included estimates of the participant’s lifetime risk of developing type 2 diabetes expressed as a percentage and verbal estimation of risk (i.e., “below average,” “average,” or “above average”). Moreover, research suggests that comparative risk estimates may have a greater influence on behavior than absolute risk estimates [28,29], and that visual representations of risk elicit greater recall and understanding of risk [26,30,31]. Thus, estimates were framed in comparison to the average risk within each participant’s age and sex-specific group, and participants were told what percentage of the study sample had a risk estimate higher, lower, and equal to their own. Each piece of information was represented using a visual scales [27].
Methods for calculating the genetic and phenotypic risk estimates have been described in detail previously [20]. We calculated genetic risk using methods similar to those outlined by several direct-to-consumer genetic testing companies [32–34]. The 23 SNPs utilized in the calculation were identified through adequately powered genome-wide association studies, had associations with type 2 diabetes that reached the genome-wide significance level (p-values for associations less than 5x10−8), and had associations that were replicated in at least one independently published study. We took the odds ratio for each SNP from replication samples and the allele frequency from the HapMap population. We calculated phenotypic risk using the Cambridge Diabetes Risk Score [35,36]. Age, sex, smoking status, family history of diabetes, and prescription of steroid or antihypertensive medication were assessed via questionnaire. We measured height and weight using standardized procedures and calculated body mass index as weight (in kg) divided by the square of height (in m). All data used in the estimation of risk were collected during the participant’s Fenland Study health assessment.
The primary outcome was physical activity, defined as physical activity energy expenditure (kJ/kg/d), measured objectively using the Actiheart continuously for six days and night [22]. We used a submaximal exercise test for individual calibration of heart rate response [37] and a branched equation model to estimate physical activity energy expenditure from acceleration and heart rate [38]. This approach has high validity for estimating the intensity of physical activity [39,40] and overcomes some of the key limitations associated with either accelerometers or heart rate monitors alone [22].
Baseline physical activity was measured in the Fenland Study (median 1.76 years prior to enrollment in the DRCT), and follow-up occurred 8 wk postintervention. This relatively short follow-up period was chosen on the basis that it is unlikely that a long-term effect would exist in the absence of an impact in the short term. All prespecified secondary outcomes were measured via questionnaire and included self-reported diet, self-reported weight, self-rated health, worry (measured at baseline and follow-up), anxiety, behavioral intention, perceived risk, self-efficacy, response efficacy, perceived severity, and diabetes risk representations (measured at baseline, immediately post receipt of the intervention, and follow-up; more details are available in S1 Appendix).
All analyses were performed on an intention-to-treat basis (i.e., analysis of data according to randomized study group, regardless of whether or not the intervention was received) using STATA software [41]. We used univariate descriptive statistics (means, standard deviations [SDs], numbers, and percentages) to summarise the characteristics of the study sample at baseline. We used analysis of covariance to assess differences between groups in physical activity at follow-up, adjusted for baseline. Prespecified exploratory analyses were conducted to examine whether sex, age, body mass index, time since the Fenland Study, and baseline measurements of the trial outcomes moderated the intervention effects on physical activity. A further subgroup analysis explored whether a high or low risk estimate moderated the effect of the type of risk estimate (i.e., genetic or phenotypic) on physical activity. The study protocol and statistical analysis plan specified that the analyses should include only participants with complete postintervention or follow-up data (i.e., a complete case analysis). We included participants with missing baseline data in the analyses using the missing-indicator method [42]. Similar regression procedures were used to examine differences in all secondary outcomes. The acceptability of the interventions was assessed by summarizing recruitment rates, loss to follow-up, and reasons for loss to follow-up. Additionally, differences in responses to questions regarding the perceived accuracy of the risk estimates, as well as the retention and discussion of the risk estimates were examined.
Estimates used in the sample size calculation were taken from the Feedback, Awareness and Behavior (FAB) study, which had a similar sample population and the same primary outcome as proposed here (i.e., physical activity energy expenditure) [43]. The mean (SD) physical activity energy expenditure at follow-up in the FAB study was 46.2 (15.4) kJ/kg/d, and the correlation between baseline and follow-up was high (0.69). After making a Bonferroni adjustment for multiple comparisons in a three-group trial, we determined that in order to detect a between-group difference of 4.1 kJ/kg/d at follow-up (which equates to approximately 20 to 25 min of walking per day), with 98.3% significance and 80% power, approximately 465 participants would need to complete the trial [44,45]. We conservatively allowed for an attrition rate of 20% and targeted the recruitment of 580 participants.
Between February 11, 2011 and September 5, 2011, we sent invitations to take part in the DRCT to 1150 Fenland Study volunteers and 635 (55%) replied positively and were assessed for eligibility. Between March 8, 2011 and September 14, 2011, 569 (49%) participants were randomized (Fig 1). Reasons for exclusion included not responding after an initial positive reply (68%), responding after enrollment closed (12%), being unavailable prior to the expected trial completion date (8%), reporting a rash while wearing the Actiheart during the Fenland Study (1%), participating in another study (6%), and being diagnosed with diabetes (5%). Those who did not reply and those who were excluded from participation did not differ from those randomized according to sex, body mass index, glycated haemoglobin (HbA1c) or phenotypic risk. However, they were slightly younger than participants at the time of their Fenland Study health assessment (mean [SD] age 45.0 [6.9] y versus 47.2 [7.4] y) and had a higher genetic lifetime risk estimate (mean [SD] of 18.8% [8.2%] versus 17.8% [8.1%]).
After randomization, 12 (2.1%) participants were lost to follow-up, and we excluded 7 (1.2%) from the primary analysis because they had insufficient physical activity data (the monitor did not record more than three days of data) (Fig 1). The complete case sample comprised 550 participants: 184 received standard lifestyle advice alone, 184 received standard lifestyle advice and a genetic risk estimate, and 182 received standard lifestyle advice and a phenotypic risk estimate.
There were no significant between-group differences in objectively measured physical activity at the 8-wk follow-up (Fig 2). Prespecified exploratory analyses showed that the interventions had no effect on physical activity within subgroups defined by age, body mass index, physical activity, self-reported diet, self-reported weight, self-rated health, behavioral intention, perceived risk, anxiety, worry, time since participation in the Fenland Study, or receipt of a high or low risk estimate (Table A in S1 Appendix). However, when compared to the control group, the genetic risk estimate was associated with a greater increase in physical activity among women than among men (women: β = 4.29, 95% CI = 0.27 to 8.33, p = 0.037; men: β = −2.69, 95% CI = −6.92 to 1.55, p = 0.213). The phenotypic risk estimate did not have a differential effect by sex (women: β = 1.70, 95% CI = −2.45 to 5.86, p = 0.421; men: β = 1.33, 95% CI = −2.75 to 5.41, p = 0.523).
No significant differences were observed between trial groups in self-reported diet, self-reported weight, self-rated health, and worry at follow-up, nor were there any significant differences in behavioral intention and anxiety immediately postintervention or at follow-up (Tables 2 and 3). Participants who received a risk estimate had a lower perceived risk immediately postintervention than those who did not receive an estimate. These effects were attenuated at follow-up, but remained statistically significant, and did not differ by the type of risk estimate received (Tables 2 and 3). Additionally, immediately postintervention, both diet response efficacy and illness understanding were significantly lower in the phenotypic risk group compared to the control group, and perceived severity was significantly lower in the genetic risk group compared to the control group (Tables 2 and 3).
Among those who received a risk estimate, the majority (93.0%) reported that they believed their risk estimate to be either fairly or very accurate. Most participants (90.5%) stated that they had kept their risk estimate, and many (63.7%) reported discussing it with others (for example family members, friends, or health professionals).
In a sample of healthy, middle-aged men and women who were given information about type 2 diabetes and standard lifestyle advice, there was no effect of communicating a genetic or phenotypic estimate of the risk of developing type 2 diabetes on objectively measured physical activity. We did not observe significant intervention effects on self-reported diet and weight, self-rated health, behavioral intentions, anxiety, or worry. This is an important observation, given the expectations that such communications might facilitate behavior change and the concerns about the potential adverse psychological consequences of predictive genetic testing. We also did not observe significant intervention effects on a range of other cognitive and emotional theory-based antecedents to behavior change. We examined several potential moderators, and only sex was found to interact with the intervention effect on physical activity, raising the possibility that genetic risk information may be more influential among women than among men. More research is needed to explore whether women and men respond differently to genetic risk information.
Risk information was received and understood, and had a sustained effect on participants’ perceptions of their risk. However, the volunteers tended to overestimate their risk at baseline and may therefore have been somewhat reassured by the information that they received, albeit not to the extent that they adopted unhealthy behaviors [46]. Nevertheless, we cannot exclude the possibility that provision of risk estimates that exceed participants’ perceived risk might influence behavior, although previous trials have not reported effects among high-risk subgroups [18,19]. Furthermore, it is possible that information concerning the genetic risk of diseases other than diabetes, such as various cancers or chronic neurodegenerative diseases, might elicit a greater response, although this has not been the finding of published trials [47].
This trial provides much needed robust evidence on the behavioral impact of communicating genetic risk information. A systematic review identified only two clinical trials that assessed physical activity and diet as outcomes [47]. The authors concluded that given the limited number of low quality studies, strong conclusions could not be drawn and larger, higher quality studies were needed. The findings of this study suggest that the provision of a genetic risk information, which reduced perceived risk in the majority of participants, did not motivate healthy changes in behavior over and above phenotypic risk information or standard lifestyle advice alone. Findings are consistent with those of a cohort study of the impact of direct-to-consumer genome-wide testing [48] and recent trials of type 2 diabetes genetic risk information [18,19] and add to existing evidence showing that those who undergo testing seldom experience psychological harm [13,49]. While risk information appears not to motivate changes in health behaviors, there is some evidence that it may influence decisions about use of medication [50,51].
Previous research indicates that the effect of communicating genetic risk information on perceived risk, a central construct in many health behavior theories, is unclear. Most studies report that perceived risk decreases after receipt of risk information, usually towards a more accurate perception of risk [52]. We found that the provision of a genetic risk estimate was associated with lower and more accurate perceived risk both immediately and after eight weeks. However, there were no differences in risk perception between those receiving genetic information and a phenotypic risk estimate that can be calculated using routinely collected clinical data. Small but statistically significant immediate effects of risk information on diet response efficacy, illness understanding, and perceived severity may have arisen through multiple testing.
The strengths of this trial include the recruitment of a relatively large sample, a randomized design with sufficient power to assess clinically important impacts on objectively measured behavior, and a high rate of study completion (97%). We presented risk information in a manner similar to that of several direct-to-consumer genetic testing companies. Potential limitations are that participants were recruited from one location in the UK, were well-educated, physically and psychologically healthy, and exhibited limited socioeconomic and ethnic diversity. Consequently, the results might not generalize to other settings or groups. Other limitations include the use of a baseline measure of physical activity that occurred prior to enrollment in the study, self-report questionnaires to measure all secondary outcomes, and a relatively short time to follow-up. However, it is widely assumed that the continuous measure of physical activity for three or more days accurately captures habitual physical activity levels. To the extent that this is true and baseline differences were equally distributed across study groups, the time between the baseline measure of physical activity and enrolment should not have influenced the results. Furthermore, it is unlikely that communication of risk information would have a long-term effect in the absence of an impact in the short term. We did not attempt to assess the clinical validity and utility of the genetic or phenotypic risk estimates as the objective of the trial was to determine the effect of the reported estimates on behavior, regardless of the accuracy of the estimate. However, the risk estimates have been validated in other studies. Although nearly all participants who received a risk estimate believed it to be accurate, the extent to which they were aware of the uncertain clinical validity and utility of predictive genetic tests may have influenced the results of the trial.
In conclusion, we found that communicating an estimate of the risk of type 2 diabetes, either based on genotype or phenotype, did not motivate changes in behavior in the short term, but neither did it cause an increase in worry or anxiety. These findings are consistent with systematic review evidence and should inform the ongoing debate regarding the regulatory response to the proliferation of direct-to-consumer genetic testing companies. Additional research is needed to investigate the conditions under which risk information might enhance preventive strategies. Approaches targeting individual behavior change, such as communicating genetic risk, are unlikely to be successful in isolation in an environment in which there are many impediments to being physically active and eating a healthy diet. The results of the current study thus provide further evidence for a shift in focus for promoting healthy changes in habitual, environmentally patterned behaviors, such as physical activity and diet, away from interventions solely based on provision of information and advice to individuals towards interventions that target the wider collective determinants of disease [53].
|
10.1371/journal.pcbi.1006888 | Predicting synthetic lethal interactions using conserved patterns in protein interaction networks | In response to a need for improved treatments, a number of promising novel targeted cancer therapies are being developed that exploit human synthetic lethal interactions. This is facilitating personalised medicine strategies in cancers where specific tumour suppressors have become inactivated. Mainly due to the constraints of the experimental procedures, relatively few human synthetic lethal interactions have been identified. Here we describe SLant (Synthetic Lethal analysis via Network topology), a computational systems approach to predicting human synthetic lethal interactions that works by identifying and exploiting conserved patterns in protein interaction network topology both within and across species. SLant out-performs previous attempts to classify human SSL interactions and experimental validation of the models predictions suggests it may provide useful guidance for future SSL screenings and ultimately aid targeted cancer therapy development.
| Our new algorithm SLant, uses artificial intelligence to help target future cancer drug research. In healthy cells tens of thousands of proteins work together forming large interaction networks. However, in cancerous cells genetic damage means that many of these proteins are disabled. Basic functions like DNA repair and signaling no longer work properly, and the cell replicates without proper control. Recent experience with breast cancer shows that gentler, more personalised therapies can be achieved by finding pairs of proteins which are ‘synthetically lethal’. The term means that the cell can cope if either one of the proteins does not work, but will die if neither of the proteins is functioning. Many synthetic lethal interactions are known, but there are many millions of potential pairs and finding new ones experimentally is difficult and time-consuming. SLant uses most of the experimental data that we have to identify the patterns in the protein interaction network associated with being part of a synthetic lethal interaction. By searching the network for proteins pairs that match these patterns, it can effectively predict new synthetic lethal pairs. The predictions were then checked against the rest of the experimental data. Our predictions are publicly available through the Slorth database.
| Despite sustained global efforts to develop effective therapies, cancer is now responsible for more than 15% of the world’s annual deaths. There are over 12 million newly diagnosed cases per annum and this figure continues to grow [1]. Standard chemotherapy involves non-selective, cytotoxic agents that often have limited effectiveness and strong side-effects. Consequently, the current focus in oncology drug discovery has moved towards identifying targeted therapies that promise both improved efficacy and therapeutic selectivity [2].
The development of multi-platform genomic technologies has enabled the identification of many of the genes that drive cancer [3]. These cancer driver genes can be broadly classified either as oncogenes or tumour suppressors. The protein product of an oncogene shows an increase in activity, or a change or gain of function when mutated, whereas mutations or epigenetic silencing in tumor suppressors result in an inactivation or loss of function (LOF) of the protein product [4].
Targeted therapies that act on oncogenes often work by directly inhibiting the activated protein product. This strategy has been particularly successful for targeting nuclear receptor proteins or those that contain protein kinase domains. [5–7]. Unfortunately, it is not usually feasible to repair tumour suppressor genes or their protein products, particularly if they are inactivated by a truncation [8]. Instead an emerging strategy is to target tumour suppressors indirectly by exploiting synthetic lethal interactions.
Synthetic lethality (SSL) is a phenomenon whereby individual genes in a pair can be knocked-out without affecting cell viability, whilst disruptions in both genes concurrently cause cell death [9]. Synthetic sensitive and synthetic sickness interactions are extensions of this concept where concurrent genetic interactions impair cellular fitness without necessarily killing the cell. Conversely, synthetic dosage lethality (SDL) interactions occur when over-expression of one gene, in combination with loss of function in another gene results in cell death. SSL and SDL interactions are both examples of negative genetic interactions. Negative genetic interactions are events where a deviation from the expected phenotype is observed when genetic mutations occur in more than one gene [10].
To exploit SSL interactions therapeutically one gene, the tumour suppressor, is genetically inactivated by mutation while the protein product of the other is targeted and inactivated pharmacologically [11]. Synthetic dosage lethal interactions can be used for targeting cancer cells with over-expressed, undruggable oncogenes [11]. SDL causes cell death as a result of one gene being genetically activated (GOF, the oncogene) and another being inactivated (LOF, the drug target).
PARP inhibitors are the most developed therapies that exploit SSL interactions. The PARP inhibitor Olaparib, has been approved for the treatment of patients with recurrent, platinum-sensitive, high-grade serous ovarian cancer with BRCA1 or BRCA2 mutations [12, 13]. PARP1 (poly(ADP-ribose) polymerase) is an important component in DNA single strand break repair and has been shown to share a synthetic lethal relationship with both BRCA1 and BRCA2 [14, 15], which are themselves both key in DNA double strand break repair. Complete loss of function of the protein product of either BRCA gene leaves cells extremely sensitive to PARP inhibitors presenting this therapeutic opportunity [16, 17].
Other studies have highlighted a range of SSL interactions that may provide suitable targets for therapy [18–20]. For example, PI5P4K kinases are essential in the absence of p53 [21], inhibition of ENO2 inhibits viability in ENO1 deficient glioblastoma cells [22] and APE1 inhibitors in PTEN deficient cells results in the induction of apoptosis [23].
Currently, mainly due to experimental limitations [24] exhaustive experimental identification of human SSL interactions is not tenable. However there are many studies focused on screening for genetic interactions in model organisms [25]. Unfortunately, genetic interactions are not highly conserved between lower eukaryotes and their human orthologue equivalents [26]. Instead, in order to identify novel human SSL interactions, we are left to infer and predict these pairs indirectly from existing human and model organism data through the use of models and other computational techniques [27].
Several classifiers have been developed to predict genetic interactions within model organisms. Wong et al. [28] predicted genetic interactions in Saccharomyces cerevisiae using decision tree classifiers with multiple data types and network topology. Paladugu et al. [29] focused on S. cerevisiae data; by extracting multiple features from protein interaction networks they achieved sensitivity and specificity exceeding 85% using support vector machine (SVM) classifiers. Later, Chipman et al. [30] employed random walks and decision tree classifiers on protein interaction and gene ontology (GO) data to classify both S. cerevisiae and C. elegans negative genetic interactions.
Several classifiers have been developed to predict genetic interactions between species. Zhong and Sternberg [31] classified Caenorhabditis elegans negative genetic interactions based on orthologous gene pairs in S. cerevisiae and Drosophila melanogaster. Jacunski et al. [32] developed SINaTRA (Species-INdependent TRAnslation) to classify S. cerevisiae SSL pairs based on Schizosaccharomyces pombe training data and vice versa, using features extracted from physical interaction data. The model trained on S. cerevisiae data was applied to predict 1,309 human SSL pairs with a reported false positive rate of 0.36. Similarly Wu et al [33] developed MetaSL, an ensemble machine learning mode which applied eight different classifiers on S. cerevisiae data and applied it to predict human SSL pairs.
Using an alternative approach, the DAISY workflow predicted human SSL interactions directly from human cancer and cell–line data [34]. The authors used somatic copy number variation and mutation profiles to achieve a ROC AUC score of 0.779 demonstrating a strong propensity (p-value < 1e-4) for predicting SSL pairs in H. sapiens.
There are a number of additional recent studies that use biological networks to predict genetic interactions. Mashup [35] reported an average area under the precision curve (AUPR) of 0.59 for SSL and 0.51 for SDL pair prediction in a real human dataset. Others have utilised gene ontology terms to predict SSLs. These include Ontotype [36], where the authors predict the growth outcome on double knock-out of gene pairs. Their prediction set of gene pairs related to DNA repair and nuclear lumen correlated with Costanzo et al’s [37] validated SSL dataset with a coefficient of r = 0.61. The authors of DCell [38] constructed a visible neural network embedded in the hierarchical structure of 2526 subsystems describing the eukaryotic cell and used this to predict negative genetic interactions in S. cerevisiae.
In this study we introduce SLant (Synthetic Lethal analysis via Network topology), a random forest classifier trained on features extracted from the protein-protein interaction (PPI) networks of five species. These features comprise both node-wise distance and pairwise topological PPI network parameters and gene ontology data. Using SLant we provide in-species, cross-species and consensus classification for synthetic lethal pairs in all five organisms including human. We subsequently experimentally validated three of the predicted human SSLs in a human cell-line. Finally we identify a large cohort of candidate human synthetic lethal pairs which are available with the consensus predictions for all the model organisms in the Slorth database (http://slorth.biochem.sussex.ac.uk).
A genome-wide protein-protein interaction (PPI) network was constructed for Homo sapiens and each of our model organisms (S. cerevisiae, D. melanogaster, C. elegans, and S. pombe) using PPI data from the STRING database [39]. In this network, each node represents a protein and each edge represents a physical interaction between two proteins. For each pair of proteins 12 node-wise and 7 pairwise features were extracted from the PPI network using the R igraph library [40]. Each protein in the network was labelled with its respective Ensembl gene identifier so that this physical interaction data could be matched with gene interaction data. For each gene pair 3 additional GO term related features were generated using Gene ontology (GO) data [41].
For each PPI network, pairs of proteins whose respective genes were identified as having a negative genetic interaction in BioGRID [42] were labelled as having an SSL interaction (Fig 1). Equal numbers of SSL and non-SSL gene pairs were selected independently for the training sets for each species (see methods). Similarly we created training sets for SDL and non-SDL gene pairs in H. sapiens and S. cerevisiae, the only two species where there is enough data for prediction purposes.
The features used for classification in the SLant algorithm were broadly divided into node-wise, pairwise or GO-term related categories. Node-wise features were derived from an individual node's network parameter, such as degree or centrality. These node-wise features were converted to pairwise features by taking the average distance for that feature between the nodes in each pair. Pairwise features were defined as those that apply to a pair of nodes such as shortest path or cohesion. The spin glass random walk features discussed below were included in our pairwise category. GO related features were derived from shared annotations between pairs of genes [41] (for a full list of features see Table 1).
Fig 2 shows the distribution of these features in SSL and non-SSL gene pairs in humans. In general pairwise parameters showed a greater variance between SSL and non-SSL classes than our node-wise parameters suggesting they may prove better predictors in our models. Of these pairwise parameters the most notable differences were observed in the parameters labelled: cohesion—the minimum number of nodes that would have to be removed to result in two separate sub-graphs separating the source and target nodes, shortest path—the minimum number of nodes that must be traversed in a path between the source and target gene, and mutual neighbours—the number of nodes that are shared as neighbours between the source and target gene.
The higher values exhibited by gene pairs in the SSL class for the cohesion feature (paired t-test; p = 2.2e-16 in H. sapiens) suggest that SSL pairs are generally more densely connected in a physical interaction graph than non-SSL pairs (S1A Fig).
We also note that the shortest path between gene pairs is shorter on average for SSL gene pairs compared to non-SSL gene pairs (paired t-test; p = 4.589e-11 in H. sapiens) (S1B Fig) and, related to the shortest path parameter, SSL genes more often share a large number of mutual neighbours (paired t-test; p = 4.058e-11 in H. sapiens) (S1C Fig).
In terms of node-wise features it is of some interest to note that the difference between neighbourhood sizes of two genes in an SSL pair often differ more than those in a non-SSL pair.
In an attempt to ascertain whether synthetic lethal interactions occurred within or between local clusters of genes in our physical interaction network we applied a spin-glass random walk to assign genes to 20 distinct communities separated by choke points across the graph (Fig 3A). Analysis showed that the majority of SSL interactions occurred between these communities rather than within (Fig 3B). In addition pairwise topological analysis suggests that SSL pairs of genes have shorter paths between them than non-SSL pairs and a higher occurrence of adjacency. Together these analyses suggest that SSL pairs are often at the peripheries of these communities, connecting their respective clusters.
Based on these observations we were able to create two additional features which provide further predictive power for classifying SSL pairs; whether nodes shared a community and whether the pair connected two communities.
The count of shared GO terms, that is the number of GO annotations that two genes in a pair share with each other, also varies between SSL and non-SSL observations. SSL pairs generally share, on average, less biological process GO annotations (S1 Table) than non-SSL pairs (p < 2.2e-16 in H. sapiens) and proportionately more molecular function and cellular component GO annotations (p < 2.2e-16 in H. sapiens for both biological process and cellular compartment terms). This supports the view that that SSL protein product pairs are often found in similar but distinct pathways rather than within a single pathway [43]. Damaging two complementary functional pathways is likely cause more stress to the cell than damaging one pathway twice and leaving the complementary pathway functional.
Although the GO annotation based features above provide predictive power in our models as discussed below, due to the hierarchical nature of GO annotation, comparing the absolute count of shared GO terms does present some issues. As such GoSemSim [44] was used to further measure the semantic similarity between SSL and non-SSL pairs. We found that in H. sapiens SSL pairs showed a significantly higher semantic similarity score (mean = 0.65) that non-SSL pairs (mean = 0.57) (Welch two sample t-test p = 4.6e-07).
Analysis of GO terms present in paired SSL genes found that the most commonly shared molecular functional GO annotation related to protein binding (S2 Fig). Other molecular function GO annotations commonly found associated between SSL pairs include protein complex binding, GTP binding, DNA binding and GTPase activity. At the level of biological process GO annotation for SSL gene pairs we also noted associations with terms related to positive regulation of cell proliferation and negative regulation of apoptotic process as well as those labelled with positive regulation of gene expression and positive regulation of transcription from RNA polymerase II promoter.
In an attempt to further quantify the GO annotation driving the variation between genes found in SSL pairs and those not found in SSL pairs we employed a GO enrichment analysis using the on-line GOrrila tool [45]. We found significant enrichment in a number of GO annotations including negative regulation of cell differentiation (p = 9.15e-3), positive regulation of transcription by RNA polymerase II (p = 9.53e-3) and regulation of Notch signaling pathway (p = 8.85e-3) in the biological process ontology but no further enrichment in the molecular function or cellular compartments ontologies. All p-values have been are corrected for false positives using the Benjamini Hochberg method.
Comprehensive studies of S. cerevisiae genetic interactions by Costanzo et al [37, 46] have found that essential genes that share an edge on the PPI network are enriched for genetic interactions and that is consistent with previous observations [43]. As our classifiers in part use the distance of gene pairs as a predictive feature we performed analysis to ensure our predictions were not simply picking out gene pairs enriched for essential genes.
We first noted that the range of shortest path values between SSL pairs on the protein-protein interaction (PPI) network runs from 1 to 7 with a mean of 2.43 and a standard deviation of 0.78 affirming that our training set features many SSL pairs that are not adjacent in the PPI network.
Using a set of essential human genes defined by Wang et al. [47], we found that 11% of the genes in our SSL training set were defined as essential, where as for non-SSL genes it only 0.7%. For human gene pairs ~1.7% of SSL pairs and ~1.4% of non-SSL pairs are comprised of two essential genes. We also found that 29% of SSL pairs and 22% of non-SSL pairs included at least one essential gene.
Upon comparison we found that ~22.5% of our SSL predictions included at least one essential gene and ~1.4% featured two essential genes, a ratio comparable with our training data. This suggests that our predictions are not further enriched for essential genes.
The distribution of network parameters across our four model organisms widely followed similar trends with our human feature set. Again the pairwise features for each organism appear to vary more between SSL and non-SSL classes than node-wise features. A few dissimilarities were noticeable, for example while SSL gene pairs tend to exhibit a higher levels of adhesion and cohesion in H. sapiens, S. cerevisiae (S3A Fig) and D. melanogaster (S3B Fig) the distribution for these features were notably inverted in C. elegans (S3C Fig) and S. pombe (S3D Fig) so that non-SSL pairs showed higher adhesion and cohesion than SSL pairs.
In this study we perform two classifications. First in-species classification, classifying and validating SSL gene pairs using training and test data from the same organism. Then cross-species classification where we use the models built using the training data for each organism to blindly predict SSL for each other species. Within each species, the feature data were normalised and segmented into training and test sets with 20% set aside for validation. We employed 5-fold cross validation to optimise the hyperparameters for each organism's random forest classifier and evaluated in-species classification performance (Table 2). In this study our random forest classifiers utilised just one hyper-parameter, mtry—the number of variables randomly sampled as candidates at each split for each tree. The best classifier for each species was then used to predict the SSL gene pairs in each of the other four species. Table 2 shows the ROC AUC scores for both the in-species and cross-species predictions for all of our models.
Although it is difficult to compare the performance of classifiers due to varied validation sets, the ROC AUC score of 0.965 for H. sapiens SSL gene pair classification achieved by the SLant classifier (using holdout validation data) appears to out-perform Daisy’s ROC AUC score of 0.779.
Our initial in-species classification of S. cerevisiae SSL resulted in relatively low performance (AUC 0.734) compared to other related studies. For example MetaSL, which used a much smaller data set of just 7,347 SSL pairs compared to Slant’s 395,199 pairs, achieved ROC AUC scores of up to 0.871 [33]. In order to mitigate any noise or error introduced in our large dataset we filtered out any SSL interactions reported in BioGRID supported by less than 3 supporting publications for S. cerevisiae and less than 2 papers for S. Pombe. Our training data ultimately used 17,568 out of a total 395,199 SSL pairs available for S. cerevisiae and 3,836 out of 35,391 SSL pairs for S. Pombe. These sample sizes should still be large enough to generalise well for out of sample predictions as well as preforming well in classification and validation. Filtering our yeast data improved our scores from AUC ROC 0.734 to AUC ROC 0.883 for S. cerevisiae and 0.728 to 0.889 for S. Pombe which suggests that by removing pairs that show fewer citations in the BioGRID data we are reducing variation in our training data introduced by false positives. This may be due to the relatively high false-positive rate found in large scale GI screenings, an observation supported by analysis performed by Campbell & Ryan et al. who estimated that large scale screenings can suffer a false positive rate of up to ~10% [51]. Using this value we can calculate that by removing GI pairs with less than 2 and 3 references respectively we may be reducing false positive rates from 1/10 to 1/100 in S. pombe and from 1/10 to 1/1000 in S. cerevisiae.
Cross-species predictions of SSLs were quite variable in performance. Models from both S. cerevisiae and D. melanogaster and C.elegans were successful in predicting human SSLs with AUC ROC scores of 0.713, 0.727 and 0.769 respectively.
Although the C. elegans classifier performed relatively poorly in our cross-species validation for H. sapiens classification, this variation may help improve the generalisation of our consensus model which is discussed below. To test this cross-species validation was performed without the worm model. The removal of worm data from the classifier resulted in a small but noticeable decrease in performance of the consensus classifier for humans (decreasing from ~0.985 to ~0.92).
The result here suggest that the PPI patterns between SSL genes are similar both within and between species and that network topology features used in our classifiers generalise well across organisms. We identified the most predictive features for each organism and found that the same features were most predictive in many of the species. The shared GO count features were important in all organisms except S. pombe and the pairwise features adhesion, cohesion, mutual neighbours and adjacency were important in all organisms except C. elegans. Two node-wise features, coreness and neighbourhood size are also listed as important features across 3 organisms (S2 Table).
As described below in methods each of these models use a balanced training set with a ratio of 1:1 interacting and non-interacting pairs, however in reality the ratio between interacting and non-interacting pairs is likely more in the order of 1:50 based on global yeast GI screens [37]. To ascertain that our class balance has not unduly biased our prediction in any way we re-ran our classifiers using a randomised training / validation set with approximately 1:10 and 1:50 class balance. We found that with a class balance of 1:10 our performance remained stable and with a class balance of 1:50 we found just a small drop in performance (human AUC ROC ~0.87 compared to the original ~0.965 and consensus AUC ROC ~0.90 compared to ~0.985).
It is known that our current PPI models are incomplete [52–54] and suffer from ascertainment bias. That is, some genes, and indeed some species, are better studied than others. To test our model’s robustness to the incomplete nature of the protein-protein interaction networks, we re-ran our classifiers holding out 10% and 20% of the nodes, at random, from original PPI data in H. sapiens. In the case of the 90% ‘complete’ PPI network the performance of our in-species model validation was not effected and our H. sapiens consensus showed just a small drop in performance (from AUC ROC ~0.985 to ~0.922). With a 80% ‘complete’ H. sapiens PPI network we saw another fairly small incremental drop in H. sapiens consensus performance (AUC ROC ~ 0.85) and a small drop in H. sapiens in-species performance (AUC ROC dropping from 0.965 to 0.911). This suggests both that an increasingly complete PPI network may incrementally improve our predictive performance and that the current models are fairly resilient to the incomplete nature of the PPI network.
In addition to the feature importance analysis performed in this study we also re-ran our classifiers holding out our 12 node-wise distance features, 6 pair-wise features and 3 GO-term related features in turn. We found that the model holding out pair-wise features saw the largest drop in performance in consensus with the H. sapiens consensus ROC AUC dropping from ~0.985 to ~0.730 and the in-species H.sapiens ROC AUC dropping from ~0.965 to ~0.82. In comparison to our models holding out node-wise features saw a more notable drop in the in-species performance (H.sapiens consensus ROC AUC dropping from ~0.985 to ~0.85 and in-species H.sapiens from ~0.965 to ~0.823). Similarly holding out our GO term features resulted in a decrease in predictive performance (H.sapiens consensus ROC AUC dropping from ~0.985 to ~0.882 and in-species H.sapiens from ~0.965 to ~0.890).
As discussed by Parks et al. [55] computational prediction methods that utilise gene pair observations, such as the models in this study, can be subject to positive bias in validation. They discovered that model validation performed significantly better when genes that made up the pairs in the test set were also featured in the training set compared to those models where they were not.
In order to evaluate how SLant’s validation was effected by pair-input bias we generated a test set from our raw feature data in which none of the genes featured in the test pairs were present in any of the pairs featured in the training set. We refer to these as segregated datasets.
To make sure we could make a fair comparison we generated a further control training and test set by randomly sampling the pairs created above from both segregated data sets. This ensured that the pair count and the pairs themselves remained the same while gene components could be shared between our control training and test sets.
Running our models again using these segregated training and test data we achieved a AUC ROC of 0.789 for predicting human SSL pairs, compared to 0.845 for our control datasets and 0.965 for our full training and test sets. This suggests that while our predictions may be somewhat biased towards genes that are featured in the training data our models also appear to predict SSL pairs comprised of genes that are not in our training data and, more importantly, potentially genes that have not previously been associated with SSL interactions.
To further expand our model we took a consensus from the cross-species predictions for each organism. This consensus was calculated by running a second classifier, a boosted general linear model (GLM) that was trained on the previous cross-species classifier output. This output took the form of confidence scores. For example, for any particular pair of human genes the confidence scores given to that pair by every cross-species classifier were used as features. The probability outputted by this final classifier is referred to as the consensus score.
To allow for validation this consensus dataset was segmented into a training and test set (both 0.5 the size of the original due to the smaller overall size). The ROC AUC for our consensus prediction validation was also plotted and achieved a score of up to 0.985 when predicting H. sapiens SSL pairs, a further improvement on our in-species human validation ROC AUC score of 0.965 (Fig 4).
To ascertain whether SSL and synthetic dosage lethality (SDL) interactions share topological predictors we re-purposed our models to predict SDL gene pairs. We achieved an in-species AUC of 0.78 for H. sapiens pairs and 0.89 for S. cerevisiae pairs, a significantly improved score compared to that achieved during S. cerevisiae SSL pair classification. Our consensus model, utilising just H. sapiens and S. cerevisiae data, improved our H. sapiens predictions slightly (ROC AUC 0.80) (S3 Table).
SDL and SSL pairs in H. sapiens exhibit broadly similar feature distribution and feature importance for both classifiers. Despite this only 7,531 pairs were predicted as both SDL and SSL (of 41,103 SDL pair predictions and 59,475 SSL pair predictions).
In our human SDL models cohesion and shared cellular compartment GO terms featured as important features for both classifiers though molecular functional GO term annotation proved an important feature for SDL classification while shared biological process GO term featured well for SSL classification. The closeness feature, which measures how many steps is required to reach all other nodes from a given node, performed well for SDL classification. On the other hand coreness, a measurement of how well connected a node's neighbours are compared to the graph overall provided better predictive power for SSL classification.
We next compared biological process GO terms present in SDL and SSL pairs. We found that DNA damage related processes were more frequently seen in SDL pair data than in SSL pair data (~1.00% cellular response to DNA damage stimulus, ~0.70% DNA repair in SDL pairs compared to ~0.53% and ~0.46% respectively in SSL pairs). MAPK cascade and regulation of cell proliferation processes were well represented in both groups.
As discussed in the introduction, a number of other studies have used similar methods to predict genetic interactions. Most notably, this study shares a number of similarities with SINaTRA [32]. However, SLant has been developed for a wider number of organisms, including using human data directly, uses an enhanced feature set, our predictions have been experimentally validated (see below) and all of our data are available via the SLorth database (see below).
Algorithmically, the similarities between SLant and SINaTRA include some of the features used and the treatment of normalisation to allow cross-species prediction. However the PPI data used by SLant were sourced from STRING and were filtered for reliability, while SINaTRA's PPI data were sourced from BioGRID. A number of key algorithmic differences include SLant's use of consensus models, for both SSL and SDL interactions, and the use of a large range of topological, community and GO features. SLant also treats node-wise features differently and includes the averaged difference between genes in a pair as well as the individual values for each gene. We show that the novel features present in SLant improve the results in the feature holdout section (see Our pair-wise distance features are the most predictive) and propose that the different data sets appear to be providing a large impact on the results. A comparison of the features used in the two studies are available in S7 Table.
Unfortunately, the source code for SINaTRA is not available. However we were able to assess how our algorithm performed compared to SINaTRA, by testing it on the historical yeast SSL data from BioGrid 3.2.104 that had been used in the development of the SINaTRA algorithm. SINaTRA reports impressive AUC ROC values of 0.92 for in-species S. cerevisiae SSL predictions, 0.93 for in-species S. pombe SSL predictions, 0.86 for S. cerevisiae to S. pombe cross species validation and 0.74 for S. pombe to S. cerevisiae cross species validation. We obtained similar results using cross validation (as reported by SINaTRA) with AUC ROC values of 0.98 for in-species S. cerevisiae SSL predictions, 0.98 for in-species S. pombe SSL predictions, 0.88 for S. cerevisiae to S. pombe cross species validation and 0.77 for S. pombe to S. cerevisiae cross species validation (see S8 Table).
Next, we re-implemented SINaTRA by running SLant with a close approximation of the features that SINaTRA used originally but using the current STRING PPI network and current SSL data for training (see S9 and S10 Tables). We found that SLant outperformed SINaTRA in all tests apart from the S. pombe to S. cerevisiae cross species validation (AUC ROC 0.607 versus 0.609). In particular SLant considerably outperforms SINaTRA using models generated using the pair-wise non-bias segregated training sets. This supports our theory that the additional pairwise features incorporated into SLant leads to a generalisation of the models.
Finally we analysed the 2518 predicted human SSL pairs, with a SINaTRA score of over 0.90, that were published in the original paper. Of these, none of these predictions have subsequently been reported in BioGRID, either as SSLs or as negative genetic interactions. However, the number of reported SSLs for humans is still rather low. Encouragingly, 55% of the SINaTRA high confidence SSL predictions were also predicted to be SSLs by SLant.
We employed the full cross-species consensus model to predict SSL and SDL gene pairs in all of our species. All pairs that did not achieved a consensus score of over 0.75 were removed from our final prediction list. All predictions are available in the Slorth database http://slorth.biochem.sussex.ac.uk.
The graphical visualizations of the SSL predictions and the experimentally derived SSL interactions from our training data (S4A Fig) shows that the SSL network becomes much denser around the genes represented in the initial training data from BioGRID. This suggests that genes already implicated in an SSL pairs may share more SSL interactions than currently experimentally identified.
Using the models and classifiers described above we have identified and validated previously unpublished human SSLs that could be exploited therapeutically in the treatment of cancer. To identify potential therapeutic targets using our consensus method, we identified all the SSL gene pairs in H. sapiens where one of the genes had been identified as a tumour suppressor by the cancer gene census [56] (S4B Fig, appendix Table 4) and the other was a target of a drug approved for human use.
We found an enrichment in highly scoring SSL pairs containing the tumour suppressors VHL and PTEN. SSL pairs with the highest consensus scores included SREBF1, a transcription factor that binds to sterol regulatory element-1 and VHL (confidence score 0.810) and PTEN and SFN, a gene associated with breast cancer (confidence score 0.808). Other novel, highly scoring gene pair predictions that included cancer associated genes included PARP1 with PBRM1, BRCA2, ARID1A and APC as well as PIK3CA with MAP2K1, ABL1 and EGFR.
Validation on a handful of these predicted pairs providing some evidence that PBRM1 / PARP1 and PBRM1 / ABL1 share previously undescribed SSL interactions. We also see some evidence that PBRM1 / POLA1 share a synthetic rescue interaction.
A set of predicted gene pairs, where one of the genes identified was PBRM1, was selected for experimental validation. The PBRM1 gene codes for the tumour suppressor BAF180 a protein that plays a key role in both chromatin remodelling and gene transcription. It is frequently mutated in a subset of cancers including Clear Cell Papillary Renal Cell Carcinoma and Clear Cell Renal Cell Carcinoma [57] We chose gene pairs where the second gene codes for a protein which has published inhibitors. These included; PARP1, ERBB2, RAF1, POLA1, JAK2, ABL1, GSK3B (S5 Table). Inhibitors were chosen and procured via Sellekchem (https://pubchem.ncbi.nlm.nih.gov/source/Selleck%20Chemicals).
Clonogenic survival assays [58] were prepared for a control group and a BAF180 knockout group from the U2OS cell line. Both cell groups were treated with a range of drug concentrations based on previous literature for each. The resulting cell colonies were stained and counted after 14 days of incubation.
Of the drugs tested, three showed differential effects on the BAF180-deficient cells when compared to the control cells. PBRM1 mutant cells were more sensitive to both the PARP inhibitor and, to a lesser extent, ABL1 inhibitor than the control cells (Fig 5 with plate photography in S5 Fig), whereas the PBRM1 mutant cells appeared less sensitive to the POLA1 inhibitor than the control cells (Fig 5). Interestingly, cells lacking ARID1A, which is another SWI/SNF subunit, are also selectively sensitive to PARP inhibitors [59, 60], which supports this relationship. We also note this ARID1A / PARP1 SSL interaction was not present in the BioGRID data used to generate our training set but was also predicted with a high probability by SLant. The two protein products of the two genes SSL with PBRM1; PARP1 and ABL1, share a number of similar cellular processes such as regulation of differentiation, proliferation and of DNA damage and stress response. POLA1 which potentially shares a different type of interaction, synthetic rescue, plays an essential role in the initiation of DNA replication.
In this paper we have predicted SSL relationships using features derived from both in-species and cross-species PPI network information. The SLant consensus classifier out-performs previous attempts at predicting human and model organism SSL interactions and may provide a useful tool in guiding future experimental validation of SSL pairs.
The original intention in this study was to predict cross-species without using the target species' data in the training set. However our in-species predictions generally performed so well it seemed sensible to instead use the additional cross-species data as an enhancement instead. The only in-species classifier that underperformed was that derived for S. cerevisiae. However, this result should be interpreted with caution; direct comparison of results is not possible as there are differences in the validation data. So that others may compare their algorithm to ours we have made all of the source code for SLant freely available so that our results, training data and validation can easily be recreated and repeated.
Improving the quantity and the quality of the input data will also improve the quality of the SSL and SDL predictions. For instance the amount of genetic interaction data is very limited in humans and D. melanogaster. Protein-protein interaction data is plentiful for humans and the model organisms studied, but the majority of the interactions are unlabelled. Adding additional annotation to these interactions, e.g. the direction of an interaction, may improve predictions if enough labelled data were available. Also, both the PPI and the genetic interactions reported have ‘popularity bias’; genes and proteins of biological or medical interest are frequently studied and hence more interactions involving them are reported.
Recently Abdollahpouri et al. [61] developed a flexible regularization-based framework which can be used to control for popularity. An adaptation of this method to enhance the coverage of less frequently reported genetic interations, may help mitigate this bias. Furthermore, providing a reliability score for genetic interactions and only using the more reliable ones may be particularly important for S. cerevisiae where although there is a wealth of data, the number of false positives reported experimentally may be corrupting the prediction accuracy.
In an attempt to ascertain whether synthetic lethal interactions occurred within or between local clusters of genes in our physical network we applied a spin glass random walk to assign genes to distinct clustered communities separated by choke points across the graph. Analysis showed that the majority of SSL interactions occurred between these communities rather than within them. Based on the shorter distance between SSL genes and higher occurrence of adjacency presumably SSL genes are often at the peripheries of these communities. Further exploration of how SSL pairs are distributed between clustered communities such as these may shed further light on the node wise features of genetic interactions.
Although this study does not use orthology data directly we do note that our GO annotation features may in some way serve as a proxy for orthology data and this study could be also be expanded in the future through improved analysis of the relationship between GO terms and pairwise SSL pairs.
The identification of SSL interactions is a key step in expanding and improving targeted cancer therapy. The results presented here suggest that inhibition of PARP1 or of ABL protein kinase 1 may have therapeutic value in tumours lacking functional BAF180. The computational and experimental validation of our models performance presented in this study suggests that the predictions provided by SLant, all of which have been made publicly available, will be useful in guiding future SSL screening studies and ultimately in the continued goal of generating a more complete list of human SSL pairs.
Gene and orthology data were downloaded from Ensembl [62], Genetic interaction data were obtained from BioGRID (version 3.4.156) [42] with supplementary D. melanogaster data downloaded from Flybase (version 6.13) [63]. Each gene was labelled with gene ontology (GO) data from the gene ontology consortium [41]. Protein-protein interaction (PPI) data were obtained from the STRING database (version 10) [39]. To ensure reliability only experimentally derived and curated pathway data with a reliability cut-off of 80 were utilised (S6 Table). The Ensembl ENSP protein IDs in the PPI data sets were converted to their respective Ensembl ENSG gene IDs. This enabled us to relate the physical interaction data to the genetic interaction data and label each physical interaction gene pair as SSL (if present in the BioGRID data) or non-SSL (if the pair was not present in the BioGRID data).
For each organism an equal number of non-SSL pairs were assigned randomly to constitute the negative training set. When assigning a non-SSL pair, we checked to makes sure that its orthologues had not been assigned as having an SSL as, although it is not prescriptive, there is an enrichment of SSL pairs in orthologous genes.
Similar methods were used to build the training set used for our SDL interaction classifiers but we instead extracted BioGRID pairs annotated as synthetic dosage lethal as our positive class data.
The R (version 3.4.0) igraph package (version 1.1.2) [40] was used to generate a network representation of the PPI data for each of our 5 organisms and to calculate network features. (Table 1). Whilst we extracted network features for just a subset of all possible gene pairs the entire network of protein interactions was used in each calculation.
The features generated for our models were broadly categorised as node-wise or pairwise features as listed in Table 1. In general node-wise features, such as degree, were calculated by extracting network parameters for single nodes and finding the averaged distance between them as a pairwise feature. Pairwise features such as shortest path were calculated by igraph on each pair. To calculate shared GO terms, classed as a pairwise feature, we took a count of overlapping GO terms between the genes in each pair.
To generate our community features we applied a spin-glass random walk using the R igraph communities module to assign genes to 20 distinct communities separated by choke points across the graph. The final count of communities, 20, was chosen by measuring the predictive performance of our community features with a community count incrementing in steps of 5. After 20 communities we saw no further improvement.
The entire feature generation pipeline for the full complement of available gene pairs proved computationally intense, especially the generation of pairwise features such as cohesion, and run-time took up to 120 hours for each organism on an 8x Intel Xeon 3.50GHz processor with 16Gb RAM.
Before analysis all features in each dataset were normalised so that all feature values fell between 0 and 1. The resulting feature sets were divided into training, test and unlabelled sets. For each organism the feature set was under sampled to provide a balanced training set with an equal number of SSL and non-SSL pairs. The training set was further partitioned 80:20 to create a test set. The non-SSL pairs removed from the training data as part of under sampling were set aside as unlabelled data to be used in the prediction section of this study.
Some genes are highly represented in our available SSL training data whilst some only occur once, so generating two sets with balanced classes and a requisite number of observations posed a challenge. To create balanced training and test datasets with enough observations to perform validation we first created a list of genes ranked by the number of pairs they were found in. Next we divided this list adding the first to our list of genes available in our training data, the second to our test data and so on so that both data sets had a similar distribution of gene representation. Finally we used these two gene lists to filter our feature data into two subsets with no overlapping genes and balanced class.
We used the “ranger” e1071 random forest classifier, part of the R caret library, to model and classify SSL and non-SSL interactions in our training set. 5-fold cross validation was applied to each organism's training set to tune the model's hyper-parameters and the best model was used to assess predictive performance within each species. These optimised models were then used to predict SSL pairs across species, both in H. sapiens and across all other model organisms. These predictions were outputted as the probability of each class and were validated against the test data set.
In an attempt to further improve accuracy, as well as pairwise cross-species predictions, a consensus was taken from the predictions on the test set from all other species. This consensus was calculated by running a second classifier, a boosted Generalized Linear Model (GLM) that was trained on the previous classifiers outputs. To allow for validation this consensus dataset was segmented into a train and test set (both 0.5 the size of the original due to the smaller overall size). Finally we used this consensus model to predict SSL pairs in the unlabelled data set.
All of the R source code for SLant is available publically at https://bitbucket.org/bioinformatics_lab_sussex/slant. All data used is available via public resources.
A subset of potential SSL interacting pairs featuring PBRM1 (BAF180) complemented with genes with a known inhibitor were chosen from our predictions for experimental validation (S5 Table).
|
10.1371/journal.ppat.1002416 | Autophagy Protein Atg3 is Essential for Maintaining Mitochondrial Integrity and for Normal Intracellular Development of Toxoplasma gondii Tachyzoites | Autophagy is a cellular process that is highly conserved among eukaryotes and permits the degradation of cellular material. Autophagy is involved in multiple survival-promoting processes. It not only facilitates the maintenance of cell homeostasis by degrading long-lived proteins and damaged organelles, but it also plays a role in cell differentiation and cell development. Equally important is its function for survival in stress-related conditions such as recycling of proteins and organelles during nutrient starvation. Protozoan parasites have complex life cycles and face dramatically changing environmental conditions; whether autophagy represents a critical coping mechanism throughout these changes remains poorly documented. To investigate this in Toxoplasma gondii, we have used TgAtg8 as an autophagosome marker and showed that autophagy and the associated cellular machinery are present and functional in the parasite. In extracellular T. gondii tachyzoites, autophagosomes were induced in response to amino acid starvation, but they could also be observed in culture during the normal intracellular development of the parasites. Moreover, we generated a conditional T. gondii mutant lacking the orthologue of Atg3, a key autophagy protein. TgAtg3-depleted parasites were unable to regulate the conjugation of TgAtg8 to the autophagosomal membrane. The mutant parasites also exhibited a pronounced fragmentation of their mitochondrion and a drastic growth phenotype. Overall, our results show that TgAtg3-dependent autophagy might be regulating mitochondrial homeostasis during cell division and is essential for the normal development of T. gondii tachyzoites.
| Autophagy is a catabolic process involved in maintaining cellular homeostasis in eukaryotic cells, while coping with their changing environmental conditions. Mechanistically, it is also a process of considerable complexity involving multiple protein factors and implying numerous protein-protein and protein-membrane interactions. The cellular material to be degraded by autophagy is contained in a membrane-bound compartment called the autophagosome. We have characterised the formation of autophagosomes in the protozoan parasite Toxoplasma gondii by following the relocalisation of autophagosome-bound TgAtg8. Thus, exploiting GFP-TgAtg8 as a marker, we showed that it is a process that is regulated and can be induced artificially by amino acid starvation. Autophagic vesicles were also observed in normally dividing intracellular parasites. Depleting Toxoplasma of the TgAtg3 autophagy protein led to an impairment of TgAtg8 conjugation to the autophagosomal membrane and, at the cellular level, to a fragmentation of the single mitochondrion of the parasite and to a severe growth arrest. We have thus found that TgAtg3-dependent autophagy is essential for normal intracellular development of T. gondii tachyzoites.
| Proteolysis is very important to eukaryotic cells and occurs at a considerable constitutive rate. Base line degradation regulates the levels of numerous proteins and removes misfolded proteins. Mechanistically, this process can be separated into two main pathways: one pathway mediated by the proteasome and the other pathway by the lysosome [1]. The proteasome relies on the ubiquitin system for selection of target proteins and plays a major role in the rapid degradation of short-lived proteins as well as abnormal proteins. The lysosome, which represents the terminal compartment of the endosomal pathway, is a membrane-bound organelle that contains a diverse array of hydrolases for the degradation of plasma membrane proteins and endocytosed extracellular proteins. Lysosomes are also involved in bulk degradation of cytoplasmic components, such as long-lived cytosolic proteins and organelles, and this is achieved through a process called autophagy. Autophagy has been divided into several classes of pathways, such as macroautophagy, microautophagy, and chaperone-mediated autophagy, but macroautophagy has been studied most extensively and we will refer to this specific form of the process as “autophagy” in this manuscript for simplicity.
Autophagy is evolutionarily conserved in eukaryotes from yeast to mammals and has important roles in various cellular functions [2]. The basal role of autophagy is in turnover and recycling of cellular constituents; these housekeeping functions include the elimination of defective proteins and organelles and the prevention of abnormal protein aggregates accumulation. Autophagy also plays an important role in organelles and proteins (but also lipids) recycling under nutrient starvation conditions, as a nutrient source for the cell. Finally, there are pleiotropic and more specialised roles for different eukaryotic cells (in particular in mammals) including cellular remodelling during differentiation and development, regulation of cell longevity and programmed cell death, elimination of invading pathogens and providing antigens to the immune system [2], [3].
In the autophagic process, cytosolic components are sequestered in a double-membrane vesicle known as the autophagosome. The outer membrane of the autophagosome will then fuse with the lysosomal compartment to deliver the inner contents of the vesicle, which will be subsequently degraded. Although autophagy has been initially identified in mammalian cells, the characterisation of the molecular machinery involved in this cellular process has been mainly done in yeast. Due to the ease of genetic analyses in the yeast system, screening for mutants unable to survive nitrogen starvation, allowed the identification of more than thirty Atg genes involved in autophagy and the related cytoplasm to vacuole targeting pathway (Cvt) [4]. Of these, some are only present in one organism and others represented by orthologues in different eukaryotic cell types. Among all Atg proteins, Atg8 occupies a central position: it is essential to the process of autophagosome formation, especially for the membrane expansion step [5], [6] and, possibly, the final membrane fusion steps [7]. The protein is present as a soluble form in the cytosol of eukaryotic cells, and gets recruited to the autophagosomal membrane upon induction of autophagy. Interestingly, the binding of Atg8 to the autophagosomal membrane involves two conjugation systems that resemble ubiquitin-conjugation systems [8].
Toxoplasma gondii is an obligate intracellular protozoan parasite that is virtually able to infect all species of warm-blooded animals [9]. T. gondii is a member of the phylum Apicomplexa, which also includes several other notable human pathogens such as Plasmodium and Cryptosporidium. The secretory pathway of T. gondii is highly polarised and includes several unique organelles devoted to the invasion of the host cell (the micronemes, rhoptries and dense granules) [10]. In addition the parasite harbours the apicoplast, a plastid-like organelle of endosymbiontic origin. The micropore, a cytostome-like structure formed by the invagination of the plasma membrane, can be seen laterally [11], which suggests the existence of an endocytic pathway within the parasite, although it has not been clearly defined at the molecular level. Until very recently, the parasite was thought not to contain any morphological equivalent of “classic” lysosomes, as rhoptries appeared to be the only acidified organelles containing hydrolases in T. gondii [12]. Two recent reports have nevertheless identified an acidified vacuole bearing a cathepsin-like peptidase in Toxoplasma tachyzoites that could fulfil that role [13], [14], but its precise physiological role remains to be characterised. The autophagic pathway is even less well described in T. gondii, although it can reasonably be suspected to be involved in the survival of Apicomplexa, as well as for other parasites with complex life cycles. For example, the importance of autophagy for the development and virulence of protozoan parasites such as the trypanosomatids has been shown previously [15], [16] and recent data have shown an implication of autophagy in the cellular remodelling of Plasmodium liver stages [17]. Here, we used the experimental opportunities afforded by Toxoplasma as a genetic model organism to investigate the autophagic machinery and assess its physiological role in the parasite.
Many of the molecular components involved in autophagy have been identified in the budding yeast Saccharomyces cerevisiae. We have thus used the known S. cerevisiae Atg protein sequences to identify T. gondii homologues through homology searches in the ToxoDB database ([18], Table S1). The core machinery for the assembly of the pre-autophagosomal structure appears to be present in T. gondii. This includes the regulating kinase Atg1, as well as proteins central to autophagosome formation itself, like Atg8 and its conjugating partners Atg7–Atg3. In several other eukaryotes, Atg8 requires the activity of two conjugation systems to bind the autophagosomal membrane: Atg7–Atg3 on the one hand and Atg5–Atg12 on the other (see [8], [19] for a review), the latter being dispensable in vitro [20].
Interestingly, no clear orthologues of the members of the Atg5–Atg12 conjugation system were found in our homology search. These are also absent from the genomes of several other parasitic protozoa [21]–[23]. In Leishmania, distantly related proteins might have similar function [24]. However, the orthologues of these Leishmania proteins cannot be found in the T. gondii genome. Overall, this either suggests that the Atg5–Atg12 conjugation is not essential for autophagosome formation in T. gondii, or that Toxoplasma uses different (possibly lineage-specific) proteins to perform this task.
In fact, a significant part of the yeast Atg proteins seems to be missing from the T. gondii genome. For instance, most of the Atg proteins involved in the Cvt pathway are not found in Toxoplasma. This may not be surprising as this pathway is likely fungi-specific, and is also largely lacking in other eukaryotic systems that are clearly able to perform autophagy [19].
The most reliable marker for autophagosomes is Atg8 (or LC3, its mammalian counterpart). This protein is essential to autophagosome formation, and stays associated with the autophagosomal membrane from the early membrane recruitment step until the fusion with the lysosomal compartment [25]. Hence, to further characterise the autophagic process, we raised a specific antibody against TgAtg8 (TGME49_054120, Table S1). Using this antibody on tachyzoite lysates, we detected by Western blot analysis a major protein band consistent with the predicted molecular weight of 14 kDa (Figure 1). However, the rather weak signal that was observed suggested a low level of expression for endogenous TgAtg8. To observe the dynamics of autophagosomes formation we then cloned TgAtg8 in fusion with the green fluorescent protein (GFP) gene to generate a T. gondii cell line that stably expresses GFP-TgAtg8 in addition to its endogenous copy. Stable clones of GFP-TgAtg8 transgenic parasites were obtained and showed no apparent difference in growth or morphology (data not shown). When GFP-TgAtg8 parasites were assessed by Western blot with anti-TgAtg8 and anti-GFP antibodies, both detected a fusion protein at the expected molecular weight of 42 kDa (Figure 1). The strong tubulin promoter driving the expression of GFP-TgAtg8 showed an overexpression of ∼20 fold of the GFP-fused TgAtg8 compared with the native protein, which greatly facilitated its detection.
We studied the localisation of the GFP-TgAtg8 fusion protein by fluorescence microscopy on living tachyzoites. In extracellular tachyzoites freshly released from host cells, the GFP-TgAtg8 signal was generally distributed throughout the cytoplasm and also occasionally found in one or more punctate vesicles of various sizes that could correspond to autophagic vesicles (Figure 2A). Using the anti-TgAtg8 antibody, both on the parental cell line and on the GFP-TgAtg8-expressing parasites, we verified by immunofluorescence assay (IFA) that GFP-TgAtg8 and anti-TgAtg8 labelled identical cellular compartments (Figure S1). The GFP-TgAtg8 signal was convenient to follow; we thus conducted our subsequent microscopic observations on the GFP-TgAtg8 cell line. To establish whether the observed structures are autophagic vesicles, we performed starvation experiment on GFP-TgAtg8-expressing extracellular parasites. Free tachyzoites allow a more direct control of environmental conditions and nutrient access, compared with intracellular parasites sheltered within their host cells. Strong induction through starvation is a hallmark of autophagosomes in many eukaryotic systems. To induce starvation, we incubated GFP-TgAtg8-expressing extracellular tachyzoites for increasing intervals of time in an isotonic solution devoid of amino acids (Hank's balanced salt solution, HBSS). We imaged the parasites and quantified our findings. We noticed that the number of parasites bearing GFP-labelled puncta increased over the incubation time (increasing from 15±3% to 79±8% after 8 hours, Figure 2B) as did the number of puncta per parasite (Figure S2A). As a control, incubation of the tachyzoites in complete Dulbecco's modified Eagle medium (DMEM, supplemented with 10% fetal bovine serum) for similar time periods induced no appearance of these puncta (17±2% of parasites bearing GFP-labelled puncta versus 21±2% after 8 hours).
These changes in the GFP-TgAtg8 signal were induced quickly and reached a plateau after ∼8 hours of starvation. The re-localisation of GFP-TgAtg8 from the cytosol to vesicular structures during starvation suggested that these puncta were indeed autophagosomal structures. The presence of punctate GFP-TgAtg8 signal in ∼20% of the extracellular tachyzoites at the start of the experiment might reflect basal autophagy, or a proportion of the population that have egressed early and are already in a relative nutrient stress condition. The majority of these extracellular tachyzoites showed a single relatively large vesicle, which appeared prominently localised in the anterior part of the parasite, in the Golgi apparatus/apicoplast region (Figure S2B). Similarly, we performed microscopic observations on intracellular tachyzoites and found that they also occasionally displayed vesicular GFP-TgAtg8 in normal growth conditions (Figure S2B). Additional vesicles were found to be localised in a non-polarised manner throughout the cytoplasm of the tachyzoites. Although mostly present as a punctate signal, the GFP-positive structures were found to display some heterogeneity as previously described for GFP-TgAtg8 in other eukaryotes [26].
We sought to morphologically characterise these autophagic vesicles. Tachyzoites, either freshly released or starved for 8 hours, were fixed and processed for electron microscopy analysis. Compared to the control parasites, the starved tachyzoites displayed a higher number of large cytoplasmic vacuoles (Figure 3A, V). Additionally, in the starved parasites we identified several membrane-bound vesicles of about 300–900 nm in diameter containing cytoplasmic material or organelles. These structures were usually bounded by two membranes (Figure 3, Ap). These are commonly observed structural features of autophagosomes [27]. As an example of organelles that could be found in these vesicles, part of the unique mitochondrion of a tachyzoite was seen segregated inside a membranous compartment (Figure 3A) and co-labelling of GFP-TgAtg8-decorated autophagosomes and mitochondrion showed partial co-localisation (Figure 3B), which is suggestive of mitophagy occurring in Toxoplasma. When the autophagosome has fused with the lysosomal compartment to deliver its content for degradation, it becomes an autolysosome, or autophagic vacuole. These degradative vacuoles containing partially degraded cytoplasmic material could also be identified in starved parasites (Figure 3, Av). Importantly, similar structures could be identified in a starved cell line expressing GFP-TgAtg8, by immuno-electron microscopy with an anti-GFP antibody (Figure 3C), providing a link to the observations we made by fluorescence microscopy.
Our results thus show that nutrient starvation triggers the appearance of autophagosomes in T. gondii tachyzoites through the recruitment of autophagosomal marker TgAtg8, indicating that autophagy is operating and functional in this parasite.
The conjugation process of Atg8 to the autophagosomal membrane is fairly well understood at the molecular level in yeast and mammalian cells and resembles ubiquitination [8]. In order to bind the autophagosomal membrane, Atg8 must first undergo proteolytic maturation by the cysteine peptidase Atg4 [28], [29]. This exposes a C-terminal glycine that is then conjugated to a phosphatidylethanolamine (PE) lipid moiety through the action of the Atg7–Atg3 system. Accordingly, most eukaryotic Atg8 orthologues described so far bear one or several amino acids after the C-terminal glycine. However, to our surprise, translation of the putative TgAtg8 gene as annotated by the ToxoDB genome database results in a protein that ends in a glycine in the absence of post-translational maturation. This feature was conserved in other predicted apicomplexan Atg8 proteins, with the exception of several Cryptosporidium species (Figure S3). To confirm this, we performed a 3′-RACE experiment using T. gondii cDNA as template and primers specific for Atg8 to map the stop codon. In all the clones that we obtained and sequenced, the penultimate codon was confirmed to be coding for a glycine (data not shown). It is conceivable that this peculiar feature may result in constitutive lipidation of Atg8 in T. gondii and that maturation by the Atg4 peptidase may not be required. However, analysis of the T. gondii genome reveals the presence of a protein with some similarity to Atg4 from yeast (Table S1), and a conserved active site catalytic triad appears to be present. This peptidase, which appears to be expressed in T. gondii tachyzoites according to the proteomic data available in the ToxoDB database, could be involved in TgAtg8 recycling from the autophagosomal membrane, but this remains to be investigated.
Atg8 and its lipidated form can be separated by SDS-PAGE in the presence of urea [25]. We thus used this technique to separate proteins from GFP-TgAtg8 lysates, followed by Western blot analysis with an anti-GFP antibody. In these conditions, the anti-GFP antibody revealed a major protein band with a molecular size consistent with the prediction for a GFP-TgAtg8 fusion protein (42 kDa) and an additional faster migrating protein (Figure 4A). We used the anti-Atg8 antibody on the parental cell line: we were also able to separate two isoforms of the native protein by urea SDS-PAGE at around 14 kDa (Figure 4A). Lipidated Atg8 typically migrates more rapidly than the non-lipidated form [29], thus the faster migrating protein possibly corresponds to the GFP-TgAtg8-PE form. To confirm that the faster migrating protein was indeed a membrane-associated form of GFP-TgAtg8, we performed cell fractionation experiments that showed the faster migrating form was exclusively present in the 100 000 g insoluble pellet (in contrast, the slower migrating protein was present in both fractions, but appeared to be enriched in the soluble fraction) (Figure 4B). Similar results were obtained on parental cells with the anti-TgAtg8 antibody (Figure S4). Moreover, only a treatment by a detergent (DOC), but not by chemicals disrupting low energy bonds (NaCl and urea), could lead to the complete solubilisation of TgAtg8 proteins present in the membrane fraction (Figure 4C). This has been described for other Atg8s [30] and suggests a tight association of TgAtg8 to membranes. Metabolic labelling using 3H-ethanolamine (as a precursor for PE) and subsequent immunoprecipitation of TgAtg8, revealed that TgAtg8 indeed incorporated the label, thus confirming that the membrane association was likely to be mediated by PE (Figure 4D). Also, the abundance of this membrane-associated form increased with the duration of the starvation period (Figure 4A). This confirmed that in T. gondii tachyzoites TgAtg8 becomes increasingly lipidated, and hence, potentially, more is recruited to the autophagosomal membranes during starvation (as more autophagosomes are formed).
It is to note that proportions of soluble versus membrane-associated forms of TgAtg8 are higher in the GFP-TgAtg8-expressing cell line, compared to what is generally observed for TgAtg8 in the parental cell line. One explanation is that while the global pool of GFP-TgAtg8 is higher in the GFP-TgAtg8 parasites, the number of autophagosomal vesicles harbouring the membrane-bound form is limited, leaving a significant part of GFP-TgAtg8 in the cytosol.
We then constructed, by site-directed mutagenesis, a variant of GFP-TgAtg8 were we replaced the C-terminal glycine by an alanine. We expected that this should abolish the lipidation of the protein [31]. Tachyzoites were transfected with the GFP-TgAtg8-G/A construct and GFP-positive clones were selected. Autophagy was induced by starvation in amino acids-depleted medium as described above and the appearance of autophagosomes was monitored by fluorescence microscopy (Figure 5A). While the proportion of autophagosome-bearing cells increased along time in the GFP-TgAtg8 control cell line, we did not observe the formation of GFP-labelled autophagosomes in the GFP-TgAtg8-G/A mutant. This finding suggests that the mutated protein is not recruited to autophagosomes (Figure 5A and B), confirming the essential role of the glycine for the recruitment. This finding further validates the interpretation of the GFP-TgAtg8 positive vesicular compartment as autophagosomal. We also followed by Western blot the lipidation of GFP-TgAtg8-G/A following starvation. Consistent with the microscopy results, no lipidated form could be detected for GFP-TgAtg8-G/A by Western blot analysis following urea SDS-PAGE (Figure 5C).
Overall, in spite of a constitutively exposed C-terminal glycine, our results demonstrate that TgAtg8 exists both in soluble and membrane-associated forms in the tachyzoites.
Using GFP-TgAtg8 as a marker we were able to demonstrate that autophagy occurs in extracellular parasites under starvation conditions. However, these are not necessarily reflecting physiological conditions and we wondered whether the parasite might encounter the need for this process during its normal intracellular development. After invasion of the host cell, T. gondii tachyzoites replicate inside a parasitophorous vacuole by a process called endodyogeny [32]. This process is composed of single gap phase (G1) preceding a synthesis (S) phase, which is then followed by mitosis and cytokinesis through budding [33]. Daughter cells are assembled within the mother and, as they form, encapsulate most of the maternal cell contents, leaving only a small residual body behind. We hypothesized that some of the maternal material could be digested by autophagy to the benefit of the daughter cells. We thus sought to follow the presence of autophagy in intracellular parasites developing within their host cells in tissue culture. GFP-TgAtg8-expressing tachyzoites were used to synchronously infect host cells [34] and the proportion of intracellular tachyzoites displaying autophagic vesicles was measured as previously at different time points following infection.
It is to note that intracellular parasites were found to display a less intense and a more diffuse GFP-TgAtg8 labelling than the autophagic vesicles induced by starvation, yet slightly more intense than the cytosolic background (Figure 6B, compare 4h and 48 h timepoints). Similar structures were also seen in intracellular parasites expressing non functional GFP-TgAtg8-G/A and were partly co-localising with Golgi apparatus marker GRASP visualised with a red fluorescent protein (RFP) fusion (Figure S5), allowing us to rule out that they were autophagic vesicles. Taking this into account for our quantification experiments, it appeared that there was an initial slight increase in the proportion of cells displaying intensely labelled GFP-TgAtg8 autophagic vesicles, peaking at 2 hours post-invasion. However, such labelling generally decreased in following hours of intracellular development (Figure 6A), where parasites could still be found to display one or several GFP-TgAtg8 autophagic vesicles, but the majority showed a rather homogenous cytosolic signal (Figure 6B). Also, the GFP-TgAtg8 vesicular signal was usually not found at the residual body occasionally formed within the vacuole after parasite division (only in ∼2.5% of the residual bodies) (Figure 6B, arrowed).
To better follow the timing and localisation of the GFP-TgAtg8 signal, we used the inner membrane complex protein IMC1 [35] as a marker to track the progress of cytokinesis. This protein is a component of the subpellicular network that defines the periphery of both the mature tachyzoite and of the daughter cells developing with the mother cell. We imaged GFP-TgAtg8/IMC1-RFP co-expressing tachyzoites, and observed that the vesicular GFP-TgAtg8 signal was usually present in dividing cells (Figure 6C). More precisely, live imaging observation of these parasites showed that autophagosomes were usually present until the cytokinesis process, but disappeared shortly after (data not shown).
Altogether, our data argue against a continuous autophagic activity in intracellular parasites, yet vesicle formations could occasionally be observed and transient autophagy seemed to occur.
Autophagy is a multistep process that can be modulated by upstream kinase effectors and so can be interfered with using specific pharmacological agents. The “target of rapamycin” (TOR) kinase is a serine/threonine kinase which is central in regulating cellular growth by promoting anabolic processes and antagonising autophagy. Consequently, treatment with TOR inhibitor rapamycin mimics nutrient starvation and is known to increase the level of autophagy in several eukaryotes [36]. On the other hand, the class III phosphoinositide 3-kinase (PI3K) signalling cascade has been shown to promote autophagy, thus specifically inhibiting this kinase is thought to reduce autophagy [37]. Both a TOR kinase and a class III PI3K are predicted to be present in the T. gondii genome (Table S1); we have thus sought to use specific kinase inhibitors to evaluate the biological significance of autophagy in Toxoplasma. We drug-treated GFP-TgAtg8-expressing extracellular parasites and evaluated the effect of treatment on autophagy.
Increasing concentrations of rapamycin in the starvation medium resulted in higher proportions of GFP-TgAtg8 vesicles in extracellular parasites, suggesting an increased level of autophagic activity (Figure S6A). However, the effective concentration was significantly higher than routinely observed in yeast or mammalian cells (5 µg/ml instead of 0.5 µg/ml). This suggests a modest effect of rapamycin on the modulation of T. gondii autophagy, like previously observed with the TOR kinase of plants such as rice, tobacco or Arabidopsis [38] and the parasitic protist Trypanosoma brucei [23].
We also evaluated the effects of class III PI3K on autophagy by incubating extracellular tachyzoites in the starvation solution in the presence of 3-methyladenine (3-MA) or wortmannin, two known PI3K inhibitors. In these conditions, there were lesser proportions of GFP-TgAtg8 vesicles-positive cells amongst the tachyzoites treated with the two inhibitors (particularly wortmannin), which suggested a role for the PI3K in promoting autophagy in Toxoplasma (Figure S6B). However, again the concentrations of wortmannin and 3-MA we had to use were higher than those used for mammalian and yeast cells, similarly to plants were these PI3K inhibitors have to be used at greater concentrations to inhibit autophagy [39], [40].
In conclusion, the use of inhibitors specific for kinases known to regulate autophagy in other eukaryotic systems allowed us to show that they generally follow the same trends in T. gondii (inhibition of autophagy by the TOR kinase and activation by the class III PI3K), although the use of these drugs at relatively high concentrations can alter their specificity and would preclude a use on intracellular parasites to investigate the autophagic function.
As we could not use kinase inhibitors to satisfyingly investigate the function of autophagy in Toxoplasma tachyzoites, we sought to genetically produce an autophagy-deficient cell line. To this end, we chose Atg3 as a target, a gene coding for a protein involved in the conjugation of Atg8 and essential for autophagy [41]. In other eukaryotes, Atg8 is activated by Atg7 to form an Atg8/Atg7 thioester intermediate and is then transferred to Atg3 to form an Atg8/Atg3 thioester intermediate, before being finally conjugated to the amino group of PE for binding to the autophagosome membrane.
TgAtg3 (TGME49_036110, Table S1) was identified in the T. gondii genomic sequence based on protein sequence homology searches with eukaryotic orthologues and was found to have a conserved predicted active site region with, in particular, a conserved active site cysteine (Figure S7). T. gondii genomic database predicts a 397 amino acids-long TgAtg3 protein. Yet, when aligned to yeast or human orthologues, predicted TgAtg3 shows a N-terminal extension bearing a poly-serine motif (Figure S7). The corresponding putative mRNA region has a polypyrimidine tract (promotes the assembly of the spliceosome), with a putative downstream AG splice acceptor site and a potential downstream start codon that would produce a shorter version of TgAtg3, closer in size to other eukaryotic orthologues.
The regions corresponding to the 3′ and 5′ untranslated regions (UTRs) of T. gondii TgAtg3 gene were cloned in a plasmid on either side of a selection marker gene to use for a knock-out strategy by gene replacement through a double recombination event. We tried to transfect the RHΔHX strain [42] and the RH-derived ΔKu80 strain, which is more amenable to targeted gene deletion [43]. Despite numerous independent attempts, we were unable to obtain clones in which TgAtg3 had been deleted (data not shown).
The inability to delete the TgAtg3 gene with these different strategies suggested an essential role for the corresponding protein in tachyzoites. We thus sought to produce a conditional null mutant cell line for TgAtg3. We introduced an ectopic copy of TgAtg3 into the TAti tet-transactivator line by stable transformation [44]. The ectopic copy was placed under the control of the tetracycline-regulatable promoter 7tetOSag1. To aid detection of the corresponding protein, the transgene was tagged with a sequence encoding an N-terminal c-myc epitope. We cloned both a long and short versions of putative TgAtg3 into the expression vector (named imyc-lTgAtg3 and imyc-sTgAtg3, respectively). After transfection of tachyzoites of the TAti cell line, clonal parasites were obtained. Both cell lines showed an anhydrotetracycline (ATc)-regulatable TgAtg3 expression by Western blot and by IFA with anti-myc antibody (data not shown). IFA also demonstrated a cytosolic localisation for myc-tagged TgAtg3 (long and short versions alike), which is compatible with a functional role in autophagy, where it should be conjugating cytosolic Atg8 to the autophagosomal membrane (data not shown).
To replace TgAtg3, we used a cosmid-based gene disruption strategy, allowing the use of large flanking regions to increase the chance to recombine at the appropriate locus [45]. A single cosmid clone was available for the TgAtg3 locus (TOXOU62). In this cosmid the TgAtg3 open reading frame was replaced with a chloramphenicol acetyl transferase marker by recombineering (Figure 7A). Parasites expressing a regulatable copy of TgAtg3 were transfected with the resulting cosmid and stable clones were isolated by chloramphenicol selection. Clones were tested by PCR for 5′ and 3′ disruption of the TgAtg3 locus, and presence of the resistance cassette (Figure 7B). We note that we successfully obtained a null mutant clone in the imyc-sTgAtg3 background, but not in the cell line expressing the longer version of TgAtg3. This suggests that this short transcript likely encodes the functional protein. Further analysis by Southern blot confirmed the correct integration of the cosmid-derived cassette and the disruption of native TgAtg3 locus (Figure 7C). Conditional mutant clone displayed a tight regulation of TgAtg3 expression by ATc, as shown by Western blot and IFA with anti-myc antibody (Figure 8A and B): Western blot analysis showed that a 2 days treatment with ATc was reducing almost completely the expression of the protein and that a 4 days treatment led to undetectable levels of the extra copy.
We next assessed whether the Atg8-conjugation function of Atg3 was conserved in T. gondii. To do so, we used the conditional mutant to deplete intracellular tachyzoites of TgAtg3 by a 2 days treatment with ATc, we then isolated extracellular parasites and triggered autophagy by exposing them to starvation medium for increasing intervals of time as described before. Parasites extracts were prepared and analysed by urea SDS-PAGE to visualise the lipid-conjugated form of TgAtg8. As described above, in the presence of TgAtg3 we observed an increase in the TgAtg8 lipidated form, as autophagy was induced (Figure 8C). In contrast, when TgAtg3 was repressed by ATc treatment: i) a significant proportion of the slower migrating form was present even before autophagy was induced and ii) the lipidated form was not upregulated by starvation (Figure 8C). This shows that reducing the levels of TgAtg3 leads to a reduced capacity in TgAtg8-conjugation to the autophagosomes and thus likely impairs the putative autophagic function in the parasite.
We examined the effect of TgAtg3 depletion on parasite growth using plaque assays. The conditional TgAtg3 null mutant formed markedly smaller plaques when repression of the imyc-sTgAtg3 copy was induced by ATc (Figure 9A and B). Longer repression periods increased the effect on growth, as no plaque was visible with mutant parasites pre-incubated with ATc for 4 days prior to the start of the plaque assay (Figure 9A). No viable parasite could be detected after three passages following mechanical release, in the presence of ATc (data not shown), providing additional support for a critical function of TgAtg3. Growth was also assessed at shorter times following invasion. Mutant parasites that were untreated or pre-incubated with ATc for 4 days were allowed to invade host cells. Cultures were kept in the presence of ATc and fixed 24 or 48 hours later and numbers of parasites per vacuole were counted in both samples. TgAtg3-depleted parasites showed a considerable delay in growth compared to controls, and they did not progress through cell division as they accumulated vacuoles with mostly one or two parasites (Figure 9C).
As the intracellular development of tachyzoites appeared to be affected by TgAtg3 depletion, we sought to investigate whether this was associated with specific morphological defects. We performed IFA using a battery of antibodies recognising specific subcellular structures. These included secretory organelles such as the rhoptries, micronemes and dense granules, the IMC, the apicoplast and the mitochondrion (Figure S8 and data not shown). The organelle that stood out in these comprehensive analyses as being particularly affected by the lack of TgAtg3 was the mitochondrion. T. gondii tachyzoites typically have a single mitochondrion, which forms a reticulated network extending through most of the parasite (Figure 10, top, bottom). Strikingly upon depletion of TgAtg3, we observed the loss of the mitochondrial network as judged by staining for two independent mitochondrial marker proteins: F1 beta ATPase and HSP28 (Figure 10). In parasites that were grown for 2 days in the presence of ATc, the mitochondrion appeared highly fragmented or entirely absent. We note that a significant amount of staining was now found in the residual body. This structure is, located at the center of the parasitophorous vacuole and is thought to represent the residua of mother cells left behind by the emerging daughters (Figure 10). We confirmed this phenotype in transgenic parasites pre-incubated for longer periods with ATc, but it was not generally seen in TgAtg3-expressing control parasites exposed to ATc (Figure 10). To independently assess the morphology and the functional status of the mitochondrion in this mutant we performed labelling experiments with Mitotracker Red. This cationic dye accumulates in the mitochondrion depending on an intact membrane potential. We noted a profound loss of staining consistent with a concomitant loss of mitochondrial membrane potential and/or loss of the organelle (Figure S9).
No obvious defect could be detected for the apicoplast or the secretory organelles (Figure S8). We therefore assume that this phenotype reflects a specific role of Atg3 in mitochondrial maintenance rather than a general loss of cell structure due to necrosis and cell death. However, more subtle effects might be beyond the limit of resolution of fluorescence microscopy. Therefore we sought to analyse the morphology of TgAtg3-depleted parasites by electron microscopy.
Electron microscopy confirmed that these tachyzoites possessed morphologically normal secretory organelles (rhoptries, micronemes and dense granules) and apicoplast. We examined parasites grown in the presence of ATc for up to 4 days and found no apparent detriment to these organelles over this period (Figure 11 and data not shown). In contrast, electron microscopy revealed numerous mitochondrial defects ranging from alteration of cristae, to organelles including large collapsed membranous structures next to vestigial cristae that were the only recognisable feature of the mitochondrion (Figure 11). Altered mitochondrial material was also occasionally present in the residua found in the vacuolar space (not shown).
In conclusion, loss of TgAtg3 impedes the intracellular development of parasites and the most visible effect at the subcellular level is a dramatic collapse of the parasite mitochondrion.
Autophagy plays a role in organelle and protein turnover that is important for cellular homeostasis, adaptation to starvation and overall cellular development of eukaryotic cells. Yet, it was to date poorly characterised in apicomplexan parasites. Autophagy is dependent on the formation of the autophagosome, the vesicular structure in which the material to be degraded will be incorporated. In the present study we have used a molecular marker for autophagosomes, TgAtg8, to detect these structures in T. gondii. We have identified structures that show the morphological hallmarks of autophagosomes and can be formed and induced by starving extracellular tachyzoites; more importantly, we find that autophagosomes are also formed during normal development and that their presence is most pronounced at a particular point in the parasite division cycle.
Our genome mining efforts to uncover autophagy-related genes in T. gondii and other Apicomplexa have revealed a core of the autophagosome formation machinery in these parasites. We identified genes coding for proteins involved in the pre-autophagosome formation, the Atg8 conjugation system for autophagosome elongation, as well as upstream regulating kinases. However, several proteins previously described to regulate the formation of autophagosomes in other eukaryotic systems appear to be missing. Examples of these are the Atg5–Atg12 conjugation system for Atg8, or partners of TOR-regulated Atg1 kinase (i.e. Atg13, Atg17). This suggests that Apicomplexa possess a simpler system for autophagy that lacks some of the complex layers of regulations observed in mammalian cells or yeast [46] or, alternatively, that the regulation of autophagy in Apicomplexa involves parasite-specific proteins that remain to be discovered. Along this line, our use of inhibitors to interfere in T. gondii with known upstream autophagy-regulating kinases (inhibiting mTOR kinase and activating class III PI3K) has shown that, although following the same trends as their other eukaryotic counterparts, high concentrations of inhibitors were needed to act upon autophagy, suggesting differences in the kinase-dependent regulatory network. The structure of the Atg8 protein in T. gondii and many other Apicomplexa may be consistent with this apparent differential regulation. In yeast and mammals, Atg8 first needs to be processed by the Atg4 cysteine peptidase in order to bind to the nascent autophagosome membrane. This exposes a C-terminal glycine residue that will, in turn, be conjugated to a lipid by the Atg7–Atg3 complex. Yet, most apicomplexan Atg8 orthologues already constitutively bear a C-terminal glycine, which could suggest that autophagosome formation is constitutive in Apicomplexa. However, our experiments suggest some level of regulation of TgAtg8 binding to the autophagosomes, and this could occur by a mechanism independent of C-terminal proteolysis. Firstly, using urea PAGE, we could separate the non-lipidated from the lipidated isoform of GFP-TgAtg8 and native TgAtg8, and further showed that the proportions of the membrane-bound form were increasing in autophagy-inducing conditions. Secondly, microscopic observation of GFP-fused or native TgAtg8 showed dual vesicular and cytosolic localisations of the protein. Lastly, using ectopic expression of GFP-TgAtg8 mutants, we demonstrated that the C-terminal glycine was essential for conjugation.
Overall, our data suggest that autophagy in T. gondii is more complex than it appears at first glance and could have original ways of regulation compared with other eukaryotic systems studied so far.
As they invade their host cell, T. gondii tachyzoites establish themselves into a parasitophorous vacuole, which constitutes a niche that offers protection and nutrients and thus allows for efficient parasite multiplication (see [47] for a review). Spatial reorganisation of host organelles and cytoskeleton around the parasitophorous vacuole is observed within minutes following entry and it almost certainly plays a role in parasite nutrient acquisition. It is thus unlikely that the parasite, once intracellular, is experiencing starvation. We have produced a TgAtg3 conditional mutant that appears unable to regulate the conjugation of TgAtg8 and thus is likely deficient in autophagy. TgAtg3-depleted parasites invade host cells, and divide once or twice. Therefore TgAtg3-dependent autophagy does not seem to be essential in early phase of infection; however parasites cease to grow rapidly, indicating an important role for continued intracellular development. As for extracellular tachyzoites, once they have egressed their host cell, their fate is highly dependent on their motility and their ability to quickly invade a neighbouring cell or tissue, and although little is known about the exact environmental conditions encountered within its vertebrate host, autophagy could also be promoting the ability of T. gondii to survive during the journey of the parasite within the host tissues and organs.
Autophagy can occur in response to cellular stresses such as starvation, but there is growing evidence for more specific needs for autophagy to maintain cell homeostasis during normal growth and development. This can be directed towards specific organelles such as the mitochondrion (mitophagy) or the peroxisomes (pexophagy) [48]. This type of selective autophagy is used by cells to dispose of damaged organelles, or to clear the cell from these organelles when undergoing differentiation. In that context, we thought that there could be a targeted role for autophagy in the intracellular development of Toxoplasma. Some of the organelles of the mother cell are made redundant by their de novo formation in the daughters (i.e. rhoptries, micronemes) and we hypothesized that these might be degraded and recycled through autophagy. Interestingly, the use of PI3K inhibitor 3-MA on intracellular tachyzoites has been recently shown to affect parasite division, particularly daughter bud formation [49]. Also, morphological observations of 3-MA-treated parasites showed considerable retention of cellular material in residual body-like structures during daughter cell formation. It is to note that the authors showed that wortmannin did not produce similar effects, although in our hands wortmannin was a more potent inhibitor of autophagy than 3-MA on extracellular parasites, however wortmannin is notoriously unstable in culture over long periods of time and could have been degraded in the 20 hours long incubation used in their protocol, thus preventing lasting effects [50].
When depleting the parasites of TgAtg3, no accumulation of organelles or morphological alteration of de novo synthesized organelles was observed, suggesting that TgAtg3-dependent autophagy is not required for the recycling of secretory organelles from the mother. Nevertheless, TgAtg3 depletion has a significant cellular phenotype: the structural and functional alteration of the single mitochondrion present in the tachyzoites.
Is the break up of the mitochondrion that we observe in this mutant a cause or a consequence of the growth arrest of the tachyzoites? Although mitochondria are generally considered the powerhouse of the cell, many intracellular parasites rely heavily on glycolysis for energy production and do not require oxidative phosphorylation. The relative importance of the mitochondrion for energy generation in Apicomplexa is disputed, although the presence of a mitochondrial membrane potential (detected with Mitotracker labelling for instance) suggests it is at least partly functioning. However the other, often overlooked, metabolisms possibly hosted by the mitochondrion or in close interaction with the neighbouring apicoplast (i.e. redox metabolism, heme synthesis) are essential for the survival of the tachyzoites. On the other hand, the mitochondrion plays an active role in mediating apoptosis-like cell death, even in unicellular eukaryotes such as yeast [51], but this area remains largely unexplored in Toxoplasma. When treating intracellular tachyzoites for up to two days with dihydrofolate reductase inhibitor pyrimethamine [52], we could observe that cells having already lost their overall morphology, their normal micronemal distribution and displaying a fragmented nuclear content, still retained a reticulated mitochondrial signal (Figure S10), indicating that mitochondrial fragmentation is not necessarily an early event in the course of tachyzoite cell death. Moreover, when TgAtg3-depleted parasites (i.e. after 4 days of ATc treatment) were left to invade hot cells, parasites with a fragmented mitochondrion were found intracellularly after short invasion times (15 minutes), demonstrating that, in spite of the perturbation of mitochondrial homeostasis, these tachyzoites had retained their invasive potential (data not shown). Along with this, a recent work has been illustrating that glycolysis, but not mitochondrion-associated oxidative phosphorylation, was the major contributor to ATP production and was necessary to maintain the invasive potential of extracellular tachyzoites in carbon-depleted medium [53].
Mitochondrial dynamics and mitochondria turnover are linked. In yeast, but also mammalian cells, mitochondria undergo fusion/fission events during the cell cycle, for exchange of metabolites or DNA [54]. During these events, “maintenance” mitophagy is used to eliminate damaged mitochondrion and is intimately linked with fission events [55], which lead to perturbation of mitochondrial membrane potential. This way, depolarised mitochondria are eliminated and autophagy contributes to normal mitochondria homeostasis [56]. The replication and dynamics of T. gondii mitochondrion have been poorly studied; it is not known how the organelle elongates and to which extend fusion/fission events occur in the parasite. A recent report [57] has shown that the replication of the organelle was tightly coupled with the cell division cycle and that the mitochondrion was elongating after daughter scaffold formation, before entering the daughter cells at a very late stage. The appearance of the autophagosomal structures we have observed in Toxoplasma tachyzoites seems to be within the timeframe of mitochondrion replication/division. Consistent with this, recent cell-cycle microarray data [58] searchable through ToxoDB release 6.2 (www.toxodb.org) have revealed that, although TgAtg8 levels seemed to vary little during the cell cycle, autophagosome-conjugating partners TgAtg3 and TgAtg7 had similar profiles and were predominantly expressed towards the end of the G1 phase. This is compatible with an increase in autophagosomal activity after this part of the cell cycle, with a peak during the end of the S phase and the mitosis. Our microscopic observations also suggest the occurrence of mitophagy in the parasites. It thus seems that maintaining mitochondrial homeostasis during the biogenesis of this organelle in tachyzoites necessitates TgAtg3-dependent autophagy, and that loss of this autophagic function could lead to mitochondrion break up, loss of membrane potential and cell death. Interestingly, a recent report [59] has shown that a mammalian Atg3 null mutant also displays increased levels of fragmented mitochondria, and that Atg3 regulates mitochondrial homeostasis through an association with autophagy protein Atg12 (which appears to be absent from T. gondii's genome). Another recent study has suggested that cell death induced by starvation in autophagy-defective yeast mutants was caused by a dysfunction of the mitochondria [60]; more precisely, it showed that autophagy mutants accumulated high levels of reactive oxygen species and experienced defects in their respiratory functions.
In summary, our study shows that autophagy is present and functioning in the apicomplexan parasite T. gondii and that protein TgAtg3, through its role in TgAtg8 conjugation to the autophagosome, is likely essential for autophagy. Moreover, its function is crucial for maintaining mitochondrial homeostasis and for parasite growth. Many questions remain unanswered regarding the physiological roles of autophagy for T. gondii and Apicomplexa in general, or the machinery they use for this particular cellular function. For instance, the drastic phenotype we have observed with TgAtg3-depleted parasites is not exclusive and autophagy could still have a role in regulating parasite organelles other than the mitochondrion in specific physiological conditions. It is for instance possible that autophagy plays a role for organelle clearance during conversion from a parasitic stage to another. In related Apicomplexa Plasmodium, sporozoite to trophozoite conversion in the liver involves quite an extensive clearance of superfluous organelles and shape remodelling, in which autophagy is possibly playing a role [17]. Infectious stages of T. gondii, sporozoites, tachyzoites, and bradyzoites, are rather similar ultrastructurally but differ in certain organelles (for instance tachyzoites have more dense granules than bradyzoites, but have less micronemes and amylopectin granules) as well as their metabolic state and could be dependent on autophagy for differentiation and cell remodelling. Also, the role of autophagy during oocyst sporulation should be investigated. Again, in vivo studies using the TgAtg3 conditional mutant we have generated could reveal interesting features.
Also, one important question concerns the degradation and recycling of autophagocytosed material in Toxoplasma. Indeed, once completed, the double-membrane autophagosome is transported to a hydrolases-containing compartment and the outer membrane of the vesicle fuses with this compartment for degradation of its content. Typically, this lytic compartment is the lysosome in mammalian cells (or the vacuole in yeast), but so far the presence of such a compartment has remained elusive in T. gondii tachyzoites. Recent findings have nonetheless identified a compartment that seems to bear several characteristics of a bona fide late endosomal/lysosomal compartment (i.e. acidified, containing a peptidase) [13], [14], however this compartment is changing both in aspect and contents during the tachyzoite cell cycle, which renders it difficult to grasp. Moreover, the interactions between this compartment and autophagosomes are supposedly transient, so the interplay between the two would require dynamic studies on live cells. The repertoire of hydrolases present in the lytic compartment of tachyzoites and the ones particularly involved in the degradation of autophagic material also remain to be characterised.
Our discovery that autophagy protein TgAtg3 is essential for intracellular development of the parasite opens a new area for looking into possible parasitic drug targets, especially given the apparent peculiarities in the regulation of the parasite autophagic machinery compared with its host counterpart and the fact that this machinery contains enzymes (kinases, peptidases) for which inhibitors could be screened.
This study was conducted according to European Union guidelines for the handling of laboratory animals and the immunisation protocol for antibody production in rabbits was conducted at the CRBM animal house (Montpellier) and approved by the Committee on the Ethics of Animal Experiments (Languedoc-Roussillon, Montpellier) (Permit Number: D34-172-4, delivered on 20/09/2009).
Tachyzoites of the RHΔHX strain, deleted for hypoxanthine guanine phosphoribosyl transferase [42] or ΔKu80 strain [43] and derived transgenic parasites generated in this study, were propagated in vitro under standard procedures by serial passage in human foreskin fibroblasts (HFF) monolayers in Dulbecco's modified Eagle medium (DMEM, with 4500 mg/l D-Glucose, sodium pyruvate, Invitrogen) supplemented with 10% fetal bovine serum and 2 mM L-glutamine.
Total RNA was isolated from T. gondii tachyzoites using the RNAqueous kit (Ambion), according to the manufacturer's instructions. cDNA was synthetized from the isolated RNAs by reverse transcription using random hexamers and the SuperScript II kit (Invitrogen) or using the SMART RACE cDNA Amplification Kit (Clontech Laboratories). DNA corresponding to T. gondii Atg8 orthologue (TGME49_054120, http://toxodb.org) was obtained by PCR from cDNA with primers ML303/ML304 (see Table S2 for primer sequences) bearing the PstI and PacI restriction sites, respectively. The fragment was cloned into the pTGFP vector [61] to bear the pGFP-TgAtg8 plasmid for expression of Atg8 in Toxoplasma with the green fluorescent protein (GFP) fused at its N-terminus.
C-terminal glycine mutant version of the GFP-TgAtg8 construct was obtained by site-directed mutagenesis with the QuikChange mutagenesis kit (Stratagene), with primers ML308/ML309 to yield plasmid pGFP-TgAtg8-G/A with the C-terminal glycine mutated to an alanine. All constructs were checked by sequencing.
To generate stable transformants, 5×107 extracellular tachyzoites of the of the RHΔHX strain were transfected and selected as previously described [42]. GFP-TgTgAtg8-expressing parasites were obtained by electroporation of 100 µg of plasmids for the expression of GFP-TgTgAtg8 or its mutated version. After overnight growth, transfectants were selected with 25 µg/ml mycophenolic acid and 50 µg/ml xanthine for three passages, before cloning by limiting dilution under drug selection. After expanding the clones, GFP-expressing parasites were selected by observation with a fluorescent microscope.
A DNA sequence corresponding to the full TgAtg8 protein was obtained by PCR from T. gondii RH tachyzoites cDNA with primers ML697 and ML698. It was then cloned into pGEX-4T-3 (GE healthcare) and the construct was transformed into E. coli BL21 cells to produce a recombinant protein with an N-terminal glutathione-S transferase tag, which was used to immunise a rabbit. The antibody was subsequently used at 1/500 for Western blot or IFA.
Two approaches were used for knock-out of autophagy-related genes by double homologous recombination events. First, a plasmid bearing 5′ and 3′ untranslated regions (UTR) of the gene of interest flanking a selection cassette was generated. 5′ and 3′ UTR were obtained by PCR. They were cloned on either side of the chloramphenicol acetyl transferase (CAT) gene in the pTub5/CAT vector, serving as a selection marker [62]. Primer pairs used for PCR amplification were ML358/ML355 ML339/ML340 for 5′ and 3′UTR, respectively, of TgAtg3.
Second, cosmid with larger flanking regions to increase the frequency of homologous replacement was generated. Cosmid recombineering was performed as described previously [45]. Briefly, a cosmid overlapping TgAtg3 (TOXOU62) was recombineered with a cassette bearing a selection marker and obtained from plasmid template pH3CG by PCR, with primers ML537/ML538. DNA constructs for gene replacement were transfected into either RHΔHX or ΔKu80 strains and selected with the appropriate antibiotic.
For conditional knock-out strategy, an ectopic copy of TgAtg3 was introduced under the dependence of a SAG1 promoter. Two alternative start codons (see Figure S5) were tried for the constructs, corresponding to a long (amplified with ML454/ML456) or shorter (ML455/ML456) version of the TgAtg3. Constructs were introduced into the TAti tet-transactivator cell line by stable transformation [44]. For repression of the expression of the extra-copy, anhydrotetracycline (Clontech) was put at 1.5 µg/ml in the culture medium for two to four days.
Correct disruption of the TgAtg3 locus was verified by PCR using primers ML595/ML654 (PCR1), ML386/ML387 (PCR2), ML595/ML596 (PCR3). Primers ML650/ML651 and ML648/ML649 were used to generate the 5′ and 3′ probes, respectively, used for Southern blot analysis.
To induce autophagy, extracellular tachyzoites were put in starvation conditions. Extracellular parasites were obtained from freshly lysed HFFs, sedimented by centrifugation and washed twice in Hank's Balanced Salt Solution (HBSS) before being resuspended in pre-warmed HBSS and incubated at 37°C for up to 16 h.
Autophagy was occasionally modulated by incubation of the cells with several effectors: PI3K inhibitors Wortmannin and 3-methyladenine (Sigma) at 10 µM and 10 mM, respectively; TOR kinase inhibitor rapamycin (Santa Cruz), at concentrations up to 5 µg/ml.
Autophagosomes were quantified in live or paraformaldehyde-fixed GFP-TgAtg8 expressing parasites, by microscopic observation and counting of the punctate GFP signals. At least 200 cells were counted in each experimental set. Alternatively, the presence of the lipidated, autophagomal membrane-associated, form of GFP-TgAtg8 was assessed by Western blotting with anti-GFP antibody after separation by urea SDS-polyacrylamide gel electrophoresis (see below). Parasites extracts were normalised on counts of viable parasites (by trypan blue assay) at the end of the incubation time.
3.107 GFP-TgAtg8 tachyzoites starved for 8 hours in HBSS were solubilised in 1 ml of Tris HCl 50 mM pH 7.5 and sonicated twice for 30 seconds. Cellular debris were removed by centrifugation at 500 g for 10 minutes. The supernatant was submitted to an ultracentrifugation at 100 000 g for 30 minutes to yield a membrane-enriched high speed pellet and high speed supernatant soluble fractions, respectively. The supernatant fraction was TCA-precipitated and extracts were resuspended in SDS-PAGE loading buffer prior to Western blot analysis.
Alternatively, the high speed pellet was further extracted by 1 M NaCl, 2 M urea or 1% deoxycholate (DOC) for 4 hours at 4°C and submitted to another ultracentrifugation to yield a pellet and supernatant fraction.
Western blots were performed as described previously [63], with the modification that urea was included at a concentration of 6 M in the SDS-polyacrylamide gel to separate lipidated and non-lipidated forms of GFP-TgAtg8. The primary antibodies used for detection and their respective dilutions were: anti-GFP monoclonal mouse antibody (Roche) at 1/500, and anti-ROP5 [64] at 1/1000 as a loading control.
One 75 cm2 flask of HFF was grown for 24 hours with 0.1% fetal bovine serum, in the presence of 120 µCi of [1–3H] Ethan-1-ol-2-amine hydrochloride (GE Healthcare). HFFs were then infected for 24 hours with 5.107 GFP-TgAtg8 parasites; HFF layer was scrapped and parasites were syringed out. Isolated parasites were washed once in HBSS and incubated for 8 hours in 10 ml of HBSS, still in the presence of 120 µCi of labelled ethanolamine. They were washed once in HBSS and the pellet was resuspended in 1.5 ml of lysis buffer (PBS with 1% Nonidet 40 (NP40), 0.5% DOC and 0.1% SDS with a protease inhibitors cocktail (Roche)) and incubated at 4°C for 1 hour. The lysate was centrifuged for 20 min at 15 000 g and the supernatant was collected for subsequent immunoprecipitation.
Protein A-sepharose beads (Sigma) were prepared by putting together 50 µl of polyclonal rabbit anti-TgAtg8 antibody with 20 µl of beads for 1 hour. They were then washed 3 times in PBS to eliminate unbound antibodies.
1.5 ml of radiolabeled lysate was incubated with protein A-bound anti-TgAtg8 antibody overnight at 4°C under gentle agitation, and then washed five times with lysis buffer. Remaining buffer was discarded and the beads were resuspended in 20 µl of SDS-PAGE loading buffer and boiled for 5 minutes before analysis by urea SDS-PAGE. The gel was treated with Amplify (GE healthcare), dried and used for fluorography.
As a control, 5.107 parasites were treated in a similar way except that no radioactive ethanolamine was used during growth and starvation and urea SDS-PAGE, followed by Western blot analysis with anti-GFP antibody, were used after immunoprecipitation.
IFAs were performed either on extracellular tachyzoites recovered from freshly lysed HFF, or intracellular parasites at their various stages of development. They were fixed in 4% (w/v) paraformaldehyde in PBS and processed for immunofluorescent labelling as described previously [63], with the modification that the extracellular parasites were made to adhere onto poly-L-lysine slides for 20 minutes prior to processing for immunofluorescent labelling.
The following antibodies were used at 1/1000 dilution unless mentioned: anti-mitochondrial F1 beta ATPase (P. Bradley, unpublished), anti-mitochondrial HSP28 [65], anti-acyl carrier protein [66], anti-MIC3 [67], anti-ROP5 [64], anti-c-myc at 1/250 (Santa Cruz).
For co-labelling with fluorescent markers in live cells, constructs allowing the expression of IMC1 fused to the Tomato variant of RFP (B. Striepen, unpublished) and GRASP-RFP [57] were transfected in GFP-TgAtg8-expressing tachyzoites.
Fluorescent labelling of the mitochondrion was performed on extracellular parasites using MitoTracker Red CMXRos (Invitrogen) at 50 nM for 30 minutes at 37°C. Tachyzoites were then washed extensively in HBSS, fixed in 4% (w/v) paraformaldehyde in PBS and adhered onto poly-L-lysine slides before microscopic observation. For labelling of the mitochondrion in intracellular parasites, MitoTracker Red CMXRos was used at 500 nM for 45 minutes at 37°C and chased for 15 minutes 37°C before cells were processed for immunolabelling and microscopic observation.
Slides were mounted with Immumount (Calbiochem) and observed either with a Leica DMRA2 microscope, and images acquired with a MicromaxYHS1300 camera (Princeton Instruments) using the Metamorph software (Molecular Devices) or with an Axiovert/200M Zeiss inverted microscope equipped with an Axiocam MRm CCD camera (Zeiss) driven by the Axiovision software. Image acquisition was performed on workstations of the Montpellier RIO Imaging facility.
Parasite pellets were fixed for 2 hours with 2.5% glutaraldehyde in 0.1 M Na cacodylate buffer pH7.2, washed in buffer, post fixed in 1% OsO4 in the same buffer for 2 hours. After dehydration with graded ethanol series followed by propylene oxide, they were embedded in Epon. Ultrathin sections were prepared with a Leica ultracut E microtome, stained with Uranyl acetate and lead citrate and observed with a JEOL 1200E electron microscope. Immuno-electron microscopy on ultrathin cryosections was performed as described elsewhere [68] using anti-GFP antibodies on GFP-TgAtg8 transfected parasites.
Confluent monolayers of HFF grown in 6-well plates were infected with ∼50 tachyzoites per well and incubated for 6–7 days at 37°C. They were then fixed in cold methanol for 20 minutes and stained with Giemsa stain. Images were obtained with an Olympus MVX10 macro zoom microscope equipped with an Olympus XC50 camera. Plaque area measurements were performed with CellA software (Olympus).
TgAtg8 (TGME49_054120, http://toxodb.org); TgAtg3 (TGME49_036110, http://toxodb.org)
|
10.1371/journal.ppat.1000853 | Three Members of the 6-cys Protein Family of Plasmodium Play a Role in Gamete Fertility | The process of fertilization is critically dependent on the mutual recognition of gametes and in Plasmodium, the male gamete surface protein P48/45 is vital to this process. This protein belongs to a family of 10 structurally related proteins, the so called 6-cys family. To identify the role of additional members of this family in Plasmodium fertilisation, we performed genetic and functional analysis on the five members of the 6-cys family that are transcribed during the gametocyte stage of P. berghei. This analysis revealed that in addition to P48/45, two members (P230 and P47) also play an essential role in the process of parasite fertilization. Mating studies between parasites lacking P230, P48/45 or P47 demonstrate that P230, like P48/45, is a male fertility factor, consistent with the previous demonstration of a protein complex containing both P48/45 and P230. In contrast, disruption of P47 results in a strong reduction of female fertility, while males remain unaffected. Further analysis revealed that gametes of mutants lacking expression of p48/45 or p230 or p47 are unable to either recognise or attach to each other. Disruption of the paralog of p230, p230p, also specifically expressed in gametocytes, had no observable effect on fertilization. These results indicate that the P. berghei 6-cys family contains a number of proteins that are either male or female specific ligands that play an important role in gamete recognition and/or attachment. The implications of low levels of fertilisation that exist even in the absence of these proteins, indicating alternative pathways of fertilisation, as well as positive selection acting on these proteins, are discussed in the context of targeting these proteins as transmission blocking vaccine candidates.
| Sexual reproduction for malaria parasites is an essential process and is necessary for parasite transmission between hosts. Fertilisation between female and male gametes occurs in the midgut of the mosquito and proteins on the surface of gametes are principle targets in transmission blocking strategies. Despite their importance, relatively little is known about the malaria proteins involved in fertilisation. In this study we show that two gamete proteins, one expressed on the surface of males, the other on the surface of females, have important roles in the mutual recognition and attachment of gametes. Mutant parasites that lack the presence of these surface proteins show a strong reduction in fertility. Comparison of these gamete surface proteins in different malaria parasites showed that these proteins are evolving rapidly either across their length or at discreet regions/domains. We found, that despite the drastic reduction in zygote formation, low levels of fertilisation can still occur in the absence of these surface proteins, indicating that gametes can use alternative proteins to recognize each other. Both genetic variation of gamete surface proteins and the presence of different fertilisation pathways have important implications for transmission blocking vaccines targeting gamete surface proteins.
| Sexual reproduction is an obligate process in the Plasmodium life cycle and is required for transmission of the parasites between the vertebrate and mosquito hosts. The sexual phase is initiated by the formation of male and female cells (gametocytes) in the blood of the vertebrate host. Gametocytes are the precursors to the haploid male and female gametes that are produced in the mosquito midgut where fertilisation takes place. Successful fertilisation requires an ordered series of gamete-gamete interactions, specifically, the recognition of and adhesion to the female gamete by the motile male gamete, followed by a cascade of signalling events resulting from the fusion of the two gametes.
Despite their fundamental importance, relatively little is known about gamete receptors/ligands and their involvement in the process of gamete interactions of eukaryotes [1], [2], which is partly due to their rapid evolution and species-specific characteristics [3]. In Plasmodium the involvement of two gamete specific surface proteins P48/45 and HAP2/GCS1 has been demonstrated in male fertility and these proteins are to date the only known proteins with a demonstrable role in gamete-gamete interaction [4], [5], [6]. Parasites lacking P48/45 produce male gametes that fail to attach to fertile female gametes [4] while male gametes lacking of HAP2/GCS1 do attach to females, but they do not fuse due to an absence of membrane fusion between the two gametes [5]. P48/45 is one member of a family of proteins encoded within the genome of Plasmodium and this family is characterised by domains of roughly 120 amino acids in size that contain six positionally conserved cysteines (6-cys). The 6-cys family of proteins appears to be Apicomplexan specific and has a predicted relationship to the SAG proteins in Toxoplasma gondii [7], [8], [9], [10], [11]. Ten members of the 6-cys family have been identified. Most members are expressed in a discrete stage-specific manner in gametocytes, sporozoites or merozoites [8], [12], [13], [14], [15], [16]. The surface location of members of this family and their expression in gametes or in invasive stages (sporozoites and merozoites) suggests that they function in cell-cell interactions as has been shown for P48/45 in gamete adhesion. In addition to P48/45, five other 6-cys genes are transcribed in gametocytes, three of which (p230, p230p and p47) are exclusively expressed in the gamete stages of the malaria parasite [4], [8], [10], [12], [16], [17], [18], [19], indicating that these members of the gene family may also play a role in the process of gamete recognition and fertilisation. Indeed specific antibodies against the sexual stages of the human parasite Plasmodium falciparum, P48/45 and P230 can prevent zygote formation and thus block transmission of the parasite [19], [20], [21], [22], [23], [24], [25], [26]. Interestingly, P. falciparum mutants lacking P230 expression produce male gametes that fail to attach to erythrocytes resulting in a reduced formation of the characteristic ‘exflagellation centres’ and reduced oocyst formation in mosquitoes [27]. In order to investigate the role of the 6-cys proteins in parasite fertilisation we performed genetic and functional analysis on the five 6-cys proteins that are expressed in gametocytes. In this paper, we present evidence that in addition to P48/45, two 6-cys members (P230 and P47) also have an essential role in parasite fertilization. Interestingly, in P. falciparum evidence has been published that P48/45, P47 and P230 are under positive selection resulting in non-neutral sequence polymorphisms [28], [29], [30], [31]. By sequence analysis, we provide evidence that these three 6-cys proteins are undergoing strong but different rates of positive selection, either as a consequence sexual-selection driven by the competition between gametes or from natural selection exerted by the adaptive immune system of the host on proteins expressed in gametocytes.
The gametocyte-producer clone cl15cy1 (HP) of P. berghei ANKA was used as the reference parasite line [32]. In addition, the following mutant lines of the ANKA strain were used: 2.33, a non-gametocyte producer (NP) line [33] and 137cl8 (RMgm-15, www.pberghei.eu), a mutant lacking expression of P48/45 [4].
To disrupt genes encoding different members of the 6-cys family, we constructed a number replacement constructs using plasmid pL0001 (www.mr4.com) which contains the pyrimethamine resistant Toxoplasma gondii (tg) dhfr/ts as a selectable-marker cassette (SC). Target sequences for homologous recombination were PCR amplified from P. berghei genomic DNA (ANKA, cl15cy1) using primers specific for the 5′ or 3′ end of the different 6-cys genes (see Table S1 for the sequence of the different primers). The PCR–amplified target sequences were cloned in plasmid pL0001 either upstream or downstream of the SC to allow for integration of the construct into the genomic target sequence by homologous recombination. DNA constructs used for transfection were obtained after digestion of the replacement constructs with the appropriate restriction enzymes (Table S1). Replacement constructs pL1138 (p47) and pL0123 (p36), were constructed using replacement plasmid pDB.DT∧H.DB [34] and plasmid pL0121 (p47&48/45) was constructed in the previously described replacement plasmid for disruption of pb48/45 (plasmid p54 is renamed here to pL1137; [4]). This plasmid was made by exchanging the 5′ pb48/45 targeting sequence with the 5′ targeting sequence of pb47. The p230pII replacement construct pL0120 is a derivative of plasmid pL0016 [35] containing the tgdhfr-ts SC, gfp (under control of the pbeef1aa promoter and 3′UTR of pbdhfr/ts) and p230p 5′ and 3′ targeting sequences [36]. Transfection, selection and cloning of mutant parasite lines were performed as described [32], [37] using P. berghei ANKA cl15cy1 as the parent reference line. For all mutants with an observable phenotype, mutants were generated and selected in two independent transfection experiments (Table S1). Of each transfection experiment we selected one cloned line for further genotype and phenotype analysis. Correct integration of the construct into the genome of mutant parasites was analysed by standard PCR analysis and Southern blot analysis of digested genomic DNA or of FIGE separated chromosomes [32]. PCR analysis on genomic DNA was performed using specific primers to amplify either part of the wild type locus (primers WT1 and 2) or the disrupted locus (primers INT1 and 2). See Table S2 for the sequence of these primers.
Total RNA was isolated from the different blood stage parasites of the gametocyte-producer clone cl15cy1 of P. berghei ANKA (HP), the non-gametocyte producer line 2.33 (NP) and the different mutant lines according to standard methods. To determine stage-specific transcription of the 6-cys family members, Northern blots containing RNA from different blood stages were hybridised with different gene specific probes, which were PCR-amplified using the primers shown in Table S2 (primer pairs WT1+ 2). To detect expression of the P48/45 protein we used polyclonal antiserum raised against recombinant P. berghei P48/45 as described [4]. For detection of P47 we generated the following polyclonal antiserum; a fragment of the Pb47 ORF (encoding amino acids 80–411) was PCR-amplified using primers L964 and L965 (Table S2) and cloned into the NdeI/BamHI sites of the expression vector pET-15b (Novagen) providing an N-terminal 6-Histidine tag. Polyclonal antiserum was raised in New Zealand rabbits by injection of 200 µg of gel-purified recombinant protein. Boosting was carried out subcutaneously with 3-weeks intervals using 200 µg protein in incomplete Freund's adjuvant. Serum (P47) obtained 2 weeks after the third boost was immuno-purified on immobilised purified recombinant P47. To detect P48/45 and P47 in the different mutant lines, total protein samples of purified gametocytes were fractionated on non-reducing 10% SDS polyacrylamide gels.
The fertility of wild type and mutant gamete populations was analysed by standard in vitro fertilisation and ookinete maturation assays [4], [17] from highly pure gametocyte populations [38]. The fertilisation rate of gametes is defined as the percentage of female gametes that develop into mature ookinetes determined by counting female gametes and mature ookinetes in Giemsa stained blood smears 16–18 hours after in vitro induction of gamete formation. Fertility of individual sexes (macro- and micro-gametes) was determined by in vitro cross-fertilisation studies in which gametes are cross-fertilised with gametes of lines that produce only fertile male (Δp47; 270cl1) or only fertile female gametes (Δp48/45; 137cl1 [4], [17], [39]. All fertilisation and ookinete maturation assays were done in triplicate on multiple occasions in independent experiments. In vivo ookinete, oocyst and salivary gland sporozoite production of the mutant parasites were determined by performing standard mosquito infections by feeding of Anopheles stephensi mosquitoes on infected mice [40]. Oocyst numbers and salivary gland sporozoites were counted at 7–10 days and 21–22 days respectively after mosquito infection. For counting sporozoites, salivary glands from 10 mosquitoes were dissected and homogenized in a homemade glass grinder in 1000µl of PBS pH 7.2 and sporozoites were counted in a Bürker-Türk counting chamber using phase-contrast microscopy [41]. Infectivity of sporozoites was determined by infecting mice through bites of 25–30 infected mosquitoes at day 21–25 after mosquito infection.
The formation of exflagellation centres (i.e. male gamete interactions with red blood cells) was determined by adding 10µl of infected tail blood to 100–300 µl of standard ookinete culture medium pH 8.2 to induce gamete formation. Ten minutes after induction of gamete formation a droplet of 5–10 µl was placed on a cover slip and analysed under a standard light microscope (40× magnification) as a hanging-drop using a well slide. When red blood cells were settled in a monolayer, the number of exflagellating male gametocytes was counted that form or did not form exflagellation centres. An exflagellation centre is defined as an exflaggelating male gametocyte with more than four tightly associated red blood cells [27]. The formation of exflagellation centres was performed using tail blood collected at day 6 or 7 from mice that were infected with 105 parasites without treatment with phenylhydrazine. For quantification of male-female interactions tail blood was collected from phenylhydrazine-treated mice with high numbers of gametocytes [42]. Tail blood (10µl) was collected at gametocytemias ranging between 4–8% and added to 100µl of standard ookinete culture medium pH 8.2 to induce gamete formation. Ten minutes after induction of gamete formation, the cell suspension was placed in a Bürker-Türk counting chamber and during a period of twenty minutes the male-female interactions were scored using a phase-contrast light microscope at a 40× magnification. Attachments of males to females were scored if the male had active (attachment-) interactions with the female for more than 3 seconds. Penetration of a female by the male gamete was scored as a fertilisation event.
Pairwise alignments were generated between the orthologous sequences of p48/45, p47 and p230 genes in P. berghei, P. yoelii and P. chabaudi; sequences were obtained from PlasmoDB (http://www.plasmodb.org version 6.1; see Table S3 for the accession numbers of the 6-cys gene family members). Complete gene sequences for a number of these genes were obtained from the Sanger Institute (A. Pain, personal communication). Maximum-likelihood estimates of rates of non-synonymous substitution (dN) and synonymous substitution (dS) between pairwise alignments were generated using the PAML algorithm (version 3.14; [43], [44]) using a codon-based model of sequence evolution [45], [46], with dN and dS as free parameters and average nucleotide frequencies estimated from the data at each codon position (F3×4 MG model [47]). For this analysis we assumed a transition/transversion bias (i.e. kappa value) that had been estimated previously and found to be similar in case of P. falciparum and P. yoelii, i.e. 1.53 [48]. A sliding window analysis of dN/dS ratios was performed of p230, p47 and p48/45 from the three rodent parasites. We analysed the dN/dS values of these genes across their length by analysing sequentially 300bp of the gene in 150bp steps. This analysis is essentially the same as the calculation of π (i.e. the number segregating or polymorphic sites) described for p48/45 in distinct P. falciparum isolates described by Escalante et al. [29]. We obtained the single nucleotide polymorphisms (SNPs) data identified from field and laboratory isolates of P. falciparum (excluding all P. reichenowi SNPs) from PlasmoDB (www.PlasmoDB.org). The alignment of these SNPs along the different genes (to scale) was extracted from the Genome Browser page of PlasmoDB. The locations of the SNPs were aligned onto the schematic representation of the 6-cys genes of the rodent parasites. It should be noted that the alignment of the p230 gene of the different Plasmodium species was only possible around 1008bp after the putative start site. In order to determine which residues of p230, p47 and p48/45 genes were under positive selection in the rodent malaria parasites, a Bayes Empirical Bayes (BEB) analysis was performed using sequences from the 3 rodent genomes and was calculated as described in Yang et al. [49]. To test which genes were undergoing positive selection the likelihood ratio test (LRT) was performed using a comparison of site specific models of evolution [50], [51]. This test compares a ‘nearly neutral’ model (without any residues under positive selection) and a ‘positive selection’ model (with residues under positive selection and therefore under adaptive evolution). Both models assume that there are different categories of codons, which evolve with different speeds. The ‘nearly neutral’ model assumes two categories of sites at which amino acid replacements are either neutral (dN/dS = 1) or deleterious (dN/dS<1). The ‘positive-selection’ model assumes an additional category of positively selected sites at which non-synonymous substitutions occur at a higher rate than synonymous ones (dN/dS>1). Likelihood values indicate how well a model fits to the analyzed alignment and answers the question if the ‘positive selection’ model fits better to the analyzed alignment than the ‘nearly neutral’ model.
All animal experiments were performed after a positive recommendation of the Animal Experiments Committee of the LUMC (ADEC) was issued to the licensee. The Animal Experiment Committees are governed by section 18 of the Experiments on Animals Act and are registered by the Dutch Inspectorate for Health, Protection and Veterinary Public Health, which is part of the Ministry of Health, Welfare and Sport. The Dutch Experiments on Animal Act is established under European guidelines (EU directive no. 86/609/EEC regarding the Protection of Animals used for Experimental and Other Scientific Purposes).
Ten members of the 6-cys family have been identified in Plasmodium and are found in all Plasmodium species (Table S3). We analysed the transcription profile of the 10 members during blood stage development of P. berghei by Northern blot analysis and combined this analysis with a search of publicly available literature, transcriptome and proteome datasets. This method established that multiple members are transcribed in gametocytes of which four members, p48/45, p47, p230, p230p, are transcribed exclusively in the gametocyte stage (Fig. 1A). The gametocyte specific expression of p48 and p230p has been shown before [4], [8]. Transcription of p38 occurs both in gametocytes and in asexual blood stages as has also been reported [8], whereas p12 is transcribed in all blood stages. The relative weak band observed in gametocytes might be due to low contamination of the gametocyte preparation with asexual blood stages (gametocyte samples always contain a small degree of contamination with schizonts when density gradients are used for gametocyte purification). Transcription of p41 and p12p show a complex pattern of multiple transcripts in all blood stages. The close paralogue pair p36 and p36p have quite different transcriptional profiles: p36p is not transcribed in blood stages but transcription is exclusive to sporozoites [14], [15] whereas p36 is transcribed both in gametocytes (Fig. 1B; [8], [52]) and in sporozoites [14], [15].
Since no polyclonal or monoclonal antibodies exist for most of the 6-cys family members of P. berghei, except for P48/45 [4], P47 (this study) , P36 and P36p [14], data on expression of these proteins in different life cycle stages mainly comes from large-scale proteome analyses. For most members of the 6-cys family which have been detected by proteome analysis, the presence of the protein coincides with transcription of its gene (Fig. 1B). The exclusive presence of P48/45, P47, P230 and p230p in the proteomes of gametocytes corresponds to the transcription pattern of their respective genes. The presence of P48 and P47 in P. berghei gametocytes has been confirmed using polyclonal antibodies against these proteins (Fig. S1; [4]). P12, P38 and P41 have been detected in the proteome of merozoites which agrees with their transcription in the asexual blood stages and with their identification in the raft-like membrane proteome of the P. falciparum merozoite surface [13]. Also the presence of P36 in proteomes of both gametocytes and sporozoites [41], [52] and P36p in sporozoites [14], [41] fits with the transcription profile of these genes. Up to now only P12p has not been detected in any proteome of Plasmodium. Comparison of the transcription and expression patterns of the 10 conserved members of the 6-cys family of P. berghei with those of P. falciparum from large scale transcriptome and proteome analyses demonstrates that the expression patterns are conserved between the rodent and human parasite (Fig. 1B) and also confirms that four out of the 10 members are specific to the gametocyte stage.
We previously reported the functional analysis of mutant P. berghei parasites that were deficient in expressing P48/45, generated by targeted disruption of p48/45 through a double crossover homologous recombination event [4]. Here we have used the same approach, schematically shown in Fig. 2A, to disrupt 5 other members of the 6-cys family that are transcribed in gametocytes. We excluded p12, p12p, p41 and p36p from this analysis since the results obtained from transcriptome and proteome analyses indicate a role for the first three of these genes during the asexual blood stage development (Fig. 1B). We have previously demonstrated in both, P. berghei and P. falciparum, that P36p is involved in liver-cell infection and disruption of its gene had no effect on development of gametes and fertilisation [15], [53]. Mutant parasite lines have been generated deficient in P47 (Δp47), P230 (Δp230), P230p (Δp230p), P38 (Δp38) or P36 (Δp36) and for each gene, mutants were selected from two independent transfection experiments (Table S1). Two different Δp230p mutant lines were generated, Δp230p-I and Δp230p-II, differing in which regions of 230p have been disrupted. In mutant Δp230-I a fragment is deleted from the second 6-cys domain (i.e. first 894aa still present) onwards whereas in mutant Δp230-II the deleted fragment includes part of the first 6-cys domain (i.e. first 492 amino acids still present). In addition we generated a mutant line deficient in the expression of both P48/45 and P47 (Δp48/45&Δp47). Correct disruption of the target-genes was verified by diagnostic PCR analysis (Fig. 2B) and Southern blot analysis of separated chromosomes and/or digested genomic DNA (data not shown). To demonstrate that the mutant parasite lines were deficient in expression of the targeted gene we analysed transcription of the corresponding genes by Northern blot analysis using mRNA collected from purified gametocytes (Fig. 2B). No transcripts of p47 and p38 could be detected in ΔP47 and Δp38 mutants, and no p48/45 and p47 transcripts are present in the DKO mutant Δp48/45&Δp47. Only small, truncated transcripts were detected for p230 and p230p in gametocytes of the Δp230 and Δp230p lines and also in Δp36 a truncated p36 transcript was found. Full length transcripts of wt p230 and p230p are 8.5 and 9.5 kb respectively, whereas truncated transcripts are approximately 2.5 kb in size. Since several of the disrupted genes are organised as pairs within the genome (i.e. p230&p230p and p48/45&p47), we analysed whether disruption of one member of a pair affected transcription of the other gene. For Δp48/45 parasites it has been shown before that disruption of p48/45 had no effect on expression of its paralog P47 [4]. In this study we similarly show for p47, p230 and p230p that disruption had no effect on transcription of its paralogous member (Fig. S1 A&B). In addition to the transcription analysis of the disrupted genes, we analysed the presence or absence of the proteins P47 and P48/45 in the mutant parasites by Western analysis using polyclonal antiserum (Fig. S1C). P47 is present in wt gametocytes and gametocytes of the Δp48/45 but is absent in Δp47 and Δp48/45&Δp47 gametocytes. P48/45 is present in wild type and absent in the Δp48/45&Δp47 gametocytes.
We next analysed the phenotype of the different mutant lines during gametocyte and gamete development as well as during fertilisation, ookinete and oocyst formation using standard assays for phenotype analysis of the sexual- and mosquito stages of P. berghei. Surprisingly, three of the six mutants lacking expression of genes that are transcribed in gametocytes did not exhibit a phenotype that was different from wild type parasites during these stages of development. These mutants, Δp230p, Δp38 or Δp36, showed a normal growth of the asexual blood stage (data not shown), sexual development and development of the mosquito stages up to the mature oocysts (Table 1). All these mutant lines produced wild type numbers of gametocytes and gametes and showed normal fertilisation rates as measured by in vitro zygote/ookinete production (Table 1; Fig. 3). In contrast to the absence of a discernable fertilisation phenotype with the Δp230p, Δp38 and Δp36 mutants, we found that the capacity of fertilisation is severely affected in the other three mutants, (Fig. 3A). Specifically, Δp47, Δp230 and Δp48/45&Δp47 lines showed a fertilisation rate that was reduced by more than 99.9% compared to wt, as shown by the inhibition of zygote/ookinete production in vitro (Table 1; Fig. 3A). These mutants produced normal numbers of mature gametocytes during blood stage development. The analysis of in vitro gamete formation (exflagellation of males; emergence of female gametes from the erythrocyte) by light-microscopy also revealed that the process of gametocyte and gamete formation was not affected, resulting in the production of motile male gametes and female gametes, emerged from the host erythrocyte by more than 80% of the mature gametocytes (Table 1). At 16–18h after activation of gamete formation, the in vitro cultures of Δp47, Δp230 and Δp48/45&Δp47 lines contained many (clusters of) unfertilized, singly nucleated, female gametes. This phenotype of a strong reduction of fertilisation despite the formation of male and female gametes closely resembles the phenotype of Plasmodium parasites lacking P48/45 [4]. As had also been previously observed with the P48/45 deficient mutant, the fertilisation rate of gametes of the three mutant lines seems to be more efficient in the mosquito compared to in vitro fertilisation [4]. Compared to wild type parasites, the in vivo fertilisation of the mutants is reduced by 93–98% as calculated by ookinete and oocyst production in mosquitoes (Table 1), whereas the reduction of in vitro fertilisation rate is greater than 99.9%. Infections of naïve mice through bite of 20–30 mosquitoes infected with parasites of Δp47, Δp48/45&Δp47DKO and Δp230 parasites, resulted in blood stage infections containing only gene disruption mutants (i.e. mutant genotype and no ‘wild type’ parasites), as determined by PCR and Southern analysis of genomic DNA (results not shown). These results show that gametes of all three mutant lines still have a low capacity to fertilise, resulting in the production of viable and infective ookinetes, oocysts and sporozoites. Moreover, the results obtained with the double knock-out mutant Δp48/45&Δp47 indicate that the few fertilisation events in single knock-out mutants deficient in expression of either P47 or P48/P45 (this study and [4]) cannot be explained by a compensation effect due to its paralogous protein because the Δp48/45&Δp47 mutant still shows a comparable, albeit greatly reduced, ability to fertilise and to pass through the mosquito.
Fertility of the male and female gametes produced by the mutant lines can be determined by in vitro cross-fertilisation studies, where gametes are cross-fertilised with gametes of parasite lines that produce either only fertile male gametes or female gametes. Such an approach was used to establish that Δp48/45 parasites produced infertile male gametes, whereas the female gametes are completely fertile [4]. We performed different in vitro cross fertilisation experiments to determine whether the reduced fertilisation capacity of the Δp47 and Δp230 mutants was due to affected male gametes, female gametes or to both sexes. Gametes of both mutants were cross-fertilised with female gametes of Δp48/45 (males are infertile) to determine male fertility of Δp47 and Δp230. Male gametes of Δp47 were able to fertilise Δp48/45 females (at wild-type levels) whereas the males of Δp230 were unable to fertilise the Δp48/45 females (fertilisation rates <0.01%; Fig. 3B). These results demonstrate that male gametes of Δp47 are viable with wild type fertilisation capacity and therefore the fertilisation defect of Δp47 must be due to infertile females. The normal fertility of male gametes of Δp47 has also been shown in previous studies in which the males of this mutant have already been used in other cross-fertilisation studies [17], [39], [54], [55]. The lack of fertilisation in the crossing experiments of gametes of Δp230 with Δp48/45 shows that P230 plays a role in male fertility. In order to test the fertility of Δp230 females we crossed the gametes of this line with the fertile male gametes of Δp47 (as mentioned above the females are infertile). We find that Δp47 male gametes are able to fertilise Δp230 female gametes in a manner identical to their ability to fertilise Δp48/45 females (Fig. 3B). This demonstrates that female gametes of Δp230 have a fertility that is comparable to wild type female gametes and that the fertilisation defect is the result of infertile males. Crossing experiments performed with gametes of the double knockout mutant, Δp48/45&Δp47 with gametes of either Δp230, Δp47 or Δp48/45 did not result in increased fertilisation rates (<0.01%), demonstrating that gametes of both sexes are infertile in the double knock-out mutant (Fig. 3B).
In P. falciparum it has been shown that male gametes lacking P230 expression have a reduced capacity to adhere to red blood cells, as measured by the formation of ‘exflagellation centres’ [27]. We therefore examined the ability of P. berghei male Δp230 gametes to attach to erythrocytes, by microscopic examination of exflagellation centre formation under standardized in vitro conditions. In these experiments 76–92% of exflagellating wt males and 72–90% exflagellating Δp230 male gametocytes, formed such centres (Table 2), indicating that in contrast to P. falciparum Δp230 in P. berghei both wt and Δp230 male gametes have a similar ability to interact with red blood cells. Gametocytes that did not form exflagellation centres were often floating on/above the red blood cell layer during exflagellation. Further analysis of single, free male gametes of Δp230 revealed that they were highly motile and often attach to red blood cells, producing characteristic red blood cell shape deformations due to the active interactions between the male gamete and the erythrocyte. Male gametes lacking expression of P48/45 do not attach to female gametes as has been previously shown by analysing male-female interactions by light microscopy [4], [5]. We therefore analysed the interactions between male and female gametes of Δp230 or Δp47, between 10 and 30 minutes after induction of gamete formation using phase-contrast microscopy. In wt parasites attachment of males to females was readily detected with a mean of over 25 attachments during a 20 minutes period of observation, with a mean of more than 6 confirmed fertilisations (i.e. male gamete penetrations; Table 2). In preparations of gametes of both Δp230 and Δp47 not a single fertilisation event was detected and the number of male and female gamete attachments was drastically reduced (Table 2). We observed that while male gametes of both mutants undergo active interactions with red blood cells and platelets, attachment of males to female gametes are hardly ever observed. These results show that P230 like P48/45 is a male fertility factor involved in recognition or attachment to females and that P47 is a female fertility factor involved in recognition or adherence by the male gamete. Whether P48/45 and P230 once on the surface of the male gamete directly interact with P47 on the surface of the female gamete is unknown. Unfortunately, repeated immuno-precipitation experiments with anti-P. berghei P48/45 antibodies and wt gamete preparations, in order to identify interacting partners, were unsuccessful (data not shown).
Analyses of sequence polymorphisms of p48/45, p47 and p230 of laboratory and field isolates of P. falciparum has provided evidence that these proteins are under positive selection [28], [29], [30], [31]. We analysed synonymous (dN) and non-synonymous (dS) polymorphisms of p48/45, p47 and p230 by comparing these genes in three closely related rodent parasites P. berghei, P. yoelii and P. chabaudi by making use of the newly available gene sequences (www.PlasmoDB.org version 6.1). The updated dN/dS values for these genes obtained here, which is commonly used as an indicator of positive selection, were in all comparisons higher than the mean dN/dS value of all genes within the respective genomes (Table S4). However, only the dN/dS ratio of p47 in the P. berghei/P. yoelii comparison showed a significant difference with the mean dN/dS value (0.82 compared to the mean dN/dS of 0.26). Overall, P47 is in the top 4–6% of fastest evolving proteins in the rodent parasite genomes as compared to top 10–16% for P230 and 15–50% for P48/45 (Table S4). In addition, we have used the likelihood ratio test (LRT) to analyse if these genes were undergoing neutral or positive selection (see Materials and Methods). This test shows that p47 is indeed under positive selection (P = 0.006) when comparing the site/residue specific models of evolution.
We next examined sequence mutations in the same genes in more detail by performing a comparative dN/dS ratio analysis across these genes using small and corresponding regions of these genes using a ‘sliding window analysis’ (i.e. 300bp in 150bp intervals; Fig. 4; Table S4). This analysis showed that p47 has an exceptionally elevated dN/dS value (i.e. 1–2) in one area corresponding to the truncated B-type domain II as defined by [7]. Interestingly, although P230 had a relatively low overall dN/dS value (0.33–0.44), the sliding window analysis revealed that P230 contains several areas where the dN/dS ratio is higher than 1.0 with an increased ratio in all 3 species in particular around the B-type domain IV as defined by Gerloff et al. (2005). In order to analyse similarities in the location of sequence polymorphism between P. falciparum and the three rodent parasites, we aligned all known single nucleotide polymorphisms (SNPs) described for P230, P47 and P48/45 in P. falciparum (i.e. www.PlasmoDB.org; [56], [57], [58]) with the dN/dS ratios determined by the ‘sliding window analysis’ (for details see Materials and Methods; Fig. 4). Interestingly, the elevated dN/dS ratios of p47 domain II and domain IV of P230, both correspond with the location of high SNP densities in the orthologous P. falciparum genes. These findings would suggest that similar regions in the p47 and p230 genes of rodent parasites and P. falciparum are subject to positive selection. To predict which residues of the three P. berghei genes are under positive selection we performed a Bayes Empirical Bayes analysis (BEB; [49]). This analysis calculates dN/dS values (ω values) on each residue of a particular protein when the genes encoding these proteins are compared in least 3 similar species and an ω>1 indicates positive selection on a residue. For P47 ten residues were identified undergoing positive selection with ω values ranging between 4 and 7 (Table S5). Nine of these 10 residues are confined to the first two domains of P47 including the region B-type domain II. In P48/45 four residues were identified (ω values ranging between 1 and 2) and for P230 only one amino acid (ω = 1.3). Interestingly, this one residue in P230 (i.e. residue 845V) maps to the corresponding region of the P. falciparum P230, domain IV, where 6 of the 27 non-synonymous polymorphisms described by Gerloff et al. map (Table S4).
Until recently the only protein proven to play a direct role in merging of the male and female gamete of Plasmodium gametes in Plasmodium was P48/45, a surface protein principally of male gametes shown to play an essential role in recognition of and attachment to females [4], [5]. Recently, two studies have identified a second protein, HAP2/GCS1 with a role early in fertilisation [5], [6]. Male gametes of mutant parasites lacking this protein can attach to female gametes but the subsequent fusion of the gametes is absent [5], a process which is clearly after the mutual recognition and attachment of gametes. Our studies provide evidence for the direct involvement of two additional proteins, P47 and P230, which like P48/45 play a key role in the initial phase of gamete-gamete recognition and attachment. The phenotype of mutants lacking P230 expression is identical to the phenotype of mutants lacking P48/45, i.e. male gametes do not recognize and attach to female gametes whereas the female gametes are fertile. These results show that the P230 protein, like P48/45, is a male fertility factor. A similar role of P48/45 and P230 in male fertility is perhaps not surprising since evidence has been reported that both proteins interact with each other. Unlike P48/45, P230 does not contain a glycosylphosphatidylinositol (GPI) anchor and in P. falciparum evidence has been found that P230 forms a complex with P48/45 at the surface of gametocytes and gametes [18], [27], [59], [60]. Indeed, analysis of P. falciparum mutants has shown that in the absence of P48/45 the P230 protein is not retained on the surface of gametes, a result which may indicate that tethering of P230 to the surface of the male gamete is mediated by P48/45 [27]. In contrast, in the absence of P230 the surface location of P48/45 is not affected in P. falciparum [27], [61]. If in P. berghei the same interaction occurs, and Δp48/45 gametes also lack surface expression of P230, then the failure of Δp48/45 and Δp230 males to attach to females might be solely due to the absence of P230 on the male gamete surface. This would imply that P230 and not P48/45 is the major male protein that is responsible for recognition of and attachment to the female. However, it has been shown that antibodies directed against P48/45 strongly reduce oocyst formation [19], [20], [24], [25], [26], indicating that either P48/45 antibodies disrupt the attachment of the translocated P230 to P48/45 after gamete formation or it may play a more direct role in fertilisation and that its function is not exclusively as a membrane anchor for P230.
Interestingly, in P. falciparum it has been shown that male gametes with a disrupted p230 gene are incapable of interacting with erythrocytes and do not form the characteristic exflagellation centres and these mutants show a strong reduction in oocyst formation [27]. These observations, in P. falciparum, indicate that P230 not only plays a role in gamete-gamete interactions but male gamete interactions with erythrocytes may be required for gamete maturation resulting in an optimal fertilisation capacity [27], [62]. Our analyses of Δp230 P. berghei male gametes in live preparations did not reveal any difference in their capacity to interact with red blood cells, suggesting that there are functional differences between P230 of P. berghei and P. falciparum. As the interaction between male and female gametes has not been analysed in the P. falciparum Δp230 mutants it is unknown whether the decreased oocyst formation results from the reduced gamete-erythrocyte interactions or is due to the lack of gamete recognition and attachment, as we have observed in P. berghei. Therefore, further research is needed to unravel whether P. falciparum P230 is also involved in gamete-gamete interactions like P. berghei P230. Moreover, additional research is required to identify the proteins at the surface of P. berghei male gametes that are responsible for the adherence of the male gametes to erythrocytes. Disruption of the close paralogue of p230, p230p, did not have any effect on fertilisation or on red blood cell attachment. The distinct phenotypes of Δp230 and Δp230p gametes demonstrate that the proteins encoded by these genes are not functional paralogues that are able to complement each others function as has been demonstrated for the paralogous protein pair P28 and P25 on the surface of zygotes [63]. The same is true for the paralogous proteins P48/45 and P47 (see below) or P36 and P36p [15], [64].
In addition to the important role of P230 in male fertility, our studies demonstrate that P47 plays a key role in P. berghei female gamete fertility. Both proteome analyses of P. berghei gametocytes [17] and IFA analysis of P. falciparum gametocytes using anti-P47 antibodies [12] have shown the female-specific expression of P47. In P. falciparum, P47 is located on the surface of the female gametes following emergence from the host erythrocyte. Our studies demonstrate that P. berghei females lacking P47 are not recognized by wild type males. These observations may suggest that P48/45 or P230 on the male gamete directly interact with P47 on the female for recognition and attachment. However, P48/45 and P230 may alternatively interact with additional, as yet unknown protein/s on the surface of the female that are dependent on the presence of P47, in an analogous manner to the interaction between P230 and P48/45 on the surface of the male gamete. Both P48/45 and P230 are also expressed in the female gametes of P. berghei and P. falciparum [17], [27]. The presence of these proteins on the female gamete surface does not result from male proteins that are released by the male during activation and subsequent binding to the female since ‘pre-activated’ female gametocytes also express these proteins (B van Schaijk, personal communication and [65]. However, an essential role for P48/45 and P230 in female gametocytes is not implicated in P. berghei since both Δp230 and Δp48/45 females demonstrate normal fertilisation, i.e. to wild-type levels, when incubated with wild type males.
Unexpectedly, the lack of expression of P47 in P. falciparum mutants appears not to have a role in fertilisation as determined by oocyst formation in mosquitoes [12]. This difference between P. berghei and P. falciparum suggests that the proposed model of the interactions between male P48/45 and/or P230 with female P47 (and/or P47-interacting proteins) being key for the recognition and attachment of gametes does not hold true for all Plasmodium species. However, these differences between P. falciparum and P. berghei might also be explained by the presence of an additional set of protein ligands in both species that mediate additional mechanisms of gamete recognition and attachment. Indeed by analysing P. berghei Δp48/45 mutants [4] and mutants lacking expression of P47 and P230 (this study) we found that low levels of fertilisation did occur. Surprisingly, in all mutants significant higher fertilisation rates were observed in mosquito midguts compared to in vitro rates of fertilisation. Even in the mutant lacking expression of both P48/45 and P47, the same low fertilisation rates are observed. Assuming that P. berghei Δp48/45 gametes lack P230 surface expression as has been shown for P. falciparum Δp48/45, then gametes of the double knock-out mutant can fertilise in the absence of essentially all three fertility factors of the 6-cys family, albeit at a reduced rate. These observations indicate the presence of additional proteins that secure fertilisation in the absence of the three members of the 6-cys family. For unidentified reasons this alternative fertilisation pathway appears to be much more efficient in vivo than in vitro, suggesting that mosquito factors influence this alternative route of fertilisation. The observed oocyst formation in Δp48/45 and Δp47 P. falciparum parasites [4], [12] might therefore also be explained by this route of fertilisation and the presence of relatively high numbers of oocysts might indicate that this alternative pathway is more efficient in P. falciparum in A. stephensi compared to P. berghei in A. stephensi. Such alternative pathways of fertilisation may have implications for development of transmission blocking vaccines that block fertilisation using antibodies directed against members of the 6-cys family of proteins and therefore it is important to identify the additional proteins involved in the process of recognition and attachment of gametes. It is possible that other members of the 6-cys family that are expressed in gametocytes (P230p, P38 and P36) may be involved in the alternative pathways of fertilisation. Although we found that gametes lacking expression of these proteins did not show a significant reduction in fertilisation, the effect of their absence on gamete fertility may only become evident in the absence of P48/45, P47 and P230. Further research using mutants lacking multiple 6-cys members is required to reveal whether other 6-cys family members or other unrelated proteins play a role in alternative routes of fertilisation.
For P48/45, P47 and P230 in P. falciparum evidence has been published that these proteins are under differing rates of positive selection resulting in non-neutral sequence polymorphisms [28], [29], [30], [31]. Polymorphisms in gamete proteins may be a consequence of sexual selection as is the case for gamete proteins of other organisms [3], [66]. However, sequence polymorphism in these Plasmodium genes may also result from natural selection exerted by the adaptive immune system of the host. These three proteins are expressed in mature gametocytes, and as only a very small percentage of gametocytes ever get passed on to a mosquito, the vast majority of gametocyte proteins (including these 6-cys members) are eventually released into the hosts circulation where they are exposed to the host immune system. Indeed it has been shown that P48/45 and P230 both elicit humoral responses in infected individuals that can mediate transmission blocking immunity [22], [24], [67], [68], [69], [70]. Our analyses on dN/dS values of the three rodent parasites provide additional evidence that directional selection pressures affect sequence polymorphisms of gamete surface proteins, especially evident for the female specific p47 which belongs to the top 4–6% fastest evolving genes in the rodent parasite genomes. Analysis of dN/dS variation across the genes by the sliding window approach on P230 identifies one region that is evolving rapidly in all the rodent parasites and, interestingly, this correlates with the same region in P. falciparum (B-type domain IV) that has the highest density of SNPs [7]. The correlation of the location of P. falciparum SNP's with increased dN/dS ratios in both P230 and P47 may indicate that similar selection pressures exists in different Plasmodium species. Whether this positive selection on these gamete proteins is driven by immune responses and/or mating interactions is presently unknown. However, insight into sequence polymorphisms in gamete surface proteins that are targets for TB vaccines and the influence of these polymorphisms on mating behaviour of parasites in natural populations of P. falciparum should help to improve TB vaccines development.
|
10.1371/journal.pntd.0005217 | In vivo Distribution and Clearance of Purified Capsular Polysaccharide from Burkholderia pseudomallei in a Murine Model | Burkholderia pseudomallei is the causative agent of melioidosis, a severe infection prominent in northern Australia and Southeast Asia. The “gold standard” for melioidosis diagnosis is bacterial isolation, which takes several days to complete. The resulting delay in diagnosis leads to delayed treatments, which could result in death. In an attempt to develop better methods for early diagnosis of melioidosis, B. pseudomallei capsular polysaccharide (CPS) was identified as an important diagnostic biomarker. A rapid lateral flow immunoassay utilizing CPS-specific monoclonal antibody was developed and tested in endemic regions worldwide. However, the in vivo fate and clearance of CPS has never been thoroughly investigated. Here, we injected mice with purified CPS intravenously and determined CPS concentrations in serum, urine, and major organs at various intervals. The results indicate that CPS is predominantly eliminated through urine and no CPS accumulation occurs in the major organs. Immunoblot analysis demonstrated that intact CPS was excreted through urine. To understand how a large molecule like CPS was eliminated without degradation, a 3-dimenational structure of CPS was modeled. The predicted CPS structure has a rod-like shape with a small diameter that could allow it to flow through the glomerulus of the kidney. CPS clearance was determined using exponential decay models and the corrected Akaike Information Criterion. The results show that CPS has a relatively short serum half-life of 2.9 to 4.4 hours. Therefore, the presence of CPS in the serum and/or urine suggests active melioidosis infection and provides a marker to monitor treatment of melioidosis.
| An outer membrane component, capsular polysaccharide (CPS), is a virulence factor expressed by many Gram-negative bacteria including Burkholderia pseudomallei, the causative agent of melioidosis. Recently, B. pseudomallei CPS was identified as a useful diagnostic biomarker, leading to the development of a lateral flow immunoassay (LFI) targeting CPS for B. pseudomallei detection. In this current work, we studied the in vivo fate of CPS using a murine model, to better understand the clinical applications and potential limitations of the LFI. Interestingly, we found that B. pseudomallei CPS has a unique set of characteristics (as compared to other bacterial capsule antigens) including rapid kidney clearance from serum, no deposition in major internal organs, and ability to be cleared without degradation. Clinically, these findings suggest that CPS may be a potential biomarker for detecting active melioidosis and monitoring melioidosis treatment outcome. Additionally, urine may be used as a non-invasive sample for detecting melioidosis.
| Burkholderia pseudomallei is a Gram-negative, soil-dwelling bacillus and the etiologic pathogen of melioidosis, a severe infection endemic in tropical areas with the highest incidence in Southeast Asia and northern Australia [1]. In early 2016, it was predicted that approximately 165,000 individuals worldwide would suffer from melioidosis, while 89,000 of them would die from the infection [2]. B. pseudomallei has also been acknowledged as a potential agent of biological warfare and terrorism because of its ability to cause severe disease via airborne transmission [3,4]. Due to the possibly significant impact on public health and the inherent potential for misuse, the Centers for Disease Control and Prevention (CDC) has classified this organism as a Tier 1 select agent [5]. Currently, there is no licensed vaccine available to prevent the infection. In addition, since B. pseudomallei is resistant to common antibiotics, the success of melioidosis treatment greatly relies on rapid point-of-care diagnosis [6].
At present, bacterial isolation using Ashdown’s selective medium remains the diagnostic gold standard for melioidosis. This technique is only 60% sensitive along with being time consuming, causing treatment delays and increased mortality risk [7]. Rapid diagnostic methods such as latex agglutination, immunofluorescence assay (IFA), ELISA, and PCR have been developed for B. pseudomallei detection [8]. In addition to these techniques, a lateral flow immunoassay (LFI) targeting the capsular polysaccharide (CPS) of B. pseudomallei developed by our group has been shown to be one of the most promising methods for rapid point-of-care detection of melioidosis, especially in resource poor settings [8–10].
The LFI uses a murine monoclonal antibody (mAb) specific to CPS to detect the presence of the bacterium (by detecting CPS) in patient samples. Capsular antigens are outer membrane components expressed by many Gram-negative bacteria, and CPS is known to be one of the most important virulence factors for B. pseudomallei. Structurally, B. pseudomallei CPS is an unbranched homopolymer of 1, 3-linked 2-O-acetyl-6-deoxy-β-D-manno-heptopyranose with an approximate molecular weight of 300 kDa [11,12]. Previous animal model studies have found that a CPS-specific antibody provides protection against lethal challenge with B. pseudomallei, suggesting that CPS is a candidate target for melioidosis vaccine development [12–14]. In addition, a recent study from our laboratory revealed that CPS antigen circulates in the bloodstream during infection; this led us to develop the CPS-targeting LFI [15]. Currently, clinical performance of the LFI is being assessed in several endemic areas. However, relatively little is known about the ultimate fate of CPS in vivo. The main focus of this study was to investigate the in vivo distribution and clearance of CPS, information that is essential for improving the clinical use of the LFI.
Culture media was inoculated with B. pseudomallei RR2683 (O-polysaccharide mutant; select agent-exempt strain, originating in the Brett laboratory) and incubated overnight at 37°C with vigorous shaking [12]. Cell pellets were obtained by centrifugation and extracted using a modified hot aqueous-phenol procedure [11]. Purified CPS was obtained as previously described [12].
Female, 8-week old CD1 mice (Charles River Laboratories, Inc., Frederick, MA) were injected with 200 μL of dPBS (Mediatech, Inc., Manassas, VA) containing 4, 20 or 100 μg of purified CPS via the tail vein. The CPS doses were chosen according to previous research investigating the clearance of capsule components of Bacillus anthracis [16]. At 30 min, 2 hours, 4 hours, 8 hours, 12 hours, 1 day, 2 days, 4 days and 8 days post-injection, mice were euthanized using CO2 for sample collection. Urine samples were collected just prior to death. Immediately after euthanasia, blood samples were collected via cardiac puncture and sera were separated. Internal organs including lungs, liver, spleen and kidneys were harvested, weighed and homogenized in 2 mL of dPBS using a PRO250 homogenizer (Pro Scientific, Oxford, CT). Homogenates then were centrifuged and supernatants were collected. All samples were stored at -80°C until quantitative ELISAs were performed.
The use of laboratory animals in this study was approved by the University of Nevada, Reno Institutional Animal Care and Use Committee (protocol number 00024). All work with animals at the University of Nevada, Reno is performed in conjunction with the Office of Lab Animal Medicine, which adheres to the National Institutes of Health Office of Laboratory Animal Welfare (OLAW) policies and laws (assurance number A3500-01).
An antigen-capture (sandwich) ELISA for CPS quantification was developed using CPS-specific mAb 4C4. Isolation of mAb 4C4 was described previously [17]. Microtiter plates were coated overnight with 100 μL of mAb 4C4 (2.5 μg/mL in PBS) at room temperature. The plates were washed with PBS-Tween (PBS containing 0.05% Tween 20) and blocked with a blocking solution (PBS containing 5% skim milk and 0.5% Tween 20) at 37°C for 1 hour. After blocking, the plates were washed with blocking solution, and then incubated at room temperature for 90 min with 100 μL of a twofold serial dilution of samples (sera, urine or supernatants from tissue homogenates) diluted in blocking solution. A standard CPS sample was prepared by spiking purified CPS into untreated samples diluted in blocking solution. The final concentration of the standard CPS samples was 30 ng/mL. The CPS standard then was added to the plates and serially diluted along with samples to generate the standard curve. After incubation, the plates were washed again with blocking solution, incubated with a mAb 4C4-horseradish peroxidase (HRP) conjugate (0.5 μg/mL in blocking solution) at room temperature for 1 hour, followed by washing with PBS-Tween. The plates were developed by adding 100 μL of tetramethylbenzidine (TMB) substrate (KPL, Gaithersburg, MD) into each well. The reaction was stopped with 1 M H3PO4, and then the optical density was read at 450 nm (OD450). CPS concentrations in samples were determined by comparison with the standard curve using an OD450 of 0.5 as the endpoint. The limit of detection of the assay is approximately 0.25 ng/mL.
The amounts of CPS in organ homogenates are reported as micrograms CPS per organ. The CPS amounts reported were corrected by subtraction of the amount of CPS found in the plasma volume in each organ [18], and resulting negative values after subtraction were adjusted to zero. The organ analysis results are presented in comparison with CPS amounts present in serum, which were calculated based on the following information: 1) the blood volume of a mouse is estimated at 5.77 mL/100g, 2) half of the blood volume is plasma [19], and 3) the average weight of mice used in the study is 25.9 g.
The kinetics of CPS clearance from serum was analyzed using four different exponential decay models: 1) a two-parameter monophasic exponential decay model,
y=ae−bx,
where a is the Y intercept (concentration of CPS present at time zero) and b is the rate constant of clearance, 2) a three-parameter monophasic exponential decay model,
y=ae−bx+y0,
where a is the Y intercept (concentration of CPS present at time zero), b is the rate constant of clearance, and y0 is plateau (concentration of CPS persistent in serum), 3) a four-parameter biphasic exponential decay model,
y=ae−bx+ce−dx,
where a is the proportion of CPS that clears rapidly during the initial clearance step, b is the rate constant of the initial clearance, c is the proportion of CPS that clears more slowly, and d is the rate constant of slower clearance step, and 4) a five-parameter biphasic exponential decay model,
y=ae−bx+ce−dx+y0,
where a is the proportion of CPS that clears rapidly during the initial clearance step, b is the rate constant of the initial clearance, c is the proportion of CPS that clears more slowly, d is the rate constant of slower clearance step, and y0 is plateau (concentration of CPS persistent in serum). The model fitting was carried out using SigmaPlot 13.0 (Systat Software Inc., San Jose, CA). Corrected Akaike's Information Criterion (AICc), a standard for model selection, was used to evaluate how well each model represents the data. The model that best describes the data (lowest AICc value) among the four equations was selected and used to determine the half-life of CPS in serum.
To analyze excreted CPS, urine samples from mice injected with 100 μg CPS were analyzed by immunoblot analysis. Only samples collected at 30 min, 2 hours, and 8 hours contained enough CPS for the analysis. The samples were diluted in SDS-PAGE sample buffer, treated with proteinase K (Fisher Scientific, Waltham, MA) at 60°C for 1 hour, and boiled for 10 min prior to electrophoresis on 7.5% TGX precast gels (Bio-Rad, Hercules, CA). The volume of each sample loaded onto the gel was adjusted so that an equal amount of CPS (approximately 1 μg) was present in each lane. Western blotting was performed with mini-nitrocellulose transfer packs and a Trans-Blot Turbo transfer system (Bio-Rad). The membranes were blocked with 5% skim milk in TBS-Tween (TBS-T: 50 mM Tris, 150 mM NaCl, 0.1% Tween 20, pH 7.6) at 4°C overnight, followed by incubation with 1 μg/mL of mAb 4C4 for 90 min at room temperature. After washing with TBS-T, the membranes were incubated with an anti-mouse IgG-HRP conjugate (Southern Biotech, Birmingham, AL) for 60 min at room temperature to facilitate detection. The final development was carried out using Pierce ECL Western Blotting Substrate (Pierce Biotechnology, Rockford, IL) and a ChemiDoc XRS imaging system (Bio-Rad).
The chemical structure of the B. pseudomallei CPS antigen was obtained from previously published work [11]. The structure was drawn in ChemDraw Prime version 15.0 (PerkinElmer, Waltham, MA) and exported to ChemBio3D Ultra software (PerkinElmer). Due to limitations of the software in processing a large molecule like CPS, only a fragment of CPS that consists of 100 units of 2-O-acetyl-6-deoxy-β-D-manno-heptopyranose was built. Three-dimensional (3D) structure of the CPS was predicted using MM2 energy minimization mode implemented in the ChemBio3D software and exported in protein data bank (PDB) format. The PyMOL program (Schrödinger, LLC, www.pymol.org) was used to visualize and analyze the 3D structure of CPS.
Active Melioidosis Detect™ (AMD™) LFI (InBios International, Inc., Seattle, WA) was used to detect excreted CPS in a urine sample. The urine sample was collected from a mouse injected with 4 μg CPS at 30 min post injection. The sandwich ELISA was used to determine the CPS concentration in this sample. The sample then was serially diluted with mouse control urine to yield the desired concentrations of CPS (0.04–625 ng/mL). Each dilution of urine (5 μL) was mixed with 20 μL of chase buffer (included in AMD™ kit). The mixture then was applied to the sample pad, followed by an additional 100 μL of chase buffer. The tests were allowed to develop for 15 min. The results were assessed by four examiners in a semi-blinded, randomized manner and photographed. Intensities of the test lines were also quantified using an ESE-Quant lateral flow reader (QIAGEN, Helden, Germany).
Mice were intravenously injected with various doses of purified CPS. At the designated time points, serum and urine samples were collected, and CPS concentrations were determined using antigen capture ELISA. Preliminary analysis was performed to ensure that CPS detection was not significantly affected by the presence of serum or urine (S1 Fig). The kinetics data for CPS in serum were fitted to the exponential decay models, and AICc then was used for model selection. According to the AICc results, clearance of CPS from serum is best described by the two-parameter monophasic exponential decay model (y = ae-bx) (S1 Table). Data analysis showed that CPS was cleared rapidly from serum with a half-life (95% confidence interval) of 4 hours (2.5–6.6 hours), 4.4 hours (2.0–9.7 hours), or 2.9 hours (2.3–3.9 hours), for the doses of 100 μg, 20 μg, or 4 μg, respectively, suggesting that the half-life values derived from all three doses were comparable (Fig 1).
Fig 2 shows the concentrations of CPS in urine at different time intervals. The CPS concentrations in urine were found to be highest at 30 min (the first time point of sample collection), and then decreased rapidly, corresponding to the decrease of CPS in serum.
Kidneys, livers, lungs and spleens were collected from the same CPS-injected mice that were used for the CPS clearance study, but only the samples from mice injected with 100 μg CPS were used. The organs were homogenized in PBS for analysis of CPS concentrations by ELISA. The preliminary analysis showed that the presence of tissue homogenates had no effect on assay performance (S1 Fig). CPS amounts in each organ were reported in comparison with amounts of CPS found in serum (Fig 3). The results showed no significant amounts of CPS deposited in the organs. Since we could not detect CPS in organs from mice injected with 100 μg CPS, the internal organs from mice receiving 20 μg or 4 μg CPS were not analyzed in this study.
Altogether the results demonstrated that the circulating CPS was eliminated rapidly and predominantly through urine, without accumulation in the major organs. To find out how a large molecule like CPS was excreted, the urine samples from CPS-treated mice were analyzed by Western blot (Fig 4). The Western blots detected full-length CPS in the urine samples, while degraded CPS was not detected. When interpreting these results, however, it is important to note that CPS epitopes might be affected by degradation, so the Western blot analysis might not be able to detect degraded products of CPS. Therefore, it is appropriate to deduce that some (but not necessarily all) of the injected CPS was eliminated without degradation. In some instances, two bands of CPS were observed in our blots. The higher molecular weight band seemed to comprise the largest fraction of CPS in urine samples from these mice; however, how and why this occurred needs to be further investigated (Fig 4).
Analysis of the total urine protein was used to assess whether or not exogenous CPS affected renal function, which could result in leakage of high molecular weight compounds into urine. The results showed that there was no difference in urine protein profiles between CPS-treated and untreated mice, suggesting that kidney impairment was unlikely the cause of rapid renal excretion of CPS (S1 Fig). In order to explain how CPS was excreted through the kidneys without apparent degradation, a three-dimensional structure of CPS was predicted (Fig 5). However, due to limitations of the software, only a short fragment (~22 kDa) of CPS was constructed (full-length CPS has a molecular weight of ~300 kDa). The computational 3D model demonstrated that CPS has a rod-like shape with a diameter of approximately 1.2 nm. The length of a single molecule of CPS, which was calculated from the length of the 22 kDa fragment model, was approximately 490 nm.
Previous experiments demonstrated that CPS is cleared rapidly and predominantly through urine, suggesting that urine has the potential to be used as a non-invasive sample for diagnosis of melioidosis. AMD™ LFI is an assay designed to detect B. pseudomallei by targeting CPS molecules in various types of biological samples. To assess whether or not the sensitivity of AMD™ LFI was impacted when it was used to detect excreted CPS in urine samples, serial dilutions of urine from a CPS-treated mouse were tested with the LFI. The results showed that AMD™ LFI could detect excreted CPS as low as 0.2 ng/mL (Fig 6), comparable with the AMD™ LFI sensitivity reported previously [10].
Like many other pathogenic microorganisms such as Bacillus anthracis, Haemophilus influenzae type b, Streptococcus pneumoniae, and Cryptococcus neoformans, B. pseudomallei expresses a capsular antigen that is shed into the bloodstream during infection [15,20–22]. Based on this finding, immunodiagnostic methods (IFA and LFI, among others) targeting B. pseudomallei CPS were developed. These diagnostic tools showed similar specificity but higher sensitivity when compared to the ‘gold standard’ diagnostic bacterial culture, suggesting that these techniques have the potential to be used clinically [7,10,23]. To use these tools as a routine diagnostic method, however, it is important to understand the fate of CPS in vivo. This could provide insight into the retention and processing time for CPS in a patient during infection, and the potential patient sample types that can be targeted for testing.
To understand how CPS is processed in vivo, mice were intravenously injected with purified B. pseudomallei CPS. CPS concentrations in samples (blood, urine, lungs, liver, spleen, and kidneys) at various time points post-injection were determined using an antigen capture ELISA. The doses of CPS used in this study were chosen according to previous research investigating the clearance of B. anthracis capsule antigen [16]. We acknowledge that the concentration of CPS in patient serum reported previously was much lower than the lowest dose of CPS we used in this study [10]. However, the range of CPS concentrations reported was from a limited number of serum samples, and was not associated with the stage of infection. In addition, the concentrations reported previously were too low to allow us to collect accurate and sufficient kinetic data in our study. Thus, the experiments were conducted using 4, 20, or 100 μg CPS per mouse.
According to our results, B. pseudomallei CPS was rapidly cleared from serum with a half-life of 2.9–4.4 h (Fig 1), and not deposited in kidneys, lungs, liver, or spleen (Fig 3). However, relatively high concentrations of CPS were detected in urine shortly after the injection (Fig 2). Notably, at 30 min post-injection, we found that the CPS concentrations in urine were higher than those found in serum for mice receiving 20 or 100 μg CPS (Fig 2). These results indicated that the kidney is the major organ responsible for CPS elimination.
Comparison of our findings with the fate of capsular antigens from other organisms suggests that B. pseudomallei CPS has a unique set of characteristics. C. neoformans produces glucuronoxylomannan (GXM) as a major capsule component [24], while B. anthracis produces a capsule composed of a poly-γ-D-glutamic acid (PGA) polypeptide [25]. B. pseudomallei capsule is composed of polysaccharides and is somewhat similar in chemical composition to GXM; however, B. pseudomallei CPS and PGA are apparently similar in geometry, as both of them have a rod-shaped structure [26]. Grinsell et al. demonstrated that GXM has a long serum half-life (~1.6 days) [19]. However, CPS from B. pseudomallei behaved more like the capsular polypeptide PGA, as both of them showed rapid serum clearance [16]. Previous studies have also found that pneumococcal polysaccharide, GXM and PGA accumulated in many mouse tissues [16,19,27]. In addition, the liver and spleen were found to play important roles in clearance of both GXM and PGA. B. pseudomallei CPS, however, was not deposited in any mouse organs (kidneys, lungs, spleen and liver), and it was cleared predominantly by the kidneys. Altogether, the results suggest that B. pseudomallei CPS exhibits certain characteristics distinct from capsular antigens of other previously reported microbes.
Sutherland et al. also showed that PGA was found in urine at high concentrations, which we also found to be true with B. pseudomallei CPS. The study revealed that PGA was excreted in a degraded form [16]. Since B. pseudomallei CPS has a high molecular weight, which is much greater than the molecular cutoff for glomerular filtration [28], we anticipated that we would find degraded CPS in urine, as previously seen in the PGA study. However, our results illustrated that a portion (if not all) of circulating B. pseudomallei CPS was apparently excreted without degradation (Fig 4). The CPS molecule, thus, was further investigated using 3D computer modeling that allowed us calculate a structure of the molecule. The result showed that CPS has a rod-like shape with the dimensions (diameter x length) of 1.2 nm x 490 nm (Fig 5). We found that the structure of CPS resembles a high molecular weight molecule (~350–500 kDa) of a single-walled carbon nanotube (SWCN, dimension = ~1 nm x ~500 nm) [29]. Ruggiero et al. demonstrated that in vivo SWCN was excreted rapidly, and predominantly by glomerular filtration of the kidneys, even though its molecular weight exceeds the known glomerular cutoff [29]. As explained by Ruggiero et al., SWCN has a diameter smaller than a glomerular pore (~10 nm), and capillary flow orients the major axis of the rod to align with the glomerular orifice, thereby allowing it to flow through the kidneys. Since B. pseudomallei CPS molecules and SWCN are geometrically similar, it is possible that CPS and SWCN are eliminated via the kidneys by the same mechanism.
Our results show that B. pseudomallei CPS has a short serum half-life like PGA from B. anthracis. However, while the half-life of PGA was dose-dependent, we found that the B. pseudomallei CPS half-life was dose-independent (Fig 1). This reflected the possibility that CPS and PGA might be eliminated by different mechanisms. We know that PGA was eliminated following degradation; the degradation capacity, however, can be saturated when a treatment dose of PGA exceeds a certain concentration. As a consequence, PGA accumulated faster than it could be cleared; thus the half-life of PGA became longer when the doses were higher (dose-dependence) [16]. For B. pseudomallei CPS, however, half-life was independent from the treatment doses. We interpret these results to indicate that the CPS elimination capacity was not saturated, at least by the highest concentration used in this study. We found this interpretation fit our proposed mechanism of CPS elimination, in describing that B. pseudomallei CPS is eliminated passively through glomerular filtration, rather than by degradation or carrier proteins.
Finally, our results have several implications for the clinical use of immunodiagnostics detecting B. pseudomallei CPS. We know from our previous study that during infection CPS is shed into the blood circulation, indicating that serum can be used as a sample for diagnosis of melioidosis [15]. In this study, we have revealed that CPS can also be detected in urine samples at a high concentration. This finding is consistent with other studies where urine samples from human [10] and non-human primates infected with B. pseudomallei (K. C. Brittingham, A. Leon, P. A. Braschayko, M. S. Anderson, K. A. Knostman and R. E. Barnewall, presented at the ASM Biodefense and Emerging Diseases Research Meeting, Arlington, VA, 8 to 10 February 2016) were analyzed; results showed a large amount of CPS was present in the urine. In this study, we also discovered that CPS has a half-life of approximately 2.9 to 4.4 h in serum. We noted that our experiments were performed in non-infected animals that have no antibody to CPS. Antibody is known to play an important role in clearing various exogenous antigens, including capsular antigen [30]. Thus, it is possible that, in infected animals or patients with CPS-specific antibody, the serum half-life of CPS could be even shorter than the half-life reported in this study. Since CPS is cleared rapidly from serum by the kidneys, the presence of CPS in serum or urine may suggest an active source of B. pseudomallei antigen, i.e. acute B. pseudomallei infection. It also suggests that CPS could be a potential biomarker for monitoring efficacy of melioidosis treatment. In addition, we also found that the AMD™ LFI can efficiently detect eliminated CPS in a urine sample (Fig 6). These findings together suggest that perhaps urine, a noninvasive sample that contains a high concentration of CPS, as a sample for melioidosis diagnosis is more appropriate than using serum samples.
In summary, the in vivo clearance of B. pseudomallei CPS has a unique set of characteristics, including i) rapid serum clearance, ii) no significant accumulation in internal organs, iii) potentially passive excretion by glomerular filtration, and iv) presence at a high concentration in urine. Rapid serum clearance of CPS suggests that CPS is a significant biomarker for identifying active melioidosis and monitoring treatment progress. In addition, urine, a noninvasive sample, also has a potential to be used as a clinical specimen for melioidosis diagnosis.
|
10.1371/journal.pbio.2006134 | KDM5 histone demethylases repress immune response via suppression of STING | Cyclic GMP-AMP (cGAMP) synthase (cGAS) stimulator of interferon genes (STING) senses pathogen-derived or abnormal self-DNA in the cytosol and triggers an innate immune defense against microbial infection and cancer. STING agonists induce both innate and adaptive immune responses and are a new class of cancer immunotherapy agents tested in multiple clinical trials. However, STING is commonly silenced in cancer cells via unclear mechanisms, limiting the application of these agonists. Here, we report that the expression of STING is epigenetically suppressed by the histone H3K4 lysine demethylases KDM5B and KDM5C and is activated by the opposing H3K4 methyltransferases. The induction of STING expression by KDM5 blockade triggered a robust interferon response in a cytosolic DNA-dependent manner in breast cancer cells. This response resulted in resistance to infection by DNA and RNA viruses. In human tumors, KDM5B expression is inversely associated with STING expression in multiple cancer types, with the level of intratumoral CD8+ T cells, and with patient survival in cancers with a high level of cytosolic DNA, such as human papilloma virus (HPV)-positive head and neck cancer. These results demonstrate a novel epigenetic regulatory pathway of immune response and suggest that KDM5 demethylases are potential targets for antipathogen treatment and anticancer immunotherapy.
| Pathogens often find ways to turn down cell-intrinsic antipathogen immune responses by the host. Similarly, cancer cells use various mechanisms to evade attack by immune cells. One of the common mechanisms is suppression of the stimulator of interferon genes (STING)-dependent innate immune response. Using potent and specific small-molecule inhibitors and genetic-depletion approaches, we found that the silenced STING pathway can be reactivated in breast cancer cells by suppressing KDM5 demethylases. Activation of the STING pathway led to a robust interferon response, which blocked viral infection, and was associated with increased tumor-infiltrated lymphocytes and better patient survival in multiple cancer types. This discovery has major clinical implications for treating both pathogen infection and cancer because KDM5 inhibition provides a fast, robust, and reversible control of innate immune response. Since the discovery of histone demethylase activity of KDM5 proteins a decade ago, significant efforts have been dedicated to developing KDM5 inhibitors for clinical applications. In fact, a KDM5 inhibitor recently entered phase I clinical trial for treatment of hepatitis B infection. Here, we provide mechanistic insights on how KDM5 inhibitors block viral infection. Moreover, our results suggest that KDM5 inhibitors can also be combined with other cancer immunotherapies.
| Evasion from immunosurveillance by cancer cells is a major cancer hallmark [1], and restoration of immunosurveillance has been demonstrated as an effective antitumor strategy. For example, antibodies targeting inhibitory checkpoint molecules, including programmed cell death protein 1 (PD-1) and cytotoxic T-cell lymphocyte-associated protein 4 (CTLA-4), have achieved remarkable efficacy in the clinic [2]. However, only a small percentage of patients respond to these therapies. Thus, the mechanisms for lack of response to these treatments are areas of intense investigation. Lack of T-cell infiltration (also known as immunologically “cold” tumors) appears to characterize a major subset of patients who do not respond to treatment [3]. Identification of strategies that convert tumors from an immunologically “cold” to “hot” state could enhance immune checkpoint inhibitor therapies and potentially result in the effective treatment of patients who otherwise would not have responded.
Pattern recognition receptors (PRR) are cell surface and intracellular sensors that recognize pathogen-associated and abnormal-self molecular patterns, e.g., nucleic acids, and trigger intracellular signaling cascades to activate cell-intrinsic antipathogen or antitumor responses [4]. Cyclic GMP-AMP (cGAMP) synthase (cGAS) senses pathogen- or abnormally released self-DNA [5, 6] and signals through stimulator of interferon genes (STING) [7]. RNA helicases retinoic acid inducible gene I (RIG-I) and melanoma differentiation-associated gene 5 (MDA5) are the main cytosolic RNA sensors—and activate the interferon pathway through mitochondrial antiviral signaling protein (MAVS)—whereas toll-like receptors (TLRs) respond to pathogen-associated molecular patterns on the cell surface or in endosomal compartments [4]. The downstream pathway of these diverse receptors converges on a few key transcription factors called interferon regulatory factors (notably IRF3 and IRF7) and protein kinases (such as TANK-binding kinase 1 [TBK1]) responsible for the phosphorylation and nuclear translocation of IRF3 and IRF7 [8]. Activated IRFs drive the transcription of type I interferons, which bind to their cognate cell surface receptors and lead to the formation of the canonical signal transducer and activator of transcription 1 (STAT1)–STAT2–IRF9 (also known as interferon-stimulated gene factor 3 [ISGF3]) complex. The ISGF3 complex binds to the promoters of interferon-stimulated genes (ISGs) and activates these genes, many of which mediate the immune response [8]. Emerging evidence suggests that the cGAS-STING pathway plays a critical role in bridging innate immunity and adaptive immunity in tumors [9–11]. However, this pathway is silenced in many tumors, and the mechanisms of their silencing remain largely unknown [12–15].
Tri-methylation on histone H3 lysine 4 (H3K4me3) is enriched near transcription start sites and strongly correlates with active transcription [16]. Methylation on H3K4, like other histone marks, is dynamically controlled through the concerted action of lysine methyltransferases, the writers, and demethylases, the erasers [16]. The lysine demethylase 5 (KDM5) family proteins—including KDM5A-D (also known as JARID1A-D)—are Fe (II)- and α-ketoglutarate-dependent dioxygenases and catalyze the removal of the methyl groups from H3K4me3 [17]. The KDM5 family demethylases play major roles in human cancers. KDM5A physically and functionally interacts with tumor suppressor pRb [18]. KDM5B is up-regulated in breast cancer cells overexpressing the ERBB2/HER2 oncogene [19]. Gene amplification of both KDM5A and KDM5B were found in various human cancers [20, 21]. Studies using cancer cell lines and mouse models demonstrated their functions in promoting tumorigenesis in multiple cancer types [17, 21–29]. However, the mechanisms by which KDM5 proteins contribute to these phenotypes are still largely unclear.
Here, we report that KDM5 demethylases suppress STING-induced innate immune response in tumor cells. We found that KDM5B and KDM5C bind to the STING locus and maintains a low level of H3K4me3 to suppress STING expression. Inhibition or depletion of KDM5B and KDM5C led to increased STING expression in a wide range of cancer cells. In the presence of abnormal cytosolic DNA, the increased STING led to a robust induction of ISGs in breast cancer cells and antiviral response through the cGAS-STING-TBK1-IRF3 pathway. Lastly, we found a strong negative correlation between KDM5B expression and STING expression in The Cancer Genome Atlas (TCGA) tumor samples. Our findings reveal a novel epigenetic suppressive mechanism of innate immune response and suggest KDM5 demethylases as attractive targets to boost antitumor immune response.
All 4 family members of KDM5 demethylases (KDM5A-D) share sequence and structure similarity [17], have similar in vitro kinetic parameters [30], and display functional redundancy [31]. Depletion of individual KDM5 enzymes usually alters histone modification level and gene expression in a context-dependent manner [17], but the effects of inhibiting multiple KDM5 enzymes remain unclear. Multiple potent pan-KDM5 inhibitors—including KDM5-C49 (cell active form is KDM5-C70) [30, 32], Dong-A-167 (patent WO2016068580), GDC-50 [33], and CPI-48 [34]—have been reported. These inhibitors are known or predicted to compete with the cofactor α-ketoglutarate in the active site of KDM5 enzymes (S1A–S1F Fig and S1 Table) and inhibited KDM5 enzymes with half maximal inhibitory concentration (IC50) values in the nM range (S1G and S1H Fig and S1 Data). We examined the effects of these small-molecule inhibitors on histone modifications and gene expression in MCF7 breast cancer cells. First, global levels of H3K4me3 increased in inhibitor-treated cells (Fig 1A), consistent with previous results [30, 32–36]. Second, these inhibitors showed minimal effects on other histone methylation marks, including tri-methylation on histone H3 lysine 9 (H3K9me3—a substrate for the KDM4 family), lysine 27 (H3K27me3—a substrate for the KDM6 family), and lysine 36 (H3K36me3—another substrate for the KDM4 family), as well as di- or mono-methylation on histone H3 lysine 4 (H3K4me2/me1, substrates for the KDM1/LSD and KDM5 family) (Fig 1A). Third, KDM5-C70 treatment induced KDM5B and KDM5C protein levels without affecting KDM5A protein level (S2A Fig). It is possible that the induction of KDM5B and KDM5C is due to a feedback regulation, and the mechanism of their differential induction will require further investigation. Fourth, despite the global increase of H3K4me3, RNA sequencing (RNA-seq) analysis of MCF7 cells treated with inhibitors KDM5-C70 and CPI-48 revealed major up-regulation of gene expression only in limited pathways (S2B Fig). The top up-regulated genes are involved in the interferon response pathway (Fig 1B, S2B Fig and S2 and S3 Data). Reverse transcription followed by quantitative PCR (RT-qPCR) analysis detected a robust increase of ISGs with direct antiviral activities, such as OAS2, IFI44L, IFI44, IFIT1, and IFIT3, and chemokine genes involved in immune cell recruitment, such as CXCL10, upon treatment with inhibitors (Fig 1C and S2C Fig).
Phosphorylated STAT1, which is often required for induction of ISGs [8], increased along with total STAT1 (Fig 1D). Consistently, other genes involved in type I interferon response were up-regulated, including cytosolic RNA sensors RIG-I and MDA5, and interferon-regulatory factors IRF7 and IRF9 (Fig 1D and S2C Fig). Treatment of other breast cancer cells SKBR3 and BT474 by compound KDM5-C70 also induced expression of OAS2, IFI44L, and IFI44, but to a lesser extent (S2D and S2E Fig). We noted that compound KDM5-C70 at 1 μM significantly induced a global change of H3K4me3 level and targeted gene expression, whereas the other 3 compounds at 10 μM showed similar (or less) potency (Fig 1A and 1D), therefore we used 1 μM KDM5-C70 in the remaining study.
Depletion of KDM5B or KDM5C, but not KDM5A, mediated by clustered regular interspaced short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) led to moderately increased expression of ISGs, and knockout of KDM5B and KDM5C synergistically enhanced their expression (Fig 1E and 1F). KDM5D is located in the Y chromosome [17] and thus not expressed in breast cancer cells derived from female patients. Similar effects were observed in cells with small interfering RNA (siRNA)-mediated individual and combinatorial knockdown of KDM5B and KDM5C demethylases (S2F and S2G Fig). Compared to the effects of KDM5 inhibitor treatment, the magnitude of ISG activation was slightly lower in KDM5B and KDM5C double knockout cells. It may be due to incomplete depletion of KDM5B and KDM5C in polyclonal knockout cells that we used. Activation of negative feedback pathways during the time required to generate stable cell lines could have also dampened the effects.
Ectopic overexpression of a catalytic deficient KDM5B mutant (H499A), but not wild-type KDM5B, dramatically activated expression of ISGs (Fig 1G), suggesting that this KDM5B mutant had dominant negative effects. Collectively, these results showed that the demethylase activities of KDM5B and KDM5C are required to inhibit the interferon pathway.
It is well-known that type I interferon establishes an antiviral state [8]. To assess the biological outcome of interferon response induced by KDM5 inhibition, we challenged inhibitor-treated cells with vesicular stomatitis virus (VSV, a negative-stranded RNA virus) carrying a green fluorescence protein (GFP) reporter (VSV-GFP) or vaccinia virus (a double-stranded DNA [dsDNA] virus). Infection by both viruses can be suppressed by treatment with type I interferons [37, 38]. To exclude the direct effects of KDM5 inhibition on viral infection or reproduction, KDM5-C70 was removed 1 day before infection. We found that pretreatment of cells with KDM5-C70 significantly inhibited VSV-GFP infection (Fig 2A and 2B). Similarly, analyzing the copy number of the viral genome at different time points after vaccinia virus infection revealed that viral replication was significantly restrained in inhibitor-pretreated cells (Fig 2C). As a result, inhibitor-pretreated cells resisted some lytic effects of vaccinia virus (Fig 2D) and produced much fewer viruses compared with control cells (Fig 2E). Similar results were obtained when KDM5B and KDM5C were depleted by CRIPSR/Cas9-mediated knockout (Fig 2F–2H). In summary, inhibition of KDM5 enzymes potentiates antiviral innate immunity.
We next examined which pathway is required for the interferon response triggered by KDM5 inhibition. Using the CRISPR/Cas9 system, we depleted major components in the interferon-inducing PRR pathways individually, including RIG-I, MDA5, MAVS, TBK1, IRF3, IRF7, cGAS, STING, and TLR3 (Fig 3A–3C). Efficient knockout of these genes was achieved in polyclonal setting as shown by western blot (Fig 3B and 3C) or T7 endonuclease assay (S3A Fig). Depletion of cGAS, STING, IRF3, or TBK1 largely abolished KDM5-C70-induced expression of IFI44L, ISG15, and other ISGs (Fig 3B–3D and S3B and S3C Fig). In contrast, loss of RIG-I, MDA5, MAVS, IRF7, or TLR3 had minimal effect (Fig 3B–3D). We note that some components in these pathways—including RIG-I, MDA5, and IRF7—are ISG products themselves, and KDM5-C70 treatment induced the expression of these proteins as well (Figs 1D, 3B and 3C and S2C Fig). Consistently, knockout of essential components in the KDM5-C70-triggered interferon response—such as IRF3 and TBK1—blocked the induction of RIG-I and MDA5 (Fig 3B). These data highlight the predominant roles of cGAS-STING in KDM5-inhibition–dependent activation of ISGs. To further confirm the requirement of the cGAS-STING-TBK1-IRF3 signaling pathway for the KDM5-inhibitor–induced interferon response, we conducted combinatorial knockdown of KDM5B and KDM5C in cGAS, STING, TBK1, or IRF3 knockout cells. Loss of any of the components in this signaling pathway was sufficient to blunt the KDM5B/C-loss–induced interferon response (Fig 3E and 3F and S3D Fig). Together, these data suggest that activation of interferon response by KDM5 deficiency is dependent on the cGAS-STING-TBK1-IRF3 signaling cascade rather than on direct modulation of ISG expression.
Consistent with the activation of ISGs in KDM5 inhibitor-treated cells, we observed increased expression of type I interferon and IFN-β, as well as type III interferons IFN-λ1 and IFN-λ2, in response to KDM5-C70 treatment (S3E Fig). We also compared the effects of KDM5-C70 treatment to 5 to 500 unit/ml IFN-β treatment on the expression levels of 32 ISGs, most of which have antiviral activity [39]. We found that KDM5 inhibition induced similar patterns of ISGs as IFN-β treatment, and the extent of ISG induction upon KDM5 inhibition is similar to 25 unit/ml IFN-β treatment (S3F Fig). Knockout of individual components of the cGAS-STING-TBK1-IRF3 signaling pathway significantly blocked the effect of KDM5-C70 on the induction of interferons (S3E Fig). Moreover, conditioned media collected from inhibitor-pretreated control MCF7 cells—but not from cGAS-, STING-, TBK1-, or IRF3-deficient cells—were able to activate ISG expression in inhibitor-untreated cells (S3G Fig). Furthermore, loss of any member of the ISGF3 complex, namely STAT1, STAT2, and IRF9, blocked the effects of KDM5-C70 (Fig 3G and S3H Fig). Taken together, our data suggest that inhibition of KDM5 enzymes facilitates the cGAS-STING-TBK1-IRF3 signaling cascade to trigger an interferon response, resulting in increased secretion of interferons and activation of the ISGF3 complex to induce the expression of ISGs.
To further determine whether the resistance to viral infection by KDM5 inhibition was also dependent on cGAS-STING-TBK1-IRF3 signaling, we infected inhibitor-treated knockout cells with VSV-GFP or vaccinia virus. Depletion of any member of the cGAS-STING-TBK1-IRF3 signaling cascade, which was required for a KDM5 inhibition-triggered interferon response, diminished the antiviral effects of inhibitor treatment, further confirming the requirement of the cGAS-STING-TBK1-IRF3 pathway for KDM5 inhibition-mediated interferon response (Fig 3H and S4A–S4C Fig).
We showed that the cGAS-STING-TBK1-IRF3 axis was required for KDM5 inhibition-triggered interferon response (Fig 3). The increase of STING after inhibitor treatment does not require IRF3 and TBK1 (Fig 3B), suggesting that STING is directly regulated by KDM5 enzymes in this axis. Both mRNA and protein levels of STING significantly increased after treatment with KDM5-C70 in MCF7, SKBR3, and BT474 breast cancer cells (Fig 4A and 4B) and was variably up-regulated in most of the other cell lines examined (S5A–S5D Fig). Consistently, knockout or knockdown of KDM5B and KDM5C (S5E–S5H Fig), or overexpression of KDM5B H499A mutant, but not wild-type KDM5B, led to STING increase (Fig 4C). The induction of STING by KDM5 inhibitor treatment or by siRNA-mediated combinatorial knockdown of KDM5B and KDM5C was not affected by cGAS, IRF3, TBK1, STAT1, STAT2, or IRF9 knockout (Figs 3B and 4D–4F and S5H Fig), excluding the possibility that the increase of STING was secondary to an activated interferon response. This is in contrast to the RNA sensors RIG-I and MDA5, whose inhibitor-dependent inductions were attenuated upon STING, cGAS, IRF3, or TBK1 knockout (Fig 3B). Overexpression of STING in MCF7 cells was sufficient to induce an interferon response (Fig 4G), further supporting that increased STING per se was responsible for the interferon response resulting from KDM5 inhibition.
To further dissect the mechanisms of STING activation and interferon response, we conducted time course studies to examine the effects of KDM5-C70 on H3K4me3 levels and expression levels of STING and ISGs. The global levels of H3K4me3 increased at day 1 after KDM5-C70 treatment and remained high over time (Fig 4H). STING mRNA levels began elevating at day 1 and peaked at day 3 in all 3 cell lines (Fig 4I–4K). Consistently, STING protein levels also started to increase at day 1 and further increased over time (Fig 4H). In contrast, the activation of ISGs, including RIG-I, MDA5, IRF9, and OAS2, was first seen at day 3 or day 4 (Fig 4H and 4L). Thus, STING induction preceded activation of ISGs, further supporting that STING mediates KDM5 inhibition-induced interferon response.
We next asked whether decreasing the level of H3K4me3, the KDM5 substrate, affects STING expression. The WD40-repeat protein WDR5 is a core component of H3K4 methyltransferase complexes and critical for tri-methylation of H3K4 [40]. Both WDR5 knockout or WDR5 inhibitor OICR-9429, which prevents the binding of WDR5 to the methyltransferase complexes [41], precluded H3K4me3 increase by KDM5 inhibition and abolished the effect of KDM5 inhibition on STING expression (Fig 5A–5C). In addition, chromatin immunoprecipitation (ChIP)-qPCR analysis showed that H3K4me3 at the promoter of STING is induced by KDM5 inhibitor treatment for 1 day in both MCF7 (S6A Fig) and BT474 cells (S6B Fig). In contrast, treatment by KDM5-C70 inhibitor for 1 day had minimal effects on H3K4me3 at the promoters of GAPDH and IFNβ (S6A and S6B Fig). Although H3K4me3 at the promoter of ISGs such as OAS2 and IFI44L increased at day 1, their increases were much smaller than those at day 6 (Fig 5D). These increases of H3K4me3 were abolished in STING knockout cells (Fig 5D), consistent with the idea that KDM5 loss-triggered interferon response results from increased H4K3me3 at the STING promoter and the subsequent up-regulation of STING. Furthermore, KDM5B binds to the promoter of STING in MCF7 cells (Fig 5E) and K562 cells (Fig 5F), while KDM5C binds to the promoter of STING in ZR-75-30 cells (Fig 5F). In contrast, KDM5B and KDM5C do not directly bind to the promoter of cGAS or downstream ISGs, such as OAS2, IFI44L, and IFI44 (S6C Fig). In comparison, although KDM5A binds to the promoter of a known KDM5A target NDUFA9 [29], it does not bind to the STING promoter (Fig 5E). These data suggest that KDM5B and KDM5C maintain a low level of H3K4me3 at the STING promoter, suppress STING expression, and prevent the STING-mediated interferon response.
We noticed that overexpression of STING was sufficient to trigger a robust interferon response in MCF7 cells (Fig 4G), but knockout of cGAS blocked the induction of interferon response by KDM5 inhibition in these cells (Fig 3B and 3D). These data suggested that MCF7 cells had sufficient cytosolic DNA to bind cGAS and trigger cGAMP production to activate STING but had a low level of STING protein that prevented a robust interferon response. Tumor cytosolic DNA can be derived from mitochondria, nuclear DNA leakage, micro-nuclei, or other sources such as oncoviruses [43–48]. We first examined whether MCF7 cells have cytosolic DNA. MCF7 cells were costained with dsDNA and the mitochondrial marker Hsp60. As expected, we observed dsDNA in the cytoplasm of MCF7 cells, but most of these dsDNA did not colocalize with mitochondria (Fig 6A). Treatment with dideoxycytidine (ddC), a deoxyribonucleoside analogue that specifically inhibits mitochondrial DNA (mtDNA) replication [6, 46], led to a dramatic decrease of mtDNA (Fig 6A, right panel) and disappearance of cytosolic DNA (Fig 6A, left panel). These results indicated that cytosolic DNA in MCF7 is mainly derived from mitochondria. To test the requirement of cytosolic DNA derived from mitochondria for the induction of interferon response by KDM5 inhibition, we treated MCF7 cells with KDM5-C70 and ddC. Treatment of ddC strongly inhibited the induction of ISGs by KDM5-C70 (Fig 6B and 6C). These results suggest that mtDNA is required for KDM5-inhibition–triggered interferon response in MCF7 cells. In contrast, treatment with leptomycin B (LMB), an inhibitor of nuclear DNA export, prevented the induction of ISGs by Ataxia-telangiectasia mutated (ATM) and Ataxia-telangiectasia and Rad3-related protein (ATR) inhibitor VE-821 treatment (S7A Fig) [49, 50] but did not suppress the ISG induction by KDM5 inhibition (S7B Fig). These results indicate that nuclear DNA leakage is not the major source of cytosolic DNA in MCF7 cells. Further experiments will be necessary to exclude the possibility that nonmitochondria-derived sources of cytosolic DNA contribute to ISG induction. It is worth mentioning that KDM5 inhibitor treatment did not alter the amount of cytosolic DNA in these cells (S7C Fig). In contrast to MCF7 cells, we observed limited cytosolic DNA in SKBR3 cells (Fig 6D), in which the induction of interferon response by KDM5 inhibition was less robust compared with MCF7 cells (S2D Fig), suggesting that the amount of cytosolic DNA is also a limiting factor for a potent interferon response. To further examine this possibility, we introduced additional cytosolic DNA into SKBR3 cells by transfecting dsDNA, and followed with KDM5-C70 treatment. Treatment with dsDNA or KDM5-C70 alone only led to minimal increase of ISGs, while combinatorial treatment with dsDNA and KDM5-C70 dramatically induced ISGs (Fig 6E). This induction was blocked by knockout of cGAS, STING, TBK1, or IRF3 (Fig 6F and S7D Fig). These data demonstrate that cytosolic DNA is required for full activation of interferon response upon KDM5 inhibition, suggesting that cancer cells with an elevated level of cytosolic DNA can elicit a strong interferon response upon STING induction by KDM5 loss or inhibition.
To validate the regulation of STING by KDM5 in human patients, we compared STING expression levels in “KDM5B low” and “KDM5B high” samples. We found that STING expression level is lower in “KDM5B high” samples than in “KDM5B low” samples from multiple human tumor types, including breast invasive carcinoma, bladder urothelial carcinoma, and ovarian serous cystadenocarcinoma (Fig 7A). To validate the effects of KDM5 on interferon response in tumors with an elevated level of cytosolic DNA, we analyzed human papilloma virus (HPV; a dsDNA oncovirus)-induced tumors, such as head and neck cancer and cervical cancer. In HPV+ head and neck cancer, we found significant negative correlation between KDM5B and STING expression, with a Spearman’s correlation of −0.465 (Fig 7B). Despite the inability to separate HPV+ and HPV− cervical cancer, we observed significant negative correlation between KDM5B and STING expression in cervical cancer, with a Spearman’s correlation of −0.172 (S8A Fig). CXCL10 is one of the interferon-stimulated chemokines that promotes infiltration of immune cells into the tumor microenvironment [10, 51]. We found CXCL10 expression negatively correlated with KDM5B expression in HPV+ head and neck cancer and positively correlated with STING expression in both HPV+ head and neck cancer and cervical cancer (Fig 7C and S8B Fig). Additionally, we found that CD8+ T-cell infiltration was negatively associated with KDM5B, especially in HPV+ head and neck cancer (correlation score −0.458) (Fig 7D and S8C Fig). Lastly, we found a positive correlation between CD8+ T-cell infiltration level and patient survival and a negative correlation between KDM5B expression and patient survival in HPV+ head and neck cancer (Fig 7E). These data show that tumors with high KDM5B expression levels present with low STING expression, suppressed interferon response, and decreased tumor-infiltrating lymphocytes, especially in the presence of abundant cytosolic DNA. As a result, high KDM5B expression is associated with poor prognosis, suggesting KDM5B as a potential target of immunotherapy.
Here, we identified a novel epigenetic regulatory mechanism that tumor cells use to avoid damage caused by cytosolic DNA-triggered innate immune response. Specifically, expression of STING, a key component of the interferon pathway, was silenced by KDM5 family demethylases through removal of H3K4me3 from the STING locus. Suppression of STING by KDM5 demethylase blocked the signal transduction initiated by cytosolic DNA and mediated by the cGAS-STING-TBK1-IRF3 axis (Fig 7F). Inhibition or depletion of KDM5B and KDM5C—by small-molecule inhibitors, siRNA-mediated knockdown, or CRISPR/Cas9-mediated knockout—enhanced STING expression and activated ISGs. The enhanced STING expression was dependent on the activity of H3K4 methyltransferases. This epigenetic regulation allows for a fast, robust, and reversible control of the interferon pathway and is thus expected to have major implications in controlling infection by DNA-containing pathogens and treating cancer.
Robust activation of the cGAS/STING pathway requires not only STING activation by cGAMP—generated by cGAS after it binds pathogen-derived or abnormal self-DNA in the cytosol—but also sufficient STING protein to mediate the signal cascade. Although cytosolic DNA is commonly found in tumor cells [44, 52–56], cGAS-STING signaling is disrupted or silenced in many tumors, enabling cancer cells to evade immunosurveillance [12–15]. Recent studies showed that the expression levels of cGAS and STING were inversely correlated with DNA methylation and can be activated by a DNA methyltransferase (DNMT) inhibitor in a subset of colorectal cancer and melanoma cells [12–14], indicating that DNA methylation contributes to silencing of the cGAS-STING pathway. Here, we found that STING was up-regulated by KDM5 inhibitors in a panel of cell lines, and the expression levels of KDM5B and STING were negatively associated in multiple tumor datasets. These results suggest that regulation of STING by KDM5 is another common mechanism to modulate the cGAS/STING pathway.
Epigenetic changes contribute to tumorigenesis through reprogramming of gene expression profiles [57]. Alternations of epigenetic marks, caused by dysregulation of their writers and erasers, are reversible [58]. This makes epigenetic regulators very attractive drug targets. In fact, inhibitors of epigenetic regulators are either approved or under extensive clinical development, such as inhibitors against DNMTs, Enhancer of zeste homolog 2 (EZH2), histone deacetylases (HDACs), and bromodomain proteins. Emerging evidence shows that, in addition to their effects on tumor cells, these inhibitors also affect the tumor microenvironment, including immune cells [59]. Previous studies, including ours, have shown that KDM5 family histone demethylases, especially KDM5A and KDM5B, are highly expressed and promote tumorigenesis in multiple cancer types [17, 21–29]. The mechanisms for their up-regulation in cancer remain largely unknown. KDM5B was identified as a gene up-regulated by HER2 in human breast cancers [19]. KDM5B undergoes post-translational modifications such as SUMOylation by small ubiquitin-like modifier protein (SUMO) E3 ligase hPc2 and ubiquitination by ubiquitin E3 ligase RNF4 that mediates KDM5B for proteasomal degradation [60]. KDM5B and KDM5C are also regulated by microRNA (miRNA)-137 and miRNA-138, respectively. Both miRNAs are down-regulated in several breast cancer cell lines compared with nontumorigenic human mammary epithelial cell line MCF10A, consistent with the higher expression levels of KDM5B and KDM5C in these cancer cells [61]. In line with the oncogenic roles of KDM5A and KDM5B, suppression of KDM5A or KDM5B delays tumor formation, metastasis, and drug resistance in breast, lung, melanoma, and gastric cancers [17, 21–29]. Although inhibition of KDM5C could have adverse effects on neuronal circuits [62] or promote tumor formation in clear cell renal carcinoma [63] and cervical cancer [64], KDM5C was also shown to have oncogenic roles in prostate cancer [65].
Small-molecule inhibitors of KDM5 enzymes have been developed for cancer treatment [30, 33, 34, 66, 67]. Here, we find that KDM5 inhibitors trigger a robust interferon response through a STING-dependent manner. Further development of these inhibitors could lead to a new class of cancer immunotherapeutic drugs. The cGAS/STING pathway has been targeted in the clinic to induce both innate immune response and subsequent adaptive immune response for cancer treatment. Small-molecule agonists of STING induce systemic immune responses and regression of established tumors in mice [10, 68]. However, this strategy is predicted to have limited efficacy in tumors with abnormal cytosolic DNA but silenced STING. In these tumors, such as HPV+ head and neck or cervical tumors, KDM5 inhibitors could be used to restore STING expression and induce antitumor immune responses. Furthermore, while immune checkpoint inhibitors have achieved remarkable success, most patients do not respond to these treatments. A major mechanism of intrinsic resistance to these treatments is due to lack of T-cell infiltration, which could be induced by STING activation. In fact, inhibition of the cGAS/STING pathway prevents the therapeutic effects of immune checkpoint blockade in a mouse model [69]. Therefore, KDM5 inhibitors, or a combination of STING agonists and KDM5 inhibitors, could maximize the antitumor immune response and allow for effective treatment of nonresponders to the current immunotherapies.
Antibody for KDM5A was described previously [26]. The following antibodies were obtained commercially: rabbit anti-histone H3 (ab1791) (Abcam, Cambridge, UK); rabbit anti-KDM5B (HPA027179) (Sigma, St. Louis, MO); mouse anti-tubulin (T5168) (Sigma, St. Louis, MO); mouse anti-STAT1 (sc-345), −STAT2 (sc-514193), and −IRF9 (sc-135953) (Santa Cruz, Dallas, TX); goat anti-Hsp60 (sc-1052) (Santa Cruz, Dallas, TX); rabbit anti-KDM5C (A301-034A) (Bethyl, Montgomery, TX); rabbit anti-H3K4me3 (C42D8), −H3K4me1 (D1A9), −H3K4me2 (C64G9), −H3K9me3 (D4W1U), −H3K27me3 (C36B11), −H3K36me3 (D5A7), −RIG-I (D14G6), −MDA5 (D74E4), −STING (D2P2F), −cGAS (D1D3G), −IRF3 (D83B9), −TBK1 (D1B4), −MAVS (3993), −IRF7 (4920), −Phospho-STAT1 (58D6), and −HA (C29F4) (Cell Signaling Technology, Danvers, MA); and mouse anti-dsDNA (MAB1293) (Millipore, Burlington, MA).
pcDNA3.1-3xHA-KDM5B construct was described previously [66]. An H499A mutation was introduced into KDM5B plasmid by site-directed mutagenesis. pcDNA3.1-3xHA-STING construct was generated by PCR amplification of the full length of STING coding sequence from cDNA and inserting into pcDNA3.1-3xHA vector between BamHI and XhoI sites. Compound OICR-9429, VE821 was purchased from Sigma. LMB was purchased from Santa Cruz (Dallas, TX) (sc-202210). KDM5-C70 (NCGC00371443) was purchased from Xcess Biosciences (San Diego, CA). Compounds Dong-A-167 (NCGC00487054), GDC-50 (NCGC00482457) [33], and CPI-48 (NCGC00488278) [34] were prepared according to patents WO2016/68580, WO2016/57924, and WO2015/135094, respectively.
The linked KDM5A JmjN-JmjC catalytic domain was prepared and purified by 3-column chromatography utilizing affinity, anion exchange, and sizing exclusion as previously described in detail [70]. The purified protein, in 20 mM Hepes (pH 8.0), 300 mM NaCl, 5% glycerol, and 0.5 mM tris (2-carboxyethyl)phosphine (TCEP), was mixed with MnCl2 and αKG at an approximate molar ratio of 1:5 and concentrated to approximately 10 mg/ml (280 μM) for co-crystallization as described [30]. Inhibitor CPI-48 was soaked into these preformed crystals of KDM5A-αKG-Mn(II) complexes by transferring a crystal into a new drop containing mother liquor (1.2–1.35 M [NH4]2SO4, 0.1 M Tris-HCl [pH 8.6–9.2], 0%–20% glycerol, and 25 mM [Na/K] dibasic/monobasic phosphate) and CPI-48 (approximately 500 μM), allowing the crystal to remain in this drop overnight for CPI-48 to exchange with αKG. The crystals were then mounted into nylon cryoloops (Hampton Research, Aliso Viejo, CA) and frozen in liquid nitrogen after the addition of more glycerol (up to approximately 30% total) to the mother liquor as a cryoprotectant. X-ray diffraction data were collected SER-CAT beam-line 22-ID at the Advanced Photon Source at Argonne National Laboratory at 100 K with 1-degree oscillation images, and the structure was determined by molecular replacement and refinement performed as described (S1 Table) [30].
AlphaLISA assays were performed and analyzed as described previously [30] with 25 nM KDM5A (BPS Biosciences, San Diego, CA; 50110), 10 nM KDM5B (1–755) ΔAP [70], 20 nM KDM5B [30, 66], or 25 nM KDM5C (BPS Biosciences, San Diego, CA; 50112).
MCF7 and BT474 cells were cultured in RPMI1640 supplemented with 10% fetal bovine serum and 1% penicillin and streptomycin. SKBR3 cells were cultured in Dulbecco’s Modified Eagle Medium supplemented with 10% fetal bovine serum and 1% penicillin and streptomycin. siRNA transfections were performed using RNAiMAX (Invitrogen), and plasmid transfections were performed using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. The sequence of dsDNA90 was described previously [71], transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions. siRNA universal negative control 1 and 2 were purchased from Sigma (SIC001 and SIC002). siKDM5A targeting sequences were described previously [29]. Other siRNA targeting sequences were as follows: siKDM5B-1, CAGTGAATGAGCTCCGGCA; siKDM5B-2, GGAGCTGACATTGCCTCAA; siKDM5C-1, GGAGGAAGGTGGTTATGAA; and siKDM5C-2, GGAGGAAGGTGGTTATGAA.
Histone extraction was conducted as described previously [66].
sgRNAs were designed using CHOPCHOP (https://chopchop.rc.fas.harvard.edu/) and cloned into LentiCRISPRv2. Knockout cells were generated as described previously [31]. Briefly, 293T cells in 6-well plates were introduced with 1.5 μg lentiviral plasmid, 1 μg psPAX2, and 0.5 μg pMD2.G. At 48 hours after transfection, lentivirus-containing media were collected and filtered through a 0.45 μm filter before being used to infect cells. Cells were infected with lentivirus for 24 hours, then refed with fresh medium with puromycin. sgRNA controls were described previously [67]. Other sgRNA targeting sequences are listed in S2 Table.
T7 endonuclease assays were conducted as described previously [31]. The primers for amplifying the region flanking TLR3 sgRNA targeting site were as follows: TLR3-F, TCATGAGACAGACTTTGCCTTG; and TLR3-R, GGCTATACCTTGTGAAGTTGGC.
Vaccinia viruses are recombinant vaccinia virus (vTF7-3, strain WR) expressing T7 RNA polymerase [72]. They were kindly provided by Linda Buonacore and Dr. John Rose (Yale University, New Haven, CT). VSV-GFP viruses (VSV-G/GFP, Indiana strain) were generated as described previously [73]. MCF7 cells were infected and incubated at MOI indicated in the figure legends for the indicated time. FACS analyses were performed using a Stratedigm 13-color cytometer with cells fixed in 4% paraformaldehyde. FACS plots were first gated on live cells before analyzing viral GFP fluorescence. Viral copy numbers of vaccinia virus were determined by quantification of pox14KD [74].
For immunostaining, cells were seeded on coverslips, fixed with 4% paraformaldehyde for 10 minutes, permeabilized with 0.4% Triton in PBS for 5 minutes, and then blocked with 10% FBS before incubation with primary antibodies at 4°C overnight. dsDNA staining and image processing were performed according to previous studies [54, 55]. For DNase I–treated samples, cells were permeabilized with 10 μg/ml digitonin and 50 μg/ml DNase I for 30 minutes at 37 °C before fixation with 4% paraformaldehyde. Z-stack images were taken using Leica SP5 confocal microscope. Surface rendering of 3D Z-stacks were processed using Huygens with threshold levels set based on DNase I–treated samples.
ChIP assays were conducted as described previously [75]. Total RNA was isolated using RNeasy Plus Mini Kit (Qiagen, Hilden, Germany). Reverse transcription was performed using High-Capacity cDNA Reverse Transcription Kit (ABI, Sterling, VA). For both ChIP-qPCR and RT-qPCR, qPCR analyses were performed in triplicate using Fast SYBR Green Master Mix (Applied Biosystems, Foster City, CA).
The primers for RT-qPCR analysis of ISG15, RIG-I, MDA5, IFNβ, IFNλ1, and IFNλ2 were described previously [76]. Other primers for RT-qPCR are listed in S3 Table. The primers for ChIP-qPCR are listed in S4 Table.
MCF7 cells were treated with 3 μM of KDM5-C70 or CPI-48 for 6 days. Total RNA was isolated using RNeasy Plus Mini Kit (Qiagen, Hilden, Germany). mRNA libraries for sequencing were prepared according to the standard Illumina protocol. Sequencing (100 bp, paired-end) was performed using Illumina HiSeq 2000 sequencing system at the Genomics Core of Yale Stem Cell Center. RNA-seq data were deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus database under accession number GSE108502.
The RNA-seq reads were mapped to human genome (hg38) with Bowtie2 [77] in local mode, which allows the reads spanning the exon–exon junctions to get mapped to one of the 2 exons (whichever gives the higher mapping score) independent of the transcriptome annotation. The uniquely mapped reads (cutoff: MAPQ >10) were counted to ENCODE gene annotation (version 24) [78] using FeatureCounts [79]. Differential gene expression was performed with DESeq2 [80].
Gene expression profiles of DMSO- or KDM5-inhibitor–treated cells were used for GSEA using GSEA version 2.0 software [81]. The gene set database of h.all.v6.1.symbols.gmt (Hallmarks) was used. Statistical significance was assessed by comparing the enrichment score to enrichment results generated from 10,000 random permutations of the gene set.
TCGA expression datasets were downloaded using the Broad Institute Firehose application programming interface (https://gdac.broadinstitute.org). Expression data are in log2 RSEM format. For each TCGA dataset, primary tumor samples were ranked by their expression of KDM5B and evenly divided into 4 groups. Samples with KDM5B expression less than the first quartile were deemed “KDM5B low,” while samples with KDM5B expression greater than or equal to the third quartile were deemed “KDM5B high.” Statistical comparisons were performed between the STING expression of the samples in “KDM5B low” and “KDM5B high” groups. Significance was computed using the Student t test. For box plots, the lower and upper hinges signify the first and third quartiles, respectively, while the center line depicts the median. The whisker tips correspond to the first observation beyond 1.5 times the interquartile range. Outliers are illustrated with points. R scripts are available upon request. The correlation between KDM5B and clinical impact in HPV-positive head and neck cancer or cervical cancer were analyzed using a web server TIMER (https://cistrome.shinyapps.io/timer/) [82, 83]. The correlation between KDM5B and STING, KDM5B and CXCL10, or STING and CXCL10 were adjusted by tumor purity.
KDM5B and input ChIP-seq data were obtained from the ENCODE K562 dataset (GSE29611) in bigwig format. KDM5C wild-type and knockout ChIP-seq data were obtained from GSE71327 [42], aligned with Bowtie2, and processed into bigwig using Deeptools [84]. All signal tracks were visualized using IGV [85].
Statistical significance was determined using the unpaired Student t test. Error bars represent SEM. SEM was calculated from triplicate technical replicates of each biological sample or 2 or 3 biological replicates. Data shown were representative of 3 independent experiments or biological replicates as indicated in figure legends.
|
10.1371/journal.pbio.2005372 | Is the sky the limit? On the expansion threshold of a species’ range | More than 100 years after Grigg’s influential analysis of species’ borders, the causes of limits to species’ ranges still represent a puzzle that has never been understood with clarity. The topic has become especially important recently as many scientists have become interested in the potential for species’ ranges to shift in response to climate change—and yet nearly all of those studies fail to recognise or incorporate evolutionary genetics in a way that relates to theoretical developments. I show that range margins can be understood based on just two measurable parameters: (i) the fitness cost of dispersal—a measure of environmental heterogeneity—and (ii) the strength of genetic drift, which reduces genetic diversity. Together, these two parameters define an ‘expansion threshold’: adaptation fails when genetic drift reduces genetic diversity below that required for adaptation to a heterogeneous environment. When the key parameters drop below this expansion threshold locally, a sharp range margin forms. When they drop below this threshold throughout the species’ range, adaptation collapses everywhere, resulting in either extinction or formation of a fragmented metapopulation. Because the effects of dispersal differ fundamentally with dimension, the second parameter—the strength of genetic drift—is qualitatively different compared to a linear habitat. In two-dimensional habitats, genetic drift becomes effectively independent of selection. It decreases with ‘neighbourhood size’—the number of individuals accessible by dispersal within one generation. Moreover, in contrast to earlier predictions, which neglected evolution of genetic variance and/or stochasticity in two dimensions, dispersal into small marginal populations aids adaptation. This is because the reduction of both genetic and demographic stochasticity has a stronger effect than the cost of dispersal through increased maladaptation. The expansion threshold thus provides a novel, theoretically justified, and testable prediction for formation of the range margin and collapse of the species’ range.
| The flow of genetic diversity across environments has conflicting effects. On the beneficial side, it increases the genetic variation that is necessary for adaptation and counters the loss of genetic diversity due to genetic drift. However, it may also swamp adaptation to local conditions. This interplay is crucial for the expansion of a species’ range. In this work, I develop a theory that shows how range expansion depends on two dimensionless parameters: (i) the fitness cost of dispersal—a measure of environmental heterogeneity—and (ii) the strength of genetic drift—a measure of the reduction of genetic diversity. The more heterogeneous an environment, the more challenging it is to expand into, and the lower the genetic diversity, the more limited is the scope for potential adaptation. Together, these two parameters define an ‘expansion threshold’: adaptation fails when the number of individuals accessible by dispersal within one generation is so small that genetic drift reduces genetic diversity below that required for adaptation to a heterogeneous environment. This threshold provides a novel, theoretically justified, and testable prediction for the formation of a range margin and for the collapse of a species’ range in two-dimensional habitats.
| Species’ borders are not just determined by the limits of their ecological niche [1, 2]. A species’ edge is typically sharper than would be implied by continuous change in the species’ environment (reviewed in [3, Table 2]). Moreover, although species’ ranges are inherently dynamic, it is puzzling that they typically expand rather slowly [4]. The usual—but tautological—explanation is that lack of genetic variation at the range margin prevents further expansion [5]. Indeed, a species’ range edge is often associated with lower neutral genetic variation [3, 6–11], suggesting that adaptive genetic variation may be depleted as well [12]. Yet why would selection for new variants near the edge of the range not increase adaptive genetic variance, thereby enabling it to continuously expand [5, 13]? Haldane [14] proposed a general explanation: even if environmental conditions vary smoothly, ‘swamping’ by gene flow from central to marginal habitats will cause more severe maladaptation in marginal habitats, further reducing their population density. This would lead to a sharp edge to a species’ range, even if genetic variance at the range margin is large. However, the consequences of dispersal and gene flow for evolution of a species’ range continue to be debated [15–18]: a number of studies suggest that intermediate dispersal may be optimal [19–23]. Gene flow across heterogeneous environments can be beneficial because the increase of genetic variance allows the population to adapt in response to selection [13]. Current theory identifies that local population dynamics, dispersal, and evolution of niche-limiting traits (including their variance) and both genetic and demographic stochasticity are all important for species’ range dynamics [13, 19–21, 24–28]. Yet these core aspects have not been incorporated into a single study that would provide testable predictions for range limits in two-dimensional habitats.
As Haldane [14] previously pointed out, it is important to consider population and evolutionary dynamics across a species’ range jointly, as their effects interact. Due to maladaptation, both the carrying capacity of the habitat and the population growth rate are likely to decrease—such selection is called ‘hard’ [29]. Classic deterministic theory [24] shows that when genetic variance is fixed, there are two stable regimes of adaptation to a spatially varying optimum (see Fig 1): (i) a ‘limited adaptation’, in which a population is only adapted to a single optimum or becomes a patchy conglomerate of discrete phenotypes, or (ii) continuous or ‘uniform’ adaptation, which is stable when the genetic variance, measured in terms of its cost in fitness (standing genetic load), is large relative to the maladaptation incurred by dispersal between environments (dispersal load). Under uniform adaptation, a species’ range gradually expands—a stable boundary only forms when the genetic variance is too small to allow continuous adaptation to the spatially variable environment, and hence, limited adaptation is stable.
When genetic variance can evolve, such a limit no longer exists in infinitely large populations: the population maintains continuous adaptation as the environmental gradient steepens [13]. Deterministic theory thus predicts that a sharp and stable boundary to a species’ range does not form when the environment changes smoothly. Uniform adaptation is the only stable regime when genetic variance can freely evolve in the absence of genetic drift [13], yet there is a limit to the steepness of the gradient. This limit arises because both the standing genetic load and the dispersal load increase as the gradient steepens, reducing the mean fitness (growth rate) of the population: when the mean fitness approaches zero, the population becomes extinct. Obviously, ignoring genetic drift is then unrealistic. In finite populations, genetic drift reduces local genetic variance [32], potentially qualitatively changing the dynamics. Indeed, it has been shown that for one-dimensional habitats (such as rivers), a sharp range margin arises when the fitness cost of dispersal across environments becomes too large relative to the efficacy of selection versus genetic drift [26]. However, most species live in two-dimensional habitats. There, allele frequencies can fluctuate over a local scale, as the correlations between them decline much faster across space than they do in linear habitats [33], and the effect of genetic drift changes qualitatively, becoming only weakly dependent on selection [34]. Is there still an intrinsic threshold to range expansion in finite populations when dispersal and gene flow occur over two-dimensional space rather than along a line? If so, what is its biological interpretation?
I study the problem of intrinsic limits to adaptation in a two-dimensional habitat. Throughout, I assume that the species’ niche is limited by stabilising selection on a composite phenotypic trait. This optimum varies across one dimension of the two-dimensional habitat—such as temperature and humidity with altitude. Demography and evolution are considered together. Selection is ‘hard’: both the rate of density-dependent population growth and the attainable equilibrium density decrease with increasing maladaptation. Both trait mean and genetic variance can freely evolve via change in allele frequencies and the associations among them (linkage disequilibria). The populations are finite, and both genetic and demographic stochasticity are included. The model is first outlined at a population level in terms of coupled stochastic differential equations. While it is not possible to obtain analytical solutions to this model, this formalisation allows us to identify the effective dimensionless parameters that describe the dynamics. Next, individual-based simulations are used to determine the driving relationship between the key parameters and test its robustness. The details are described in the Model section of the Methods.
The dynamics of the evolution of a species’ range, as formalised by this model, are well described by three dimensionless parameters, which give a full description of the system. The first dimensionless parameter carries over from the phenotypic model [24]: the effective environmental gradient B measures the steepness of the environmental gradient in terms of maladaptation incurred by dispersal across a heterogeneous environment. The second parameter is the neighbourhood size of the population, 𝒩, which can be understood as the number of diploid individuals within one generation’s dispersal range. Originally, neighbourhood size was defined by Wright [35] as the size of the single panmictic diploid population that would give the same probability of identity by descent in the previous generation. The inverse of neighbourhood size 1/ 𝒩 hence describes the local increase of homozygosity due to genetic drift. The third dimensionless parameter is the ratio s/r* of the strength of selection s per locus relative to the strength of density dependence, r*. Detailed description of the parameters and their rescaling can be found in the Methods sections Parameters and Continuous model: Rescaling.
In order to see how the rescaled parameters capture the evolution of a species’ range, I simulated 780 evolving populations, each based on different parameterisations, adapting to a linear gradient in the optimum. Depending on the parameters, the population either expands, gradually extending its phenotypic range by consecutive sweeps of loci advantageous at the edges, or the species’ range contracts or disintegrates as adaptation fails. Fig 2 shows the results of the projection from a 10-dimensional parameter space of the individual-based model (see Methods sections Individual-based simulations and Parameters) into a two-dimensional space. The axes of Fig 2 represent the first two compound dimensionless parameters: (i) the effective environmental gradient B and (ii) the inverse of neighbourhood size 1/ 𝒩, which describes the effect of genetic drift on the allele frequencies.
These two dimensionless parameters B and 𝒩 give a clear separation between expanding populations, in which the neighbourhood size 𝒩 is large relative to the effective environmental gradient B (shown in blue, Fig 2), and the rest, in which adaptation is failing. The separation gives an ‘expansion threshold’, estimated at 𝒩 ≈ 6.3B + 0.56 (red line). Above the expansion threshold, populations are predicted to expand (see Fig 3); below it, adaptation fails abruptly. If conditions change uniformly across space (as in these simulation runs, with constant gradient and carrying capacity), this means that adaptation fails everywhere—a species’ range then either collapses from the margins (Fig 2, red hues) and/or disintegrates (Fig 2, open circles), forming a fragmented metapopulation (i.e., a spatially structured population consisting of discrete locally adapted subpopulations with limited dispersal among them).
When a metapopulation forms, it exhibits an extinction and colonisation dynamics. The subpopulations drift freely along the neutral spatial axis. Because the trait distributions of the subpopulations are unstable, the subpopulations also drift slowly along the environmental gradient. Over time, the metapopulation very slowly collapses to a virtually single trait value, with many subpopulaitons along the neutral axis. The subpopulations forming this metapopulation have only a very narrow phenotypic range and maintain locally only minimal adaptive variance. They correspond to the limited adaptation regime identified for a phenotypic model with genetic variance as a parameter [24]. In contrast to one-dimensional habitats [26], these patchy metapopulations are stabilised by dispersal from surrounding subpopulations in the two-dimensional habitat and can thus persist for a long time. An example of such a metapopulation is given in Fig 4.
Interestingly, the third dimensionless parameter s/r* has no detectable effect on the form of the expansion threshold. In other words, whilst the expansion threshold reflects the total fitness cost of dispersal in a heterogeneous environment, it appears independent of the strength of selection per locus s: the dashed lines in Fig 2 compare the estimated expansion threshold for small and large s/r*. Increasing the strength of selection is inefficient in aiding drift-limited adaptation, in line with the expectation that the effect of genetic drift is only very weakly dependent on selection in two-dimensional habitats [27](see also S1 Fig). This suggests that genetic basis of adaptation is not important for a drift-induced limit to a species’ range. Yet it is plausible that there is another limit, in which selection per locus becomes important [27], that arises when the optimum changes abruptly and even when the population (neighbourhood) size is large (i.e., in an entirely different regime). A dedicated synthesis connecting the step-limited and drift-limited regimes would be of a clear interest. Importantly, once genetic drift starts to have an effect, the habitat needs to be fairly broad to be two-dimensional [37]. In narrow habitats (such as in [27]), some dependency of drift-induced expansion threshold on selection per loci would be expected [26]. Note that the apparent independence of the expansion threshold on s/r* does not imply that rate of range expansion should also be independent of the strength of selection.
In nature, conditions are unlikely to change uniformly. Abiotic environment (such as temperature, precipitation, solar radiation) does not, in general, change in a linear and concordant manner [38], and neither does the biotic environment, such as the pressure from competitors and predators, which affects the attainable population density and can increase the asymmetry in gene flow [39, 40]. I now investigate whether adaptation fails near the expansion threshold as conditions change across space. For example, we can imagine that the population starts well adapted in the central part of the available habitat, and as it expands, conditions become progressively more challenging (see S2A Fig); i.e., the effective environmental gradient B gets steeper. As the expanding population approaches the expansion threshold, adaptive genetic variance progressively decreases below the predicted value [13], which would be maintained by gene flow in the absence of genetic drift (Fig 5A, grey dashed line). This is a result of an increased frequency of demes with limited adaptation, leading to higher rates of extinctions and recolonisations, which reduce both adaptive and neutral diversity (see Fig 5B). Range expansion then ceases at the expansion threshold as the genetic variance drops to the critical value at which only limited adaptation is stable [24], assuming genetic variance is fixed (Fig 5A, dotted line). This is because although populations can persist with limited adaptation (Fig 4), the transient amount of genetic variance maintained under limited adaptation is almost never consistent with range expansion (see Fig 2, open circles). On a steepening gradient, a sharp and stable range margin forms. This contrasts to uniformly changing conditions (linear gradient, constant carrying capacity) in which populations steadily expand or contract.
In a large population, the ability to adapt to heterogeneous environments is independent of dispersal: this is because both the local genetic variance (measured by standing genetic load), which enables adaptation to spatially variable environments, and the perceived steepness of the environmental gradient (measured by dispersal load) increase at the same rate with gene flow [13]. Yet, in small populations, dispersal is beneficial because the drift-reducing effect of dispersal overpowers its maladaptive effect. This is demonstrated in Fig 6—the neighbourhood size 𝒩 increases faster with dispersal than the effect of swamping by gene flow (B) does; hence, as dispersal increases, the population gets above the expansion threshold at which uniform adaptation can be sustained. Around the expansion threshold, a small change in dispersal (connectivity) can have an abrupt effect on adaptation across a species’ range and the species’ persistence. A small increase in dispersal can lead to recovery of uniform adaptation with an arbitrarily wide continuous range. Further increase of dispersal is merely enhancing the rate of range expansion at the expense of a slight cost to the mean fitness due to rising dispersal load and standing load and can be associated with further costs, such as Allee effect (see, e.g., [17]). Therefore, the expansion threshold provides an interpretation for optimality of an ‘intermediate’ dispersal, benefiting the species’ persistence.
Here, I have shown that adaptation fails when positive feedback between genetic drift, maladaptation, and population size reduces adaptive genetic variance to levels that are incompatible with continuous adaptation. The revealed expansion threshold differs qualitatively from the limit to adaptation identified previously [26] for a population living along a one-dimensional habitat. This is because in two dimensions, dispersal mitigates the loss of diversity due to genetic drift more effectively, such that it becomes (almost) independent of selection [34]. The expansion threshold implies that populations with very small neighbourhood sizes (𝒩 ⪅ 1/2), which suffer a severe reduction in neutral heterozygosity, will be prone to collapse based on demographic stochasticity alone. However, even in the absence of demographic stochasticity, genetic drift reduces the adaptive genetic variance required to sustain adaptation to a heterogeneous environment. The expansion threshold describes when this reduction due to genetic drift is incompatible with continuous adaptation, predicting a collapse of a species’ range. If the expansion threshold is reached as the species expands through its habitat, a sharp and stable range margin forms. If there is a drop below the expansion threshold throughout the species’ range, as after a sudden drop in carrying capacity, adaptation abruptly collapses throughout a species’ range. The result is either extinction or a fragmented metapopulation consisting of a conglomerate of subpopulations, each adapted to a single phenotypic optimum. It follows that near a range margin, we expect increased range fragmentation and a decrease in adaptive genetic variance. The threshold gives a theoretical base to the controversial issue of the importance of evolution (genetics) and ecology (demography) for assessing vulnerability of a species [41, 42]. The predicted sharp species’ range edge is in agreement with the reported lack of evidence for ‘abundant centre’ of a species’ range, which, although commonly assumed in macroecological theory, has little support in data [3, 11, 43, 44]. Lack of abundant centre is consistent both with uniform adaptation and with limited adaptation in a metapopulation.
The expansion threshold provides a general foundation to species-specific eco-evolutionary models of range dynamics [45]. Its components can be measured in wild populations, allowing us to test the robustness of the theory. First, the effective environmental gradient B can be measured as fitness loss associated with transplant experiments on a local scale, relative to a distance of generational dispersal along an environmental gradient. The environmental gradient can include both biotic and abiotic effects and their interactions [46]—notably, the effective environmental gradient B steepens due to increased asymmetry in gene flow when carrying capacity varies across space, e.g., because of partial overlap with competitors [40]. Second, the neighbourhood size 𝒩 can be estimated from neutral allele frequencies [47, 48]. Estimates of neighbourhood size are fairly robust to the distribution of dispersal distances [49]. Though near the expansion threshold, both the noisiness of the statistics and the homozygosity will increase due to local extinctions and recolonisations [50]. An alternative estimate of neighbourhood size can be also obtained from mark-recapture studies by measuring population density and dispersal (as an approximation for gene flow) independently [47].
Because the expansion threshold is free of any locus- or trait- specific measure, the result appears independent of genetic architecture, readily extending to multiple traits regardless of their correlations (compare to [51–55])—yet the mean fitness will decline because of ‘drift load’ as the number of traits independently optimised by selection increases [56, 57]. Hence, if the fitness landscape is highly complex, the expansion threshold constitutes a lower limit. Naturally, there can be further costs arising in a natural population that I have neglected here, such as the Allee effect [17]. In general, while the numerical constants may change when natural populations deviate in their biology from our model assumptions, the scale-free parameters identified in this study remain core drivers of the intrinsic dynamics within a species’ range. Notably, the early classic studies assuming fixed genetic variance [24] predicted that dispersal into peripheral populations is detrimental because it only inflates the effective environmental gradient B. Yet, when genetic variance can evolve, dispersal into small marginal populations also aids adaptation by increasing local genetic variance and by countering genetic drift. The net effect of dispersal into small marginal populations (below the expansion threshold) is then beneficial because their neighbourhood size increases faster with dispersal than the effective environmental gradient B steepens.
I model evolution of a species’ range in a two-dimensional habitat, in which both population dynamics and evolution (in many additive loci) are considered jointly. The coupling is via the mean fitness r¯(z¯,N), which gives the growth rate of the population, and decreases with increasing maladaptation: r¯(z¯,N)=re(N)+r¯g(z¯). The ecological component of growth rate, re, can take various forms: here, the regulation is logistic so that fitness declines linearly with density N: re = rm(1−N/K), in which rm is the maximum per capita growth rate in the limit of the local population density N→0. The carrying capacity K (for a perfectly adapted phenotype) is assumed uniform across space. The second term, rg(z¯)≤0, is the reduction in growth rate due to deviation from the optimum. Selection is stabilising: the optimum θ changes smoothly with one spatial dimension (x): for any individual, the drop in fitness due to maladaptation is rg(z) = −(z−θ)2/(2Vs). Here, Vs gives the width of stabilising selection; strength of stabilising selection is γ = −VP/(2Vs), in which VP = VG+VE is the phenotypic variance. A population with mean phenotype z¯ has its fitness reduced by r¯g(z¯)=−(z¯−θ)2/(2Vs)−VP/(2Vs). The phenotype z is determined by many di-allelic loci with allelic effects αi; the model is haploid, hence z¯=∑iαipi, in which pi is the allele frequency at locus i. Phenotypic variance is VP = VG+VE. The loss of fitness due to environmental variance VE can be included in rm*=rm−VE/(2Vs); VE is a redundant parameter. Selection is ‘hard’: both the mean fitness (growth rate) and the attainable equilibrium density N^=Kr*/rm=K(1−VG/(2rmVs)) decrease with maladaptation. Expected genetic variance maintained by gene flow in the absence of genetic drift is VG=bσVs [13]; the contribution due to mutation is small, at mutation-section balance VG,μ/s=2μVsnl, in which μ gives the mutation rate per locus and nl the number of loci.
Discrete-time individual-based simulations are set to correspond to the model with continuous time and space. The space is a two-dimensional lattice with spacing between demes of δx = 1. Every generation, each individual mates with a partner drawn from the same deme, with probability proportional to its fitness, to produce a number of offspring drawn from a Poisson distribution with mean of Exp[r(z, N)] (this includes zero). The effective diploid population density Ne hence equals half of the haploid population density N, and 𝒩 = 4πNe σ2 = 2πNσ2. The life cycle is selection → mutation → recombination → birth → migration. Generations are nonoverlapping, and selfing is allowed at no cost. The genome is haploid with unlinked loci (the probability of recombination between any two loci is 1/2). The allelic effects αi of the loci combine in an additive fashion; the allelic effects are uniform throughout this study, αi ≡ α. Mutation is set to μ = 10−6, independently of the number of loci. Migration is diffusive with a Gaussian dispersal kernel. The tails of the dispersal kernel need to be truncated: truncation is set to two standard deviations of the dispersal kernel throughout, and dispersal probabilities and variance are adjusted so that the discretised dispersal kernel sums to 1 [58]. Simulations were run at the computer cluster of IST Austria using Mathematica 9 (Wolfram). The code for the simulations, together with a working example, have been deposited as a single *.cdf file at Dryad Digital Repository, https://doi.org/10.5061/dryad.5vv37 [36]. This file can be viewed with CDF Player, a free application developed by Wolfram Research, and also contains all the figures with their underlying data.
For any given additive genetic variance VG (assuming a Gaussian distribution of breeding values), the change in the trait mean z¯ over time satisfies:
∂z¯∂t=σ22(∂2z¯∂x2+∂2z¯∂y2)+σ2(∂2ln(N)∂x∂z¯∂x+∂2ln(N)∂y∂z¯∂y)+VG∂r¯∂z¯+ζ.
(1)
The first term gives the change in the trait mean due to migration with mean displacement of σ; the second term describes the effect of the asymmetric flow from areas of higher density. The third term gives the change due to selection, given by the product of genetic variance and gradient in mean fitness [59, Eq 2]. The last term ζ gives the fluctuations in the trait variance due to genetic drift: ζ=VG,LE/NdWζ(x,y,t), in which dW* represents white noise in space and time [34, 60]. VG,LE=∑iαi2piqi denotes genetic variance assuming linkage equilibrium.
The trait mean is z¯=∑iαipi for a haploid model, in which pi is the i-th allele frequency, qi = 1−pi and αi is the effect of the allele on the trait—the change of the trait mean z¯ as frequency of locus i changes from 0 to 1. For both haploid and diploid models, the allele frequencies pi change as:
∂pi∂t=σ22(∂2pi∂x2+∂2pi∂y2)+σ2(∂pi∂x∂ln(N)∂x+∂pi∂y∂ln(N)∂y)+piqi∂r¯∂pi−μ(pi−qi)+ε.
(2)
The expected change of allele frequency due to a gradient in fitness and local heterozygosity is piqi∂r¯∂pi=sipiqi(pi−qi−2Δi), in which selection at locus i is si≡αi2/(2Vs) and Δi=(z¯−bx)/αi [13, Appendix 3]. Here, the fourth term describes the change due to (symmetric) mutation at rate μ. The last term ɛ describes genetic drift [34, Eq 7]: ε=piqiNdWε(x,y,t), in which N is the haploid population density.
Population dynamics reflect diffusive migration in a two-dimensional habitat, growth due to the mean Malthusian fitness r¯, and stochastic fluctuations. The number of offspring follows a Poisson distribution with mean and variance of N; fluctuations in population numbers are given by [61]: ξ=NdWξ(x,y,t):
∂N∂t=σ22(∂2N∂x2+∂2N∂y2)+r¯N+ξ
(3)
The model can be simplified by rescaling [13, 59] time t relative to the strength of density dependence r*, distance x relative to dispersal σ, trait z relative to strength of stabilising selection 1/(2Vs) and local population size N relative to equilibrium population size with perfect adaptation: N^=Kr*/rm, T = r*t,X=x2r*σ2,Z=zr*Vs,N~=N/N^. Note that near the equilibrium of a well-adapted population, N~≈1.
The rescaled equations for evolution of allele frequencies and for demographic dynamics are
∂N˜∂T=∂N˜∂X2+∂N˜∂Y2+R¯N˜+2N˜N^σ2dWς˜(X,Y,T)∂pi∂T=∂2pi∂X2+∂2pi∂Y2+2(∂pi∂X∂ln(N˜)∂X+∂pi∂Y∂ln(N˜)∂Y)++sr*(piqi−2Z¯−BXα*)−μr*(pi−qi)+piqiN˜N^σ2dWε˜(X,Y,T)
(4)
in which R-≡r-/r*=1-N~-BX-Z2/2.
The rescaled Eqs 4 show that four parameters fully describe the system. First, the effective environmental gradient, B≡bσ/(r*2Vs). Second, the strength of genetic drift 1/N^=1/(2πN^σ2). The parameter N^ gives the neighbourhood size at an equilibrium with uniform adaptation. The third parameter is the strength of selection relative to the strength density dependence, s/r*; the scaled effect of a single substitution α* also scales with s/r*: α*≡α/r*Vs=2s/r*. The effect of this third parameter s/r* is expected to be small, because typically s≪r*. Therefore, assuming throughout that s is uniform across loci is a reasonably justified simplification. The fourth parameter, μ/r*, will typically be very small and will be neglected throughout. Table 1 (top) summarises the full set that describes the system.
|
10.1371/journal.pgen.1000793 | Inverse Correlation between Promoter Strength and Excision Activity in Class 1 Integrons | Class 1 integrons are widespread genetic elements that allow bacteria to capture and express gene cassettes that are usually promoterless. These integrons play a major role in the dissemination of antibiotic resistance among Gram-negative bacteria. They typically consist of a gene (intI) encoding an integrase (that catalyzes the gene cassette movement by site-specific recombination), a recombination site (attI1), and a promoter (Pc) responsible for the expression of inserted gene cassettes. The Pc promoter can occasionally be combined with a second promoter designated P2, and several Pc variants with different strengths have been described, although their relative distribution is not known. The Pc promoter in class 1 integrons is located within the intI1 coding sequence. The Pc polymorphism affects the amino acid sequence of IntI1 and the effect of this feature on the integrase recombination activity has not previously been investigated. We therefore conducted an extensive in silico study of class 1 integron sequences in order to assess the distribution of Pc variants. We also measured these promoters' strength by means of transcriptional reporter gene fusion experiments and estimated the excision and integration activities of the different IntI1 variants. We found that there are currently 13 Pc variants, leading to 10 IntI1 variants, that have a highly uneven distribution. There are five main Pc-P2 combinations, corresponding to five promoter strengths, and three main integrases displaying similar integration activity but very different excision efficiency. Promoter strength correlates with integrase excision activity: the weaker the promoter, the stronger the integrase. The tight relationship between the aptitude of class 1 integrons to recombine cassettes and express gene cassettes may be a key to understanding the short-term evolution of integrons. Dissemination of integron-driven drug resistance is therefore more complex than previously thought.
| Integrons are widespread bacterial genetic elements able to capture and express gene cassettes that often encode antibiotic resistance determinants. Gene cassettes are usually promoterless and are transcribed from a common promoter, Pc. Pc is located within the coding sequence of the integron integrase, IntI, which is the key element catalyzing the integration and excision of gene cassettes. Several Pc variants, associated with different integrase amino acid sequences, have been described, but the influence of these differences on integrase activity has never been investigated. Here, we show that Pc is highly polymorphic, conferring a wide range of antibiotic resistance. Furthermore, we found that different Pc variants are associated with different integrase excision activities: the weaker the Pc variant, the more active the integrase. These results point to evolutionary compromises between the expression and mobility of drug resistance determinants located on integrons.
| Integrons are natural genetic elements that can acquire, exchange and express genes within gene cassettes. The integron platform is composed of a gene, intI, that encodes a site-specific recombinase, IntI, a recombination site, attI, and a functional promoter, Pc, divergent to the integrase gene [1] (Figure 1). Gene cassettes are small mobile units composed of one coding sequence and a recombination site, attC. Integrons exchange gene cassettes through integrase-catalyzed site-specific recombination between attI and attC sites, resulting in the insertion of the gene cassette at the attI site, or between two attC sites, leading to the excision of the gene cassette(s) from the gene cassette array [2]–[6]. Multi-resistant integrons (MRI) contain up to eight gene cassettes encoding antibiotic resistance. To date, more than 130 gene cassettes have been described, conferring resistance to almost all antibiotic classes [7]. MRI play a major role in the dissemination of antibiotic resistance among Gram-negative bacteria, through horizontal gene transfer [8]. Five classes of MRI have been described on the basis of the integrase coding sequence, class 1 being the most prevalent [8].
Gene cassettes are usually promoterless, and their genes are transcribed from the Pc promoter, as in an operon (Figure 1), the level of transcription depending on their position within the integron [9],[10]. Among class 1 MRIs, several Pc variants have been defined on the basis of their −35 and −10 hexamer sequences. Four Pc variants have been named according to their sequence homology with the σ70 promoter consensus and their estimated respective strengths, as follows: PcS for ‘Strong’, PcW for ‘Weak’ (PcS being 30-fold stronger than PcW), PcH1 for Hybrid 1 and PcH2 for Hybrid 2, these two latter Pc variants containing the −35 and −10 hexamers of PcW and PcS in opposite combinations (Table 1), and having intermediate strengths [11]–[13]. More recently, a new variant was reported to be significantly stronger than PcS [14], and we therefore named it ‘Super-Strong’ or PcSS. Three other Pc variants have been described but their strength has not been determined; for simplicity, we named these Pc promoters PcIn42, PcIn116 and PcPUO, as they are carried by integrons In42 and In116 and by plasmid pUO901, respectively [15]–[17]. Nesvera and co-workers found a C to G mutation 2 bp upstream of the −10 hexamer in PcW and showed that this mutation increased promoter efficiency by a factor of 5 [18]. This mutation creates a ‘TGN’ extended −10 motif that is known to increase the transcription efficiency of σ70 promoters in E. coli [19]. Also, class 1 integrons occasionally harbor a second functional promoter named P2, located in the attI site and created by the insertion of three G residues, optimizing the spacing (17 bp) between potential −35 and −10 hexamer sequences [9] (Figure 1). Given the diversity of Pc variants and the range of their respective strengths, an identical array of gene cassettes should be differently expressed depending on the Pc variant present in the integron platform. However, the distribution of Pc variants among the numerous class 1 integrons has never been comprehensively studied.
In class 1 MRIs, the Pc promoter is located within the integrase coding sequence (Figure 1). Some of the base substitutions in the −35 and/or −10 hexamer sequences defining the different Pc variants actually correlate with amino acid changes in the IntI1 sequence. These variations in the IntI1 protein sequence could potentially influence integrase recombination activity and define different IntI1 catalytic variants.
We first performed an extensive in silico examination of all class 1 integron sequences available in databases in order to determine the prevalence of Pc variants and, therefore, the prevalence of IntI1 variants. We then estimated the strength of all Pc variants and Pc-P2 combinations in the same reporter gene assay, as well as the excision and integration activity of the main IntI1 variants. We found a very unequal distribution of the Pc variants, and a negative correlation between the strength of the Pc variant and the recombination efficiency of the corresponding IntI1 protein.
We analyzed the sequences of 321 distinct class 1 integrons containing the complete sequences of both gene cassette arrays and Pc-P2 promoters (see Materials and Methods). When considering only the −35 and −10 hexamer sequences, we found no more than the eight variants identified previously. However, their distribution was highly uneven, four variants (PcW, PcS, PcH1 and PcH2) totalling 98.4% of the sequences analyzed (Table 1). The most frequent Pc variant was PcW (41.7%), followed by PcH1 (28%), PcS (24.3%) and PcH2 (4.4%). The four other Pc variants, all more recently described, were extremely rare (Table 1). The most prevalent Pc variant among class 1 integrons appeared to be the weak PcW, but in 58% of the analyzed PcW-containing integrons this promoter was associated with either a ‘TGN’ extended −10 motif [20] (hereafter designated variant PcWTGN-10) or the second gene cassette promoter P2 (Table 2). These two features were much less frequent with the other Pc variants (Table 2). The dataset also contained two other extremely rare Pc configurations, designated PcWTAN-10 and PcH1TTN-10, in which the second base upstream of the −10 hexamer was replaced by an A or a T instead of C, respectively, as well as two other rare forms of P2, designated P2m1 and P2m2, for ‘P2 mutated form 1’ and ‘P2 mutated form 2’ (Table 1 and Table 2).
Altogether, on the basis of the −35 and −10 hexamers and the sequence upstream of the −10 box, we identified 13 Pc variants, four of which were also found associated with a form of the P2 promoter (Table 2).
Until recently, the promoter strength of only 4 of the 8 known variants (PcSS, PcS, PcH1 and PcW) had been estimated, but variants strength had never been compared in the same assay [11],[14]. We therefore examined the capacity of all the Pc variants and the different Pc-P2 configurations to drive the expression of the lacZ reporter gene cloned in a transcriptional fusion with a 254-bp fragment containing the Pc variant and the P2 promoter region (see Materials and Methods). We found, in agreement with the results of a previous study [11] and those of another study published during the course of this work [13], that PcS was about 25-fold stronger than PcW and 4.5-fold stronger than PcH1, while PcH2 lay between PcH1 and PcS, being 3.8-fold stronger than PcH1. PcPUO and PcIn42 were of similar strength to PcW, and PcIn116 was very weak (Figure 2A). The PcSS variant, previously described as being stronger than PcS [14], was about 12-fold less efficient in our experimental conditions (Figure 2A). This latter result was not wholly unexpected, as PcSS contains a down-promoter mutation in the −35 hexamer relative to PcS (Table 1; [21]).
We found that the presence of the TGN-10 motif increased PcW efficiency 15-fold, approaching that of PcH2, whereas it had no significant effect on PcS or PcH2 activity (Figure 2B), probably because these promoters are already maximally efficient. On the other hand, the C to A mutation in PcWTAN-10 severely reduced PcW activity (as already observed for the activity of an Escherichia coli promoter [19]), and the C to T mutation in PcH1TTN-10 slightly increased PcH1 efficiency (1.7-fold; data not shown).
To evaluate the contribution of P2 to gene cassette expression, we first created transcriptional lacZ fusion with sequences containing a combination of an inactive PcS (hereafter named PcS*, see Materials and Methods) and the P2 variants, in order to assess their specific strength. We found that P2 was active and 7-fold stronger than PcW (Figure 2C), in keeping with previous studies [11]. P2m1 and P2m2 appeared to be inactive (data not shown) and their influence on gene cassette expression was not investigated further. When the weakest Pc variants (PcW and PcH1) were associated with P2, β-galactosidase activity was increased but was equivalent to that of P2, indicating that, in the PcW-P2 and PcH1-P2 combinations, PcW and PcH1 do not contribute significantly to the expression of gene cassettes, which is mainly driven by P2. By contrast, when P2 was associated with the strongest variants, PcS and PcWTGN-10, β-galactosidase activity decreased slightly (Figure 2C). A recent report described a small increase in the expression of a gene cassette when PcS was combined with P2 [13], but these authors used different methods to measure promoter strength, which may explain the discrepancy with our results.
In class 1 MRIs, Pc is located within the intI1 coding sequence, and several of the substitutions generating the different Pc variants affect the IntI1 amino acid (aa) sequence. The aa changes involve aa 32 or 39 for the main variants and aa 31, 32, 38 and/or 39 for the rare variants (Table 3). Some Pc variants produce the same IntI1 variant, e.g. PcW/PcH1 and PcS/PcH2 (Table 3). Altogether, 10 IntI1 variants are generated from 13 Pc variants, three of which (IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39) represent almost 96% of the IntI1 variants (Table 3).
In order to estimate the impact of the aa differences on IntI1 activity, we first cloned the intI1 gene of the three main IntI1 variants, IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39, under the control of the arabinose-inducible promoter ParaB (see Materials and Methods). However, we anticipated that the two convergent promoters, namely Pc (contained in the intI sequence) and ParaB, might interfere with each other. Thus, to estimate IntI1 protein recombination activity independently of potential promoter interference, we introduced mutations that inactivated the Pc promoters without affecting the IntI1 aa sequence (see Materials and Methods). The resulting integrases were named IntI1*R32_H39, IntI1*R32_N39 and IntI1*P32_H39 (Table 3). We then estimated the excision activity of these integrases by measuring their capacity to catalyze recombination between two attC sites located on a synthetic array of two cassettes, attCaadA7-cat(T4)-attCVCR-aac(6′)-Ib, and resulting in the deletion of the synthetic cassette, cat(T4)-attCVCR, and the expression of tobramycin resistance mediated by the gene aac(6′)-Ib (see Materials and Methods; [22]). As shown in Figure 3, the three integrases exhibited very different excision activities (1.8×10−2 to 1.3×10−5), IntI1*P32_H39 and IntI1*R32_N39 being respectively 336- and 51-fold less efficient than IntI1*R32_H39. Thus, replacing R32 by P32, or H39 by N39 drastically reduces the capacity of the integrase to promote recombination between the attCaadA7 and attCVCR sites. The strongest effect was observed when a proline was present at position 32. P32 is also found in the integrase IntI1P32_N39, a much less frequent variant of IntI1 (Table 3). We therefore created this latter IntI1 variant and measured its excision activity. IntI1*P32_N39 was 27-fold less active than IntI1*R32_N39, showing the same negative effect of P32 on excision activity (Figure 3).
Class 1 integrase is also able to catalyze the integration of gene cassettes by promoting recombination between attI and attC sites [5]. We therefore tested the ability of the different IntI1 variants to catalyze recombination between attI and the two attC sites used for the excision activity assay (attCaadA7 and attCVCR), in an assay based on suicide conjugative transfer previously developed [6] and since extensively used [23]–[25] (see Materials and Methods). Surprisingly, the range of integration activity of the four IntI variants tested in this study was rather narrow (4.5×10−3 to 2.3×10−4) compared to their excision activity, independently of the nature of the attC site (Figure 3). IntI1*R32_H39 and IntI1*R32_N39 exhibited similar integration activities in the two reactions performed, and the R32P substitution appeared to be detrimental for the activity of both integrases, but far less than for their excision activity. This effect seemed a bit stronger with IntI1*P32_N39 than with IntI1*P32_H39 (integration frequency was reduced by roughly 8-fold compared to 3-fold, respectively; Figure 3).
To show that the observed differences in excision and integration activities of the four integrases tested were not due to variations in the amounts of integrase but indeed to the nature of the aa at positions 32 and 39, we performed SDS-Page western blot analysis. We found that IntI1*R32_H39, IntI1*R32_N39 and IntI1*P32_N39 were equally produced and that IntI1*P32_H39 was slightly more strongly expressed in our experimental conditions (Figure S1 and Text S1). However, the latter had one of the weakest recombination activities (Figure 3). Therefore, the observed differences in excision activity among the IntI1 variants were due not to differences in protein abundance but to differences in protein activity and/or folding.
In this study we found marked polymorphism of the gene cassette promoter Pc (13 variants), corresponding to ten variants of the class 1 integrase IntI1. The 13 Pc variants were defined on the basis of the −35 and −10 hexamers and the sequence upstream of the −10 box. Indeed almost 20% of the 321 integrons analyzed here harbored a TGN-10 motif that characterized an extended −10 promoter. This feature was mainly associated with the weak PcW variant (41.8% of PcW-containing integrons) and increased the efficiency of this promoter by a factor of 15. In view of its frequency and its strength difference relative to PcW, we propose that this promoter, designated PcWTGN-10, be considered as a Pc variant distinct from PcW. Furthermore, 9% of the 321 integrons contained the P2 promoter, which was almost exclusively associated with the PcW variant (17.2% of PcW-containing integrons, Table 2). As in previous studies, we found that transcriptional activity was mainly driven by P2 in the PcW-P2 combination [9],[11]. We also observed the same effect with PcH1.
Altogether, there are no fewer than 20 distinct gene cassette promoter configurations for class 1 integrons, but their frequencies are very different. Five main combinations emerged from the dataset, defining five levels of promoter strength. The distribution and strength of the gene cassette promoters were as follows: PcW-P2<PcW≈PcWTGN-10<PcS≈PcH1 (distribution, Table 2) and PcW<PcH1<PcW-P2<PcWTGN-10<PcS (respectively 4.5-, 7-, 15- and 25-fold more active than PcW; Figure 1 and Figure 2).
The multiplicity of gene-cassette promoters displaying different strengths indicates that a given antibiotic resistance gene cassette will be differently expressed depending on which Pc variant is present in the integron. For example, we used an E. coli strain containing a class 1 integron with PcW, PcS or PcWTGN-10, and with aac(6′)-Ib as the first cassette. The tobramycin MIC was 8-fold higher when the cassette was expressed from PcS or PcWTGN-10 than from PcW (data not shown). Our findings indicate that, in class 1 integrons, gene cassette expression is mainly controlled by the strongest Pc variants (PcS, PcH2, PcWTGN-10 and PcW-P2, in 55% of cases).
Another important and previously unnoticed feature of class 1 integrons is the variability of the IntI1 primary sequence linked to the diversity of Pc variants. Among the 10 IntI1 variants identified, three (IntI1R32_H39, IntI1R32_N39 and IntI1P32_H39) accounted for almost 96% of class 1 integrases (Table 3). We found that these three main IntI1s displayed similar integration efficiencies, independently of the attC sites tested, whereas they had extremely different excision activities, depending on the nature of the amino acid at position 32 and/or 39. The R32P and H39N substitutions each drastically reduced the capacity of the integrase to promote recombination between the attCaadA7 and attCVCR sites (by 336- and 51-fold, respectively). In the integrase of the Vibrio cholerae chromosomal integron VchIntIA, the aa found at the position equivalent to residue 32 is basic, while the aa at position equivalent to residue 39 is a histidine (K21 and H28, respectively [24]), showing that, among IntI1 variants, IntI1R32_H39 is its closest relative. The crystal structure of VchIntIA bound to an attC substrate showed that these amino acids are located within an α-helix involved in attC binding [26]. This α-helix is conserved in the predicted structure of IntI1 and presumably plays the same role in recombination [24]. Thus, mutations of aa 32 and 39 in IntI1 might perturb the binding and thus undermine the recombination efficiency of attC×attC. The positively charged aa R32 may also play a role in the interaction with the attC site in the attI×attC recombination reaction. Indeed, a R32P substitution in both IntI1*R32_H39 and IntI1*R32_N39 reduced the integration frequency, but to a lesser extent than in an excision reaction (Table 3 and Figure 3). In contrast, aa H39 does not seem to be involved in the integration reaction. The attI×attC and attC×attC recombination reactions may thus involve different regions of the integrase. Indeed, Demarre and collaborators isolated two IntI1R32_H39 mutants, IntI1P109L and IntI1D161G, that showed much higher integration efficiencies [24].
Interestingly, we found a correlation between Pc strength and integrase excision activity: the weaker the Pc variant, the more active the IntI1. Among the four integrases tested, IntI1R32_H39, which was the most prevalent IntI1 in our dataset (Table 3), had the most efficient excision activity and also displayed higher excision than integration activity. Integrons with this integrase contain either the PcW variant, leading to a weak expression of the gene cassette array, or the PcH1 variant, associated with slightly higher expression (4.5-fold). PcW-containing integrons could compensate for a low level of antibiotic resistance expression by the high excision efficiency of IntI1R32_H39, which confers a marked capacity for cassette rearrangement, in order to place the required gene cassette closer to Pc. In a recent study, Gillings et al suggested that chromosomal class 1 integrons from environmental β-proteobacteria might be ancestors of current clinical class 1 integrons [27]. The integrons they described all encoded IntI1R32_H39 and contained the PcW variant. We suspect that, under antibiotic selective pressure, these “ancestor” integrons may have evolved to enhance gene cassette expression, without modifying the potential for cassette reorganization, either through a single mutation (conversion of PcW to PcH1) or by the creation of a second promoter, P2, that is seven times more active. The high frequency of PcH1 (27.3%) likely reflects its successful selection. P2 probably arises less frequently, as it requires the insertion of three G. We have recently shown that the expression of IntI1 is regulated via the SOS response, a LexA binding site overlapping its promoter [22]. Interestingly, when P2 is created, the insertion of three G disrupts the LexA binding site, probably leading to constitutive expression of IntI1.
In a context of stronger antibiotic selective pressure, the need to express gene cassettes more efficiently could have led to the selection of more efficient Pc sequences (such as PcS and PcWTGN-10) at the expense of IntI1 excision activity, resulting in the stabilization of successful cassette arrays. This hypothesis is consistent with the observation that integrons bearing IntI1R32_N39 or IntI1P32_H39 tend to harbor larger gene cassette arrays than those bearing IntI1R32_H39 (Figure S2).
The tight relationship between the aptitude of class 1 integrons to recombine and to express gene cassettes may be one key to understanding short-term integrase evolution. Different antibiotic selective pressures might select different evolutionary compromises. Thus, integron-driven drug resistance is more complex than previously thought.
Compilation of the class 1 promoter sequences was performed in the entire Genbank nucleotide collection (nr/nt) using the alignment search tool BLASTn (http://www.ncbi.nlm.nih.gov/BLAST) and the sequences of the intI1 and/or attI1 from the In40 integron as reference [28] (GenBank accession number AF034958). This data extraction was performed on 2009-02-01. Three other published but non-deposited sequences [14],[29],[30] were added to the 1351 sequences collected above. Of these 1351 sequences, only 434 contained both the Pc and P2 promoter sequences. Among the latter 434 sequences, we identified the integrons that displayed both identical gene cassette arrays and identical Pc/P2 sequences, independently of their bacterial origin. This analysis led to the isolation of 321 unique class 1 integron sequences that were further studied (Table S1).
The bacterial strains and plasmids are listed in Table 4. Cells were grown at 37°C in brain-heart infusion broth (BHI) or Luria Bertani broth (LB) supplemented when necessary with kanamycin (Km, 25 µg/ml), ampicillin (Amp, 100 µg/ml), tobramycin (Tobra, 10 µg/ml), chloramphenicol (Cm 25 µg/ml), DAP (0.3 mM), glucose (1%), arabinose (0.2%).
Mutations of the Pc and P2 promoter sequences were generated by assembly PCR with overlapping primers that contained the desired mutation and two external primers, int4b and ΔORF11 (Table 5). The two primary PCR products were then used in an equimolar ratio as templates for a second PCR step with the two external primers.
Each transcriptional fusion plasmid was transformed into E. coli strain MC1061 to measure β-galactosidase enzyme activity. Assays were performed with 0.5-ml aliquots of exponential-phase cultures (OD600 = 0.6–0.8) as described by Miller [31] except that the incubation temperature was 37°C. Experiments were done at least 5 times for each strain.
A synthetic array of two cassettes attCaadA7-cat(T4)-attCVCR-aac(6′)-Ib preceded by the lac promoter is carried on plasmid p6851. This construction confers chloramphenicol resistance from the cat gene encoding chloramphenicol acetyltransferase from Tn9, here followed by a phageT4 rho-independent terminator, to prevent transcriptional read-through. The excision assay is based on the capacity of the integrase to catalyze recombination between the attC sites, resulting in the deletion of the synthetic cassette cat(T4)-attCVCR and expression of the tobramycin resistance gene aac(6′)-Ib from the lac promoter [22]. IntI1 proteins were expressed from the pBad-intI1* plasmids. A stationary-phase liquid culture of E. coli strain MG1656, carrying both p6851 and one of the pBad-intI1*, grown over-day in LB broth supplemented with antibiotics and glucose, was diluted 100-fold in LB broth supplemented with antibiotics plus either glucose or arabinose and was grown overnight. Recombinants were selected on LB-Tobra plates. Excision frequency was measured by determining the ratio of TobraR to KmR colonies.
The assay was based on the method described in [6] and since extensively used [23]–[25]. Conjugation is used to deliver the attC site carried onto a suicide vector from the R6K-based pSW family [32] into a recipient cell expressing the IntI1 integrase and carrying the attI site on a pSU38 plasmid derivative (all plasmids are listed in Table 4). Briefly, the RP4(IncPα) conjugation system uses the donor strain β2163 and the recipient MG1656, which does not carry the pir gene, and thus cannot sustain replication of pSW plasmids after conjugation. Recombination between attI and attC sites within the recipient cell leads to the formation of cointegrates between pSW and pSU38 plasmid. The number of recipient cells expressing the pSW marker (CmR) directly reflects the frequency of cointegrate formation. IntI1 proteins were expressed from the pBad-intI1* plasmids. Conjugation experiments were performed as previously described [5]. Integration activity was calculated as the ratio of transconjugants expressing the pSW marker CmR to the total number of recipient KmR clones. attC-attI cointegrate formation was checked by PCR with appropriate primers (primers 35 and 36; Table 5) on two randomly chosen clones per experiment. Background values were established by using recipient strains containing an empty pBad in place of the pBad-intI1*, and were 6×10−7 and 6×10−8 for the attI×attCVCR and attI×attCVCR assays, respectively. At least five experiments were performed for each recombination assay.
|
10.1371/journal.pcbi.1005240 | Impact of Lipid Composition and Receptor Conformation on the Spatio-temporal Organization of μ-Opioid Receptors in a Multi-component Plasma Membrane Model | The lipid composition of cell membranes has increasingly been recognized as playing an important role in the function of various membrane proteins, including G Protein-Coupled Receptors (GPCRs). For instance, experimental and computational evidence has pointed to lipids influencing receptor oligomerization directly, by physically interacting with the receptor, and/or indirectly, by altering the bulk properties of the membrane. While the exact role of oligomerization in the function of class A GPCRs such as the μ-opioid receptor (MOR) is still unclear, insight as to how these receptors oligomerize and the relevance of the lipid environment to this phenomenon is crucial to our understanding of receptor function. To examine the effect of lipids and different MOR conformations on receptor oligomerization we carried out extensive coarse-grained molecular dynamics simulations of crystal structures of inactive and/or activated MOR embedded in an idealized mammalian plasma membrane composed of 63 lipid types asymmetrically distributed across the two leaflets. The results of these simulations point, for the first time, to specific direct and indirect effects of the lipids, as well as the receptor conformation, on the spatio-temporal organization of MOR in the plasma membrane. While sphingomyelin-rich, high-order lipid regions near certain transmembrane (TM) helices of MOR induce an effective long-range attractive force on individual protomers, both long-range lipid order and interface formation are found to be conformation dependent, with a larger number of different interfaces formed by inactive MOR compared to active MOR.
| The μ-opioid receptor (MOR) is an important pharmaceutical target in the treatment of pain. In order to develop novel pain therapies, devoid of the serious side-effects of present opioid analgesics, we need to understand the fundamentals of how MOR works on the molecular level. While some studies suggest that oligomers of MOR could play a role in signaling, how MOR forms dimers, which interfaces form, and the exact role of oligomers in MOR function remain unclear. While research has shown that the membrane environment can affect membrane protein function, most previous computational work to study oligomerization has been performed in a very simple membrane. Here, we use molecular dynamics simulations of MOR in a heterogeneous plasma membrane model (comprising 63 lipid types) to investigate how the presence of the protein modulates its lipid environment, affecting species distribution and sculpting characteristic order and thickness profiles around the receptors. Such modulations, in turn, induce long-range interactions between the proteins and favor the formation of specific dimeric conformations.
| Elucidating the impact of the lipid environment on membrane proteins, including G Protein-Coupled Receptors (GPCRs), is increasingly being recognized as a crucial part of understanding how these proteins function. Cholesterol (CHOL), the lipid for which the most is known about its effect on GPCRs, has been shown to affect receptor thermal stabilization [1,2], agonist affinity [3,4], and oligomerization [5–8]. The conformational equilibrium of the prototypic GPCR rhodopsin is known to be sensitive not only to CHOL levels, but also to phospholipid headgroup and chain saturation [9]. Lipid headgroup charges have also been shown to play a role in the function of the β2 adrenergic (β2AR) [10] and the neurotensin NTS1 receptors [11].
The role of the lipids has primarily been attributed to indirect effects such as changing the physical properties of the membrane (e.g., thickness, curvature, surface tension, and elastic properties). Transmembrane proteins frequently experience hydrophobic mismatch in which the lengths of the hydrophobic chains of the lipids and the hydrophobic part of the protein are different. To rectify this mismatch, the protein adopts several strategies, including conformational changes and remodeling of the membrane thickness by changing the order of the lipids [12]. On a larger scale, heterogeneous membranes are organized into domains which are either liquid-ordered (lo) or liquid-disordered (ld) regions. Notably, some of these domains (e.g., lipid rafts) are enriched in CHOL and sphingolipids, two lipids which have a high propensity of being ordered [13], i.e., parallel to the membrane normal. While the exact role of lipids rafts is debated, they appear to aid in lateral organization of the proteins in the membrane, increasing the propensity of the necessary components of a cell signaling complex to come together [13,14].
In addition to modulating the physical properties of the plasma membrane, lipids can interact directly with membrane proteins. Several crystal structures of GPCRs, including β2AR [2] and the adenosine A2A receptor [15], have been crystallized with interacting CHOL, suggesting that these molecules are strongly bound to the protein. A conserved consensus CHOL binding motif (CCM) has been identified in a number of GPCRs [2], while a sphingolipid binding site has been proposed for the serotonin 5HT1A receptor [16]. CHOL-receptor interactions have also been suggested to play a role in oligomerization based on inferences from crystal structures of GPCRs (e.g., the β2AR [17] and the metabotropic mGlu2 receptor [18]) showing CHOL molecules at putative dimeric interfaces.
Molecular dynamics (MD) simulations of membrane mimetic systems continue to complement experiments by offering a mechanistic understanding of the dynamics not readily available to most experimental techniques. Early coarse-grained (CG) simulations of rhodopsin embedded in bilayers of lipids with different chain lengths and saturation levels showed that the membrane deforms to adapt to the protein [19]. Activation of rhodopsin was also shown to be affected by the physical properties of the membrane [20]; while lipids with unsaturated chains promote activation, CHOL, which is a much more rigid molecule, inhibits activation. Furthermore, while unsaturated lipids were shown by MD to preferentially cluster at the receptor in a non-specific manner [21,22] putative binding sites were identified for CHOL, notwithstanding the typically dynamic interaction between CHOL and GPCRs [23–26]. Notably, MD simulations of β2AR performed with different concentrations of CHOL showed that CHOL binding to conserved sites could prevent some dimer interfaces from forming [6].
While most published MD simulations have been performed in a single or dual component membrane, an average idealized multi-component plasma membrane parameterized recently within the Martini CG force-field [27] offers an unprecedented opportunity to study the impact of lipid composition on the spatio-temporal organization of GPCRs in a more realistic environment. This 63-component membrane mimetic contains combinations of all major lipid headgroups with different fatty acid tails, asymmetrically distributed between two plasma leaflets (see Ref. [27] for more details). Specifically, the following lipid species were included: i) the charged species phosphatidylinositol (PI), phosphatidylserine (PS), and phosphatic acid (PA), phosphatidylinositol mono-, bis-, and tri-phosphate (PIP1-3), and gangliosides (GM); ii) the zwitterionic lipid species phosphatidylcholine (PC), phosphatidylethanolamine (PE), and sphingomyelin (SM), and iii) the minor species diacylglycerol (DAG), ceramide (CER), and lysophosphatidylcholine (LPC). Simulating receptor self-association in this membrane model provides further insight into how lipids can affect oligomerization, adding to what we had previously concluded from simulations of opioid receptors in an environment composed of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and 10% CHOL [28].
The μ-opioid receptor (MOR) is an ideal test case to examine the role of lipids and receptor conformation on oligomerization. First and foremost, the crystal structures of both inactive [29] and activated [30] MOR have been solved with the latter showing the expected outward swing-out movement of TM6. While a crystallographic TM1,2,H8/TM1,2,H8 interface was observed in both active and inactive structures, a TM5,6/TM5,6 interface was identified in the crystal structure of the inactive receptor, but not in that of its activated form. Furthermore, although the exact role of MOR oligomerization in signaling is still under debate, the potential of developing new, more effective therapies in the treatment of pain [31] by selectively targeting MOR heterodimers makes these systems worthy of further investigation. Finally, while there is evidence both for [32] and against [33] the association of MOR with lipid rafts, CHOL has been shown to be important in the spatio-temporal signaling by MOR [34]. In fact, it has been shown to promote homodimerization of MOR [5], agonist binding [35], coupling with G-proteins [5,35], and translocation of β-arrestin [36].
Here, we present the results of CG MD simulations of arrays of 16 inactive and/or activated MORs in the aforementioned multi-component plasma membrane carried out to further evaluate the role of both lipid environment and protein conformation on MOR oligomerization.
The 63 lipid types in the plasma membrane model were grouped by their headgroup to examine the preferred regions of association between the lipids and the receptors during simulation. Specifically, for this analysis, we used the final 2 μs of membrane equilibration of four sets of simulations: 16 inactive or active receptors in either a 50×50 nm2 or 25×25 nm2 membrane (referred to as low and high receptor density membranes, respectively). To analyze the behavior of the lipids, the GL1 or AM1 bead was used for all non-sterol lipids while the ROH bead was used for CHOL (See S1 Fig for a depiction of representative lipids). The normalized 2D probability distributions of lipids with zwitterionic headgroups (i.e., PC, PE, SM) in the simulations of the plasma membrane model with low density of inactive or active receptors are shown in Fig 1 whereas the distributions of all charged lipid types (i.e., GM, PS, PI, PIP1-3, and PA) are shown in S2 Fig. Unlike the distributions of PA, PI and PIP1-3, the distributions of PC, PE, SM, PS, or GM appear to be qualitatively similar around inactive and activated receptors. To confirm these observations, we calculated the standard deviation of the distribution of each lipid type at each grid point for five runs of inactive or active receptors, and compared it to the difference between the average lipid distributions around inactive and active receptors.
The distributions of PC and PE, which are the most populated glycerophospholipids in the plasma membrane, show a depletion of lipids immediately next to certain receptor TM helices. For instance, comparing Fig 1A and Fig 2, we observe that the lipid depletion at TM5 and TM6 (yellow and green dots in Fig 1A) is due to the presence of CHOL immediately next to the protein.
The remaining glycerophospholipids, PA, PI, PS, and PIP1-3, make up approximately 19% of the lower leaflet and are not present in the upper leaflet (See S3 Fig for the composition of the membrane). Despite their random initial position, these lipids diffuse towards the proteins during equilibration. Since there are a number of positively charged Lys (specifically, residue number 98, 100, 1854.43, 174, 2605.66, 2696.24, 2716.26, and 344) and Arg (i.e., 951.59, 1653.50, 179, 1824.40, 2585.64, 263, 2736.28, 2766.31, 2776.32, 2806.35, 345, 348) residues on the intracellular side of MOR, it is not surprising to see PA, PI, PS, and PIP1-3, which are all negatively charged, localize in this region of the proteins (S2A–S2D Fig for PA, PI, PIP1-3 and PS, respectively). Specifically, PIP1-3 molecules are found to be predominantly in contact with the intracellular ends of TM4, TM6, and TM7 of inactive MOR. Upon activation, the outward swing of TM6 away from TM3 leaves the positively charged residues on TM5 and TM6 more exposed to negatively charged lipids, so that their concentration is reduced in the space between TM6 and TM7 and increased near TM5 and H8. See S1 Table for a list of the residues most frequently in contact with PIP1-3. Despite the small concentration of these PIP1-3 lipids in the membrane, the maximum difference between their distributions around inactive and active receptors (0.03) is much larger than the maximum standard deviation (0.005) of the PIP1-3 distribution at each grid point for the five runs of inactive or active receptors, suggesting that the observed conformation-dependent changes are larger than the sampling error. On the other hand, the PA and PI lipids (S2 Fig Panels A and B, respectively) do not show significant changes in distribution around the two receptor conformations, because the differences in lipid distribution around inactive and active receptors (PA: 0.003, PI: 0.006) are the same order of magnitude as the maximum standard deviation at each grid point (PA: 0.002, PI: 0.002).
In addition to the glycerophospholipids, the plasma membrane contains a large fraction of the sphingolipid SM. These two groups of lipids are distinguished by the linker beads connecting their headgroup to the lipid tails. While each of the two linker beads of the glycerophospholipids represents a nonpolar ester group, in the case of SM, one linker bead represents a hydroxyl group and the other an amide group, both of which are polar. Similar to the PC distribution (Fig 1A), SM (Fig 1C) is depleted immediately next to the protein except at TM1, TM5, and TM6, where the sidechains of several polar and charged residues form strong Lennard-Jones interactions with the polar linker beads of SM. Finally, the GM lipids (S2E Fig) are sphingolipids which have an oligosaccharide and sialic acid head group and comprise ~5% of the upper leaflet. During the initial portion of the membrane equilibration the GMs diffuse to the proteins where they remain, very rarely moving back into the bulk membrane. Determining the precise protein-GM interaction, which dictates their segregation propensity, is difficult due to the large headgroup of these lipid species, but it does not appear to be different depending on the receptor conformation. The GMs tend to cluster together in line with the observations of Ingólfsson et al. [27] and Gu et al. [37].
Composing about 30% of the idealized plasma membrane (S3 Fig), CHOL is one of its largest components. Fig 2 shows the distribution of CHOL around the simulated inactive and active MOR protomers. In agreement with experimental evidence [38], CHOL flip-flops between the upper and lower leaflets and was frequently found to be interacting with the helical bundle region of the receptors, as shown by the structures (Fig 2) colored white to blue to green by the probability of each residue being in contact with the ROH bead of CHOL. For inactive MOR, the highest density of CHOL was found near TM6 and TM7 in the hydrophobic region of the membrane. Notably, an identified CHOL hotspot near TM6 corresponds to the location of the electron density that was attributed to a CHOL molecule in the MOR inactive crystal structure (PDB ID: 4DKL). While a hotspot near the palmitoylation site C1703.55 was also observed in the simulations of the inactive MOR, in those of the active MOR, the CHOL was preferentially found between TM2 and TM5 and between TM5 and TM6. Notably, the maximum difference between the CHOL distributions around inactive and active receptors was 0.003 while the maximum standard deviation of CHOL distribution at each grid point of the individual simulation runs was 0.0006, indicating that the differences are substantial.
We quantified the local properties of the lipids in the plasma membrane model by calculating the order parameter of each non-flipping lipid species (i.e. excluding CHOL, CER, and DAG) during the final 2 μs of membrane equilibration in which the receptors are kept fixed or the production runs in which the receptors are permitted to freely move in both the high and low receptor density simulations. The lipid order parameter was defined as the angle between the membrane normal and a vector between the first and last bead of the tail. Specifically, as described in the Methods section, a value of the order parameter of 0 means that lipid tails are on average ordered and parallel to the membrane normal, while larger values imply that the lipid tails assume disordered conformations. The extreme value of π/2 means the lipid tail forms a 90° angle with the membrane normal. In Fig 3 and S4 Fig we report, as an example, the results of one of the simulation runs where the active or inactive receptors are fixed in the low or high receptor density systems, respectively. The other trajectories show a similar behavior, as do the trajectories of the mixed-arrays (i.e., 50% inactive/50% active MOR; S5 Fig), which were only simulated in the high receptor density membrane. Since the pattern of ordered and disordered regions near the inactive or active receptors in the mixed-arrays is similar to that in the arrays of only inactive or active protomers, respectively, only the results for all inactive or all active arrays are discussed. There are clear regions of order and disorder in the membrane. As expected, lipid order is correlated with bilayer thickness (see Fig 3, S4 Fig, S5 Fig and S6 Fig), since a lipid with a fully extended tail has a greater z-projection of the head to tail distance than that of a disordered lipid.
Fig 3 suggested that the regions close to TM5, TM6, and–to a lesser extent–TM1, show more order while the regions near TM4 show more disorder. To confirm this behavior, we calculated the average membrane thickness and lipid order near the individual helices (shown in top and lower panels, respectively, of Fig 4 for the low receptor density simulations in which the receptor was permitted to move freely). Interestingly, the extent of the order is not only helix dependent, but also conformation dependent (Fig 4). The region near TM5 and TM6 (yellow and green dots in Fig 3, yellow and green lines in Fig 4) is thicker (and more ordered) in the simulations with the inactive receptor than those with active receptors, while the region near TM4 (orange in Figs 3 and 4) is more disordered next to both the active and inactive protomers. The trends in the helix-dependent lipid order are not influenced by the overall protein density nor by the restraints imposed during the equilibration. In fact, similar results were obtained for the low receptor density simulations in which the proteins were permitted to move freely (Fig 4) as well as the low and high receptor density simulations of restrained receptors (S7 Fig and S8 Fig, respectively) despite the inability of the receptors to tilt when they were kept fixed (S9 Fig). The correlation between the locations of ordered regions with SM enrichment was confirmed by calculating a 55% to 45% ratio for the probability to find a SM molecule in a region with order parameter larger or smaller than the average in the high receptor density simulations. For all other lipid species, including CHOL, the two probabilities are equivalent. Notably, the presence of inactive receptors shifts the overall distribution of the order parameter towards smaller values (i.e., higher order) with respect to the active protomers (S10 Fig).
Snapshots of the production runs are shown in S11 Fig for representative high receptor density simulations of inactive and active MOR (upper and lower panels, respectively), and the first 10 μs of a representative inactive trajectory is shown in S1 Movie. All high receptor density simulations show a similar behavior. The lipids closest to the proteins are those that show the broadest distribution of angles. In contrast, in the regions far from the protein, the distribution of order values is much narrower. Thus, the introduction of proteins into a membrane promotes the formation of lipid regions that are either more ordered or more disordered than in the membrane simulated alone. When two receptors get closer together, the lipid regions between them become disordered for the interface to form.
We then investigated the interplay between the relative lateral position of protomers and the profile of membrane properties, by calculating, for selected interfaces (Fig 5 and S12 Fig), the average order and thickness of the lipids in the region separating two receptors as a function of the distance from the protein center d and of the relative protein-protein distance r. Not surprisingly, in agreement with the distinct effect of different helices reported above (Fig 4), the membrane profile is also strikingly dependent on the relative orientation of the proteins.
We show this for the two protein regions that are maximally involved in the observed dimers, i.e. TM1,2,H8 and TM5 (see below). Despite both these protein regions favoring the formation of ordered lipid domains, the dependence of such effect on distance is, interestingly, different. In the TM5/TM5 case (Fig 5A and S12A Fig, for the inactive and active receptors, respectively), for well-separated protomers (r≫r0, where r0 is protomer average radius), the membrane is perturbed up to d≈2.5 nm from the protein center, confirming that isolated protomers are accompanied in this case by a stable region of ordered lipids and a thicker membrane. A second regime is established when the protomers are within r≈7 nm of each other and cooperative effects of the nearby protomers start to appear. Thus, each protomer appears to start experiencing the influence of the other one at a distance of ~7 nm, which is larger than the sum of the range of the perturbations for the isolated protomers (~2.5×2 ≈ 5 nm). In contrast, for the TM1,2,H8/ TM1,2,H8 interface, the modulation of the membrane properties when protomers are far from each other (r≫r0) is very weak on average (Fig 5B and S12B Fig, for the inactive and active receptors, respectively), and the lipid ordering arises almost exclusively from cooperative effects when two protomers are closer than r≈6 nm. The results obtained for asymmetric interfaces, e.g., the TM1,2,H8/TM5 interface (S12 Fig Panels C and D for the simulated systems with inactive or active receptors, respectively) are similar to those obtained for symmetric interfaces, although the recorded cooperative effect is weaker for r<6 nm.
Since CHOL is a key player in the stabilization of ordered phases in plasma membranes [13] and it flips between leaflets, we investigated its ordering behavior separately. While CHOL does not have a tail, it is possible to calculate an order parameter (see Methods section for details) which reveals the orientation of the molecule with respect to the membrane normal (Fig 6). Since CHOL was seen to flip-flop between the leaflets in all simulations, its order was calculated separately for the headgroup regions and the hydrophobic core of the bilayer. Most of the CHOL molecules in the headgroup region were found to be in ordered regions (orange in Fig 6), i.e. they were parallel to the membrane normal, but there were regions close to the protein in which CHOL was tilted (purple in Fig 6), mostly between TM4 and TM5 (orange and yellow dots, respectively). The regions in which the CHOL molecules were tilted roughly correspond to the regions in which the remaining lipids were also disordered (Figs 3 and 4). On the other hand, CHOL molecules away from the protein were, on average, always ordered (i.e., close to parallel to the membrane normal), even when the other lipid regions were disordered. In contrast, CHOL molecules in the hydrophobic part of the membrane exhibited a much broader range of order values (see S13 Fig), which are skewed toward 90°.
Niemelä et al. used CG simulations to show that the mobility of lipids is reduced when they are in the vicinity of proteins [39]. To assess the mobility of lipids near MOR in the high receptor density simulations with freely moving receptors, local residence times were calculated for the lipids with the largest mole fraction (i.e., CHOL and the PC lipids; see S3 Fig) in the proximity of each helix (S14 Fig). Consistent with their homogeneous distribution around the proteins, the residence time of the PC lipids is similar for all the helices (~3.5 ns on average) of both inactive (blue points in S14 Fig) and active (red points in S14 Fig) MOR.
In contrast, the residence time of CHOL differs depending on the helix to which the molecule is in proximity. The CHOL molecules nearest to TM6 stay close to these helices for longer (~5–6 ns for inactive and active MOR, respectively) than the CHOL molecules near the other helices (~4 ns, on average). The CHOL residence time at TM6 or TM7 is longer for the inactive receptor than the active receptor, possibly due to the outward swing of TM6 upon activation.
The only lipids that moved transversely through the bilayer in our simulations, i.e. flip-flopped between the leaflets, were CHOL, CER, and DAG, which are the same lipids that were seen to flip-flop in the published simulations of the plasma membrane without proteins [27]. The rate of CHOL flipping in our low receptor density membrane (inactive: 7.23±0.07×106 s-1, active: 7.29±0.03×106 s-1) was comparable to the rate of 6.53±0.01×106 s-1 in the plasma membrane without proteins [27] as was the rate of DAG flipping (inactive: 6.34±0.30×106 s-1, active: 6.62±0.21×106 s-1, plasma membrane: 5.87±0.05×106 s-1). The flipping rates of CHOL (inactive: 3.97±0.04×106 s-1, active: 4.03±0.04×106 s-1) and DAG (inactive: 3.38±0.39×106 s-1, active: 3.37±0.46×106 s-1) in the high receptor density membrane were much slower than in the low receptor density membrane and plasma membrane without proteins. Consistent with the very low rate of flipping seen by Ingólfsson et al. [27], the CER switched leaflets very infrequently, on the time scale of our simulations.
To determine the effect of the protein on the equilibrium distribution of these lipids, we used the final 2 μs of the membrane equilibration of the high receptor density simulations to calculate the density of molecules as a function of the z-coordinate of CHOL’s ROH or CER/DAG’s linker beads (AM1 or GL1) and either (a) the minimum distance to the backbone (BB) beads of inactive or active MOR (S15 Fig) or (b) the lipid order in the plasma membrane with embedded inactive MOR, embedded active MOR, or no receptors (Fig 7 for CHOL and S16 Fig for DAG and CER). The majority of CHOL molecules were found close to parallel to the membrane normal in both the upper and lower leaflets of the plasma membrane with or without receptors. As also seen in S13 Fig, Fig 7 shows CHOL molecules in the hydrophobic region of the membrane that are perpendicular to the membrane normal. Notably, there are more of these perpendicular CHOL molecules in simulations of the plasma membrane with receptors than without them. To confirm that the CHOL distributions are substantially different near the inactive or active receptors, we repeated the analysis for each individual run and found a similar CHOL density in the hydrophobic part of the membrane for all of them. While the concentration of CHOL in the upper and lower leaflets is asymmetric in the simulations of both the inactive receptors and the plasma membrane without receptors, the distribution of these molecules is symmetric in the case of the active receptors. Lastly, an additional distribution of CHOL at an angle between π/3 and π/4 with respect to the membrane normal is seen only in the simulations of the active receptors.
The calculated z-coordinate of CHOL’s ROH beads as a function of their minimum distance from the protein’s BB beads (S15 Fig) shows minima in the middle of the bilayer (-0.8 nm < z < 0.8 nm) that are immediately next to the protein (minimum distance ~0.5 nm), suggesting that the protein binds CHOL at these sites, consistent with the regions of high cholesterol density seen in Fig 2. A kinetic model was built to determine whether these minima in the hydrophobic region of the bilayer next to the protein are involved in the flipping mechanism of CHOL. The five most frequently occurring pathways of CHOL movement in the z-direction are shown in S17 Fig. The two largest states (6 and 7 in S17 Fig) in the models for both the active and inactive simulations correspond to CHOL in the upper or lower leaflets, respectively, which have no contacts with the protein. For both the inactive and active protomers, the largest fluxes were between states 6 and 7 indicating that the main route of CHOL flipping is through the membrane away from the proteins. The most probable pathway for a CHOL molecule from the upper to lower leaflet through a bound state is via state 1 for the inactive protomer and state 4 for the active protomer, although the flux is much lower through this pathway than through the membrane.
The distributions of CER and DAG lipids were also examined, although these lipids each make up less than 1% of the membrane. The location of the deepest minimum in the plots of the order of DAG lipids as a function of z are different in the simulations of the plasma membrane with or without receptors (S16 Fig). For the inactive receptors, the deepest minimum of DAG lipids is in the middle of the bilayer, while it is in the lower or upper leaflet for the active receptors or membrane without proteins, respectively. In contrast, the distributions of the CER lipids are similar among themselves, except for a shallow minimum in the middle of the bilayer in the case of the simulations of the inactive receptors.
Removing the position restraints on the receptor BB beads allowed the receptors to move freely in the membrane and to eventually self-assemble within the initial microseconds of the high receptor density simulations. While a quantitative assessment of the dimer formation kinetics cannot be obtained with our data, the time decay of the number of monomers in the membrane (S18 Fig) shows that the inactive system forms aggregates slightly more quickly than the active. However, once formed, the interfaces did not dissociate over the 20 μs of simulation time.
To characterize the structural features of the formed complexes, the interfaces formed by the final microsecond of simulation time were clustered by their contact maps. While the simulations do not provide enough statistics to definitively quantify the stability of each interface, their formation or absence is telling. The fraction of interfaces formed is listed in S2 Table and depicted in Fig 8A, 8B and 8C for dimers formed between two inactive receptors, two active receptors, or one active and one inactive receptor, respectively.
There is only one interface that was formed by inactive/inactive, active/active, or inactive/active receptors, i.e. irrespectively of the conformation of the participating protomers: TM1,2,H8/TM5. In this interface, TM2 only forms extracellular contacts with TM5, while H8 and TM5 form intracellular contacts.
For the inactive proteins, nine different interfaces were formed, but over 50% of them involved the TM1,2,H8 region of one protomer and the TM4/TM5 region of the other. In particular, the TM1,2,H8/TM4,5 interface constituted the ~15% of the observed interfaces while the similar interfaces TM1,2,H8/TM4 and TM1,2,H8/TM5 made up ~31% and 8%, respectively, of all observed interfaces. In the TM1,2,H8/TM4 and TM1,2,H8/TM5 interfaces, the second protomer is slightly rotated either clock- or counter-clockwise relative to that of the TM1,2,H8/TM4,5 interface, such that no contacts are formed with TM5 and TM4, respectively. In all of these interfaces, TM2 forms extracellular contacts with TM4,5, while the majority of the contacts are formed by TM1. The inactive receptors formed two interfaces involving TM7, which were not observed in the active or mixed proteins. In the TM4/TM7 (~7%) interface, TM7 is in contact with TM4 on the intracellular side, while TM7 forms extracellular contacts with TM1 in the TM1,H8/TM6,7 interface (~8%). The TM1,2,H8/TM1,2,H8 (~8%), one of the crystallographic interfaces, TM4/TM5 (~8%) and TM5/TM5 (~8%) interfaces were also observed in the simulations of inactive MOR. While the TM5/TM5 interface met our criteria for interface formation (see Methods section), it is not as compact as the other interfaces with some lipid tails located in between the protomers.
The most frequently formed interfaces in the simulation of all active MORs are the TM1,2,H8/TM5,6, TM1,2,H8/TM1,2,H8, and TM1,2,H8/TM5 interfaces which are all seen with approximately the same frequency (~25%). In the TM1,2,H8/TM5,6 interface, TM6 forms contacts on the extracellular side only, which allows the interface to form without preventing the outward swing of TM6 in the active structure. The contacts formed by TM1 and TM2 in the TM1,2,H8/TM1,2,H8 interface are on the extracellular side, while H8 forms contacts on the intracellular side. Another frequently seen interface between active receptors is the TM4/TM4 interface (~17%) in which contacts are formed only by TM4. Both the TM4/TM4 and TM1,2,H8/TM5,6 interfaces were unique to the active protomers.
In the case of the dimers in which one protomer was in the inactive conformation and one was in the active conformation, only two interfaces were formed. However, what appeared as a new interface, TM2,H8/TM4, was structurally very similar to the TM1,2,8/TM4 interface formed by inactive receptors except that one protomer was slightly rotated, lengthening the distances between TM1 and TM4 beyond our threshold to form a contact. The TM1,2,H8/TM5 asymmetric interface is the most favored (~60%) and is formed by either an inactive/active or active/inactive combination. The symmetric TM1,2,H8/TM1,2,H8 interface was not formed by the combination of one active and one inactive protomer, which is surprising since it was formed by either two inactive protomers or two active protomers. In the crystal structures of the two conformations, the intracellular ends of TM7 do not overlap, shifting H8 slightly. Thus, formation of the TM1,2,H8/TM1,2,H8 interface in this mixed system may require some adjustments in the position of H8, which the model is not able to capture due to the elastic network required to maintain the receptor tertiary structure.
The highest-order oligomer seen in the simulations was a trimer, but only few of them were recorded. For instance, only three trimers were seen in the simulations of inactive receptors with the following interfaces: 1) TM7A/TM4B and TM4A/TM1,2,H8C, 2) TM5A /TM5B and TM4A /TM5C, and 3) TM4A/TM1,H8B and TM1,2,H8A/TM4C, where subscripts A, B, and C identify the three protomers participating in the trimer. Only one trimer was seen in simulations of all active receptors, and it consisted of a TM1,2,H8A/TM1,2,H8B interface and a TM5,6A/TM1,2,H8C interface. In the mixed-array simulations, one trimer formed as the result of the association of two active receptors (B and C) with an inactive receptor (A) in the middle. In this configuration, the inactive protomer contributed TM5 to a TM5A/TM1,8B interface and TM2,H8 to a TM2,H8A/TM4C interface.
The distribution of the lipids around frequently occurring interfaces in the inactive (e.g., TM1,2,H8/TM4 and TM1,2,H8/TM4,5) and active (e.g.,TM1,2,H8/TM5,6, TM1,2,H8/TM1,2,H8, or TM1,2,H8/TM5) dimers was calculated to determine possible correlations between the location of lipids near the protomers and the formation of specific dimeric interfaces. The concentration of the PA, PS, PI, and PIP1-3 lipids in the lower leaflet is too low to draw robust conclusions on any role these lipids play at the interface. While the distributions of the PC, PE, and SM lipids revealed no specific hotspots at dimer interfaces, CHOL molecules were always present at the dimer interfaces for the five interfaces listed above. The model structures in S19 Fig, which are colored according to the probability of a CHOL molecule being in contact with the helices involved in a dimer interface, provide an example of these interactions at the TM1,2,H8/TM4 and TM1,2,H8/TM1,2,H8 interfaces formed between inactive or active receptors, respectively and the residues frequently in contact with the bundle are listed in S3 Table.
All of the residues with which CHOL forms contacts for more than 50% of the simulation are on the helical bundle. As seen in S3 Table, four out of five interfaces involving TM1 exhibit contacts of CHOL with A731.37, S761.40, I771.41, and T1202.56. All five interfaces exhibit CHOL contacts with L1162.52 and S1192.55. While S1192.55 has one of the highest probabilities of being in contact with CHOL when MOR is monomeric, I2425.48 is frequently in contact with CHOL in both protomeric and dimeric configurations. The three interfaces involving TM5 have CHOL most frequently bound at I2385.44, F2415.47, I2425.48, M2435.49, and V2455.51.
Studies from several groups [6,19,40–43], including ourselves [28,44], have computationally explored the effects of the lipid bilayer on the formation of GPCR complexes. In this work, we further investigate the impact of the lipid environment on the spatio-temporal organization of inactive and/or active MOR employing, for the first time, a more realistic 63-component plasma membrane [27] model.
The effects of lipids on protein association in the membrane are classified as either specific (i.e., direct) or non-specific effects [45]. Specific effects refer to possible interactions of individual lipid molecules with specific residues on the protein surface, whereas non-specific effects are attributed to the modulation of the properties of the membrane (e.g. bilayer thickness). Our analysis of MOR simulations shows an interesting interplay between these two types of effects with individual helices of the receptor promoting regions of the membrane with different average order. Specifically, the enrichment of SM at helices TM1, 5 and 6 promotes regions of ordered lipid molecules next to these helices, while helix TM4 is most frequently adjacent to less ordered regions of the lipid.
Recently, Katira et al. [46] suggested a mechanism by which the stabilization of phases with specific order by proteins in a homogeneous membrane can lead to membrane-mediated interactions between proteins. In their model, a disordered phase of a simple DPPC membrane was stabilized around an idealized protein by setting the length of the protein hydrophobic core to be less than the thickness of the surrounding ordered membrane. When two proteins, each surrounded by their own disorder lipid phase, approach each other, it is energetically favorable for the two regions of disordered lipids to merge to minimize the size of the order/disorder interface, resulting in an effective, long-range, induced interaction between the proteins. A similar mechanism appears to be occurring in the MOR simulations reported here. However, in contrast to the homogeneous idealized proteins studied by Katira et al., MOR in the heterogeneous membrane appears to induce order/disorder depending on the helix and the lipids closest to it. In the complex plasma membrane in our simulations, the hydrophobic length of the individual helices as well as the modulation of the local bilayer compositions by the protein produce specific hydrophobic mismatches that lead to helix-dependent regions of order and disorder. Despite these important differences, the simulations reported here show how ordered lipid regions will tend to coalesce to reduce the energetic cost of merging ordered and disordered regions in the membrane.
Although the high-protein concentration in our high receptor density simulations, and the limited length of the low receptor density simulations prevent us from addressing the effect of this mechanism on the translational dynamics of the proteins since the average protein-protein distance is relatively small in our system, it is clear that the lipid order influences the orientation of protomers that eventually come together to dimerize. Specifically, interfaces formed by helices next to order-inducing regions (e.g. TM1,2,H8/TM5) are more frequently formed than those next to disorder-inducing regions (e.g. TM4/TM4) or those next to regions with opposite order preference. At a shorter range, the proteins will then proceed to form bona fide dimeric structures depending on specific residues at the interface, shape complementarity, and physico-chemical properties of the interface. The presence of ordered lipid regions next to helices TM1, 5, and 6 fades when the distance between protomers is only a few nm. On the nanometer length scale it is reasonable to assume that shape complementarity and specific interactions start playing a role in dictating the shape of the interaction free-energy and ultimately determining whether an interface can form or not. Thus, whether the helices involved in interface formation induce ordered or disordered lipid regions ceases to play a role once the protomers are close together.
Our simulations show an area of high CHOL density near the palmitoylated C1703.55 located between TM4 and TM5 of MOR, which had been previously suggested to trap CHOL molecules and to promote dimerization [5]. This palmitoylation site appears to help order the lipids around TM5 in simulations of the inactive but not the active MOR which may explain why the TM5/TM5 interface is formed between inactive but not active receptors. While increasing the lipid order and the membrane thickness near TM4 enhances the propensity of inactive MOR to form a dimerization interface involving this helix, the outward movement of TM6 upon activation appears to affect the ability of this helix to be involved in dimerization of active MOR by virtue of a decreased lipid order and membrane thickness near that helix.
The results presented here offer testable hypotheses for experimental investigation. For instance, regions of lipid order immediately next to the MORs are found to be rich in SM, which is known to contribute to ordered regions of a membrane [47]. We find that the polar linker beads of SM lipids associate with polar/charged side chains on TM 5 and 7 of MOR (e.g. Y2275.33, N2305.36, F3137.30, Q3147.31). Mutating these residues to Ala might decrease lipid order and interface formation since the attraction between SM and the protein is predicted to decrease.
Lipid rafts have long been thought to compartmentalize cellular processes by contributing to the assembly of signaling molecules [48]. There is support for localization of activated MOR in lipid rafts [32,35], but the involvement of lipid rafts in GPCR signaling is likely dependent on the signaling pathway and the cell type [49]. The plasma membrane used in the simulations reported herein is too small to eventually see formation of lipid rafts, but we can speculate that the ordered/disordered lipid regions mimic the role of lipid rafts in guiding the assembly of receptors, and may also contribute to orienting the receptors so as to guide their interaction with specific regions of the intracellular proteins, e.g., the G-proteins, that are in contact with the membrane.
Transmembrane lipid translocation, or flip-flop, is essential to maintain the required composition of cell membranes. In the simulation reported here, only three lipid types were observed to flip-flop from one leaflet to the other, specifically CHOL, CER, and DAG, which are the same three lipids found to flip-flop in published simulations of the plasma membrane without proteins [27]. Although GPCRs have recently been shown to act as phospholipid scramblases [50], helping them move from one leaflet to the other, phospholipids did not flip within the 20 μs of our simulations. This is likely due to the timescale of our simulations. Notably, published umbrella sampling simulations of different sets of lipid bilayers showed that the energy barrier for a dipalmitoylphosphatidylcholine (DPPC) lipid to flip-flop though a DPPC bilayer is increased by 26 kJ/mol for a 20% CHOL membrane relative to a pure DPPC bilayer [51]. The flipping rate of DPPC was estimated to be on the order of hours in a pure membrane and on the order of years in a DPPC/CHOL membrane.
The distribution of CHOL in the hydrophobic region of the bilayer (S13 Fig, solid lines) show that the CHOL molecules are either perpendicular to the membrane normal or close to perpendicular, which is consistent with neutron scattering experiments showing the presence of disordered CHOL in a PC membrane with polyunsaturated tails [52]. Previous simulations have shown that CHOL flips in a membrane by first tilting, and then migrating to the middle of the bilayer where it is perpendicular with respect to the membrane normal [53–55]. While we also see CHOL in the middle of the membrane where it is bound to the protein, our kinetic model shows that the main route for CHOL flipping is through the membrane away from the proteins. The mechanism by which transmembrane proteins promote lipid flip-flopping has been attributed to either thinning of the membrane near the protein or the formation of hydrophilic interactions between lipid and protein, both of which would reduce the energy barrier to flipping. Here we see that CHOL flipping via states that are bound to the protein reduces the overall translocation rate, suggesting that the same would be true for other lipids with hydrophilic heads.
The dimer structures also showed CHOL bound to the helical bundle. The helices to which CHOL binds when MOR is isolated are not the same helices to which it binds when the receptor is in a dimeric configuration. However, there is no correlation between CHOL bound at specific regions of the helical bundle and dimerization. It is unclear if the CHOL molecules bind an individual protomer before the interface forms or if they insert themselves between protomers after dimerization. The limited statistics of the simulations presented here does not allow us to derive quantitative inferences about the behavior of individual CHOL molecules at the dimer. A new set of simulations is currently being performed to answer this question.
The non-sterol lipids which make up the bulk of the membrane (PC, PE, SM; 60% in the upper leaflet and 53% in the lower leaflet) show a ring-like distribution around the individual receptor protomers. While some of these zwitterionic lipids are localized around the helices, these preferences are mostly independent of the protein conformation. In contrast, the negatively charged PIP lipids which are present only in the lower leaflet have significant differences in their distribution around the inactive and active conformations supporting experimental inferences that they favor the latter conformation. The PA, PS, and PI lipids all have a charge of -1, but the PIP1-3 lipids are derivatives of the PI lipids which have been phosphorylated once (PIP1), twice (PIP2), or three times (PIP3) and thus have charges of -3, -5, and -7 respectively. These lipid distributions are consistent with experimental and computational evidence showing that negatively charged residues promote activation not only in GPCRs [10,11,56] but also in ion channels and transporters [57]. For instance, in the case of β2AR, the presence of the negatively charged lipid phosphatidylglycerol (PG) promoted agonist binding and activation while the neutral PE lipid promoted antagonist binding and the inactive conformation [10]. Recent all-atom MD simulations of β2AR showed that a PG lipid can insert itself inside the receptor between TM6 and TM7, forming a salt bridge with R3.50 and stabilizing active conformations [56]. While the rigidity of the elastic network applied to the coarse-grained receptor structure in our simulations prevents lipids from entering the MOR, the PIP1-3 and PA lipids prefer to localize at the crevice between TM6 and TM7 formed upon activation of the receptor as a result of TM6 swinging outward. The ability of negatively charged lipids to promote activation has also been shown for another GPCR, the neurotensin receptor NTS1, as coupling between NTS1 and Gq-proteins increased with PG content [11]. The results of our simulations are consistent with a conformation-dependent role of negatively charged lipids in membrane protein function, but due to the low concentration of these lipids in the idealized plasma membrane, we cannot derive a clear conclusion about their role, if any, in oligomerization.
Early cysteine cross-linking experiments have suggested that inactive and active GPCRs do form different interfaces [58–60]. The present simulation results show that the interfaces formed by the MOR receptors are also dependent on the conformation of the protein. Specifically, while both the inactive and active receptors formed the TM1,2,H8/TM1,2,H8, TM1,2,H8/TM4, and the TM1,2,H8/TM5 interfaces, each conformation also formed mutually exclusive interfaces.
Both the inactive and active crystal structures of MOR show dimeric packing interactions involving TM1, TM2, and H8. The Cα RMSD of TM1 and TM2 of the simulated inactive/inactive dimer was 4.2 Å (referenced to the MOR inactive crystal structure 4DKL) or 2.8 Å (referenced to the MOR active crystal structure 5C1M). In the case of the active/active simulated dimer, the RMSD was 4.8 Å (referenced to 4DKL) or 2.9 Å (referenced to 5C1M). Unlike the crystal structure of active MOR, the crystal structure of inactive MOR also showed a symmetric TM5,6/TM5,6 packing interaction [29]. The fact that this interface is neither seen here nor in previous simulations we carried out on the inactive MOR in a POPC/10% CHOL environment [28] further supports the suggestion that the TM5,6/TM5,6 interface is either kinetically impaired or it may represent a crystallographic artifact.
Similar to the results of our previous simulations of inactive MOR in a POPC/CHOL membrane [28], the simulations carried out here suggest that interfaces TM1,2,H8/TM1,2,H8, TM1,2,H8/TM4,5, and TM5/TM5 are likely to form in a mimetic membrane environment. In contrast, the TM1,2,H8/TM5,6 and TM4,5/TM5,6 interfaces seen in the POPC/CHOL membrane did not form between inactive protomers in the 63-component plasma membrane, but we believe that this is most likely due to differences in the modeled intracellular loop 3 (IL3).
Several other GPCR crystal structures show packing interactions that are similar to those seen in our simulations. Both the inactive δ-OR (DOR, PDB ID: 4DJH) [61] and inactive β1 adrenergic receptor (β1AR, PDB ID: 4GPO) [62] crystal structures form TM1,2,H8/TM1,2,H8 interfaces. The Cα RMSD of the inactive/inactive simulated dimer was 3.8 Å relative to DOR and 2.8 Å relative to β1AR. Two of the asymmetric interfaces resulting from the simulations of inactive MOR were seen in the crystal packing of chemokine receptors: TM1,H8/TM6,7 in CXCR4 (PDB ID: 3OE8) [63] and TM4/TM7 in CCR5 (PDB ID: 4MBS) [64], with RMSD of 4.5 Å and 4.4 Å, respectively. Interestingly, the chemokine receptors have been suggested to form heterodimers with MOR [65]. Another interesting observation is that no higher-order oligomers other than trimers are identified in our simulations, but this may be due to our choice of strict interface criteria, which limits the higher-order complexation to trimers in the afforded timescale.
In summary, we have performed over 300 μs of CG MD simulations of inactive and/or active MOR in an idealized plasma membrane and concluded that the impact MOR conformation has on dimerization is two-fold. First, the two conformations induce different patterns of order and disorder with merging ordered regions determining protomer orientation with respect to one another. Second, the shape complementarity between different conformations affects both the number and type of interfaces formed. While indirect lipid effects are found to play a major role in receptor dimerization, direct effects through specific lipid-receptor interactions require further investigation.
The inactive and active crystal structures of the mouse MOR (PDB ID: 4DKL [29] and 5C1M [30], respectively) were used as the starting structures after removal of all non-receptor atoms (e.g. the ligands, as well as the fused T4L lysozyme and the nanobody in the inactive and the active crystal structures, respectively). The missing intracellular loop 3 (IL3) of the inactive structure was added using the high-resolution structure of the δ-OR (PDB ID: 4N6H) [61] as a template for homology modeling. The N-terminal region of the active structure was removed, while the missing residues of helix 8 (Η8) were added so that both proteins consisted of residues 65 to 352. All missing residues and side chains were added using MODELLER [66]. The resulting structures were coarse-grained according to the Martini force field version 2.1 [67–69] using the martinize.py script. Receptor tertiary structure was maintained with a modified version of the elastic network [70,71]. Specifically, a harmonic force was applied between all BB beads within a cutoff of 0.9 nm using a force constant of 1000 kJ mol-1 nm-2 for helical residues, or 250 kJ mol-1 nm-2 for residues in unstructured regions. In agreement with experimental inferences [5], C1703.55 (the superscript follows the Ballesteros-Weinstein numbering scheme [72]) was palmitoylated by adding four C1 beads with a bond length of 0.47 nm and force constant of 1250 kJ mol-1 nm-2 and angles of 180° with a force constant of 25 kJ mol-1 rad-2.
Two membrane sizes were used for the simulations reported in this work. Arrays of 16 coarse-grained proteins were placed in either a large membrane patch of 50×50 nm2 or a smaller patch of 25×25 nm2, corresponding, respectively, to 6.4×103 and 2.5×104 receptors/μm2. We use “low receptor density” to refer to the large membrane patch and “high receptor density” to refer to the small membrane patch. The proteins were evenly spaced, and each of the protomers was randomly rotated around its z-axis. Two types of arrays were created, using all inactive or all active receptors, respectively. For each of the four protein set-ups (active and inactive conformations, high and low receptor density), five sets of arrays were created for a total of 20 starting protein configurations. Furthermore, a third type of array, containing 50% inactive/50% active (hereafter indicated as “mixed arrays”) receptors, was prepared (in five replicas) for the high receptor density setup, adding 5 more starting receptor configurations. In the case of the mixed arrays, 50% of the proteins were randomly assigned to be inactive, while the others were in the active conformation (see S20C Fig for an example of an initial configuration of mixed arrays for the high density system, while S20 Fig Panels A and B show examples of all inactive and all active receptor set-ups for the high receptor density systems, respectively).
Using the insane.py [73] script, the protein arrays were then embedded in a coarse-grained 63-component plasma membrane with the same composition as that obtained by Ingólfsson et al. [27] (see S3 Fig) scaled so that the protein to lipid ratio was approximately 1:200 and 1:100, respectively for the low and high receptor density setups. The total number of lipids in each membrane was approximately 3200 (or 800) in the upper leaflet and 3000 (or 750) in the lower leaflet for the low receptor density (or high receptor density) systems. As in Reference [27], a 2 kJ mol-l nm-2 restraint was placed on the phosphate bead of POPC and PIPC lipids in the upper leaflet in the z direction to prevent large membrane undulations. The height of the box was set to 11 nm.
The protein-membrane systems were solvated with water, and ions were added to neutralize the total charge. To provide a direct comparison of the plasma membrane without proteins, the final structure from the 40 μs trajectory of Ingólfsson et al. [27] was retrieved from the Martini force field website (http://md.chem.rug.nl) and used as the starting point for a 700 ns trajectory using the same settings as the simulations with the receptors.
Following 10,000 steps of steepest descent energy minimization, the systems were simulated for 100 ns using a timestep of 10 fs keeping position restraints on the backbone beads of the receptor. To equilibrate the membrane and determine the distribution of the lipids around the individual protomers, each system was run for 5 μs using a timestep of 20 fs, while still maintaining position restraints on the receptors. In preparation for the production run, four 10 ns runs with decreasing restraints on the proteins (500, 100, 50, and 10 kJ mol-1 nm-2) were performed. The production runs were 20 μs and 3 to 5 μs long for the high receptor density and the low receptor density systems, respectively, giving a cumulative time of over 300 μs for the three simulated protein combinations. The simulations were run in the NPT ensemble, with reference temperature of 310 K controlled with the v-rescale algorithm (τt = 1.0 ps) [74], and reference pressure of 1 bar, controlled with the Berendsen algorithm (τp = 5.0 ps) [75]. The Coulomb interactions between 0 and 1.2 nm decayed smoothly to 0, while the van der Waals interactions between 0.9 and 1.2 nm decayed smoothly to 0. All simulations were performed with GROMACS 4.6 [76,77].
The analysis of the lipids around receptor protomers was performed on the final 2 μs of the membrane equilibration part of the simulation–in which the proteins were kept fixed with position restraints–and on the protomers extracted from the production runs. In the latter, a receptor was considered monomeric if the distance between the center of mass of the protein and every other protein was at least 5 nm. Since the atomic coordinates were recorded every ns and each frame has 16 receptors, each lipid distribution around the receptors was calculated from 5×2000×16 sets of lipid-protein positions. The analysis of the lipids around the dimers was performed on the frames from the final 1 μs of simulation time of the high receptor density production runs, which was also used for interface clustering. The ROH beads of the CHOL or the first linker bead (GL1 or AM1) of the non-sterol lipids were used for the lipid analysis (see S1 Fig for a depiction of representative lipids). Lateral lipid density was calculated by binning the position of the lipid beads into 50×50 square bins with a side of 0.2 nm in the membrane plane and calculating the normalized probability distribution. In order to quantify the order and disorder of the lipid tails, the angle between the average membrane normal (z direction) and the vector from the linker bead (AM1/2 or GL1/2) to the last bead of the tail was calculated for both tails of each lipid. While the metric used to calculate the lipid order, which is akin to that suggested by Katira et al. [46], does not follow the traditional definition of lipid order parameter, it is more computationally efficient, and it allows us to use the same metric to characterize the order of non-sterol lipids and CHOL. While CHOL does not have a tail, the angle was calculated between the membrane normal and the vector from the ROH bead to the final bead. For CHOL, the membrane was split in three parts with the middle of the bilayer defined to be 1.6 nm thick by plotting the histogram of the z-coordinate of the ROH bead. Order distribution plots were obtained by averaging the order in 1.75×1.75 Å2 square bins parallel to the xy plane. The local thickness of the membrane was defined as the difference between the average z-coordinate of the linker bead to which the headgroup is attached (AM1 or GL1) in each of the two leaflets, on the same 1.75×1.75 Å2 grid used to calculate the order. The dependence of the membrane modulation on the relative position of proteins was analyzed by selecting all frames in both the high receptor density and low receptor density production runs with two protomers at a given distance r = ||R1–R2|| (where R1 and R2 denote the COMs of the two proteins), and with relative orientation described by angles α and β in the regions α∈Ωα = [α0–π/6, α0+π/6] and β∈Ωβ = [β0–π/6, β0+π/6], where α0 and β0 specify the region of the two protomers facing the dimerization interface (0 or –3/4π, respectively for the TM1,2,H8 and TM5,6, interfaces). The position lipid molecules relative to the protomers was described as dvd+nnd in the frame of reference given by the versor vd∝R1–R2 and its normal nd. We report averages of membrane thickness and order over Ωα, Ωβ, and n<r0 (where r0 = 1.7 nm is the average protomer radius), as a function of r (protomer distance) and d (lipid position). Only frames for which no other protein was found within r0 from the line connecting the COMs (i.e., n<r0) of the two protomers were included in the analysis. The averages are calculated and reported for d∈[r0, r–r0].
A kinetic model of the CHOL movement was generated with PyEmma [78] using the contacts formed between the ROH and BB beads to perform the geometric clustering. A hidden markov model was used to kinetically lump the clusters into 8 macrostates.
The mean residence time of the lipids around a helix of the monomer was calculated for the production runs together with the last 2 μs of their corresponding equilibration runs. For every lipid, the length of time spent within 1.2 nm of any sidechain bead (SC) of each helix was calculated in order to obtain a distribution of residence times per lipid species and helix per run. The total mean residence times and standard deviation were obtained by averaging over the distributions from each replica.
All interface analysis was performed on the high-density simulations. To characterize the interfaces formed during the production runs, k-means clustering was performed on the final 1 μs of the trajectories, using the Euclidean norm d2 = Tr(DTD) of the difference of contacts maps D as a dissimilarity measure. An interface was considered to be formed when at least ten residues on each protomer formed contacts with the other protomer, with a contact defined as two backbone beads lying within 0.8 nm of each other. Helices with three or more residues forming contacts were used in the interface name. The relative frequency of each observed interface and their variance were estimated using a multinomial model Xj ~ Multinomial(n,pj), where Xj is the number of observed dimers for interface j and n is the total number of observed dimers. Specifically, the well-known relation to the Poisson model was exploited to rewrite the model as Xj ~ Poisson(λj) and pj = λj/∑ λj. We took a Bayesian approach, and sampled the posterior distributions of pj with normal uninformative (standard deviation 100) priors on γj = log(λj) using the observed values of Xj for each trajectory. Results are reported as mean and 95% credible intervals. Sampling was performed with the rstan 2.8.0 interface to the Stan language [79]. The RMSD of the simulated interfaces relative to the crystal structures was determined after aligning the Cα of the two dimers. The RMSD was calculated using only the Cα atoms of the helices participating in the interface to ensure that the RMSD captured only the differences between the interfaces. The structures of the interfaces were rendered in pymol. Scripts for the lipid analysis and interface clustering employed the MDAnalysis python libraries [80].
|
10.1371/journal.pgen.1004975 | The Arabidopsis DNA Polymerase δ Has a Role in the Deposition of Transcriptionally Active Epigenetic Marks, Development and Flowering | DNA replication is a key process in living organisms. DNA polymerase α (Polα) initiates strand synthesis, which is performed by Polε and Polδ in leading and lagging strands, respectively. Whereas loss of DNA polymerase activity is incompatible with life, viable mutants of Polα and Polε were isolated, allowing the identification of their functions beyond DNA replication. In contrast, no viable mutants in the Polδ polymerase-domain were reported in multicellular organisms. Here we identify such a mutant which is also thermosensitive. Mutant plants were unable to complete development at 28°C, looked normal at 18°C, but displayed increased expression of DNA replication-stress marker genes, homologous recombination and lysine 4 histone 3 trimethylation at the SEPALLATA3 (SEP3) locus at 24°C, which correlated with ectopic expression of SEP3. Surprisingly, high expression of SEP3 in vascular tissue promoted FLOWERING LOCUS T (FT) expression, forming a positive feedback loop with SEP3 and leading to early flowering and curly leaves phenotypes. These results strongly suggest that the DNA polymerase δ is required for the proper establishment of transcriptionally active epigenetic marks and that its failure might affect development by affecting the epigenetic control of master genes.
| Three DNA polymerases replicate DNA in Eukaryotes. DNA polymerase α (Polα) initiates strand synthesis, which is performed by Polε and Polδ in leading and lagging strands, respectively. Not only the information encoded in the DNA, but also the inheritance of chromatin states is essential during development. Loss of function mutants in DNA polymerases lead to lethal phenotypes. Hence, hypomorphic alleles are necessary to study their roles beyond DNA replication. Here we identify a thermosensitive mutant of the Polδ in the model plant Arabidopsis thaliana, which bears an aminoacid substitution in the polymerase-domain. The mutants were essentially normal at 18°C but arrested development at 28°C. Interestingly, at 24°C we were able to study the roles of Polδ in epigenetic inheritance and plant development. We observed a tight connection between DNA replication stress and an increase the deposition of transcriptionally active chromatin marks in the SEPALLATA3 (SEP3) locus. Finally, we tested by genetic means that the ectopic expression of SEP3 was indeed the cause of early flowering and the leaf phenotypes by promoting the expression of FLOWERING LOCUS T (FT). These results link Polδ activity to the proper establishment of transcriptionally active epigenetic marks, which then impact the development of multicellular organisms.
| Arabidopsis is a facultative long-day (LD) plant, meaning that LDs accelerate flowering whereas in short days (SD) flowering is delayed. Flowering in spring (LDs) is therefore promoted by GIGANTEA (GI), CONSTANS (CO) and FLOWERING LOCUS T (FT) which constitute the so called “photoperiod pathway”. GI activates CO which eventually accumulates in the late afternoon and early night under LD conditions and induces the expression of FT in phloem companion cells. The protein FT is an universal florigen and moves to the apical meristem to promote the transition to flowering [1–3].
In plants, epigenetic inheritance confers a cellular memory of past environmental conditions. The winter ecotypes of Arabidopsis thaliana require a prolonged exposure to near freezing temperatures, or vernalization, to become competent to flowering in spring [4]. Plants are able to “remember” the past winter because persistent cold exposure (2–4 weeks) activates an epigenetic mechanism that requires the trimethylation of Lysine 27 of histone 3 (H3K27me3), which permanently represses FLOWERING LOCUS C (FLC) [4].
The pathways that respond to photoperiod and vernalization are integrated at the level of FT, which is directly repressed by FLC [4,5]. The FT gene is also known as a “flowering integrator” because it also responds to other flowering pathways including the thermosensory and autonomous pathways [4,5]. The complex regulation of FT expression is due at least in part to epigenetic mechanisms [6–10]. The chromatin of FT is enriched in H3K27me3; curly leaf (clf) and terminal flowering 2/like heterochromatin protein 1 (tfl2/lhp1) mutants, which are defective in the H3K27 methylase and H3K27me3 binding activity, are early flowering and photoperiod insensitive due to high expression of FT [9,11–13].
The terminal differentiation of specialized tissues in multicellular organisms is strongly influenced by the previous DNA replication events of the cells that constitute each tissue. In Eukaryotes, three replicative DNA polymerases (Polα, Polε and Polδ) are responsible for the faithful duplication of the nuclear genome [14]. Polα forms a complex with a primase to initiate replication at origins and Okazaki fragments and these short stretches of RNA-DNA hybrids are extended by Polε in the leading strand and Polδ in the lagging strand [14].
Apart from their essential role in DNA replication, Polα and Polε are required for other non-essential functions such as the maintenance of transcriptional silencing in yeast. In this alternative role, Polε is part of a mechanism that processes transcripts into siRNAs to reinstall transcriptional silencing [15]. In plants, mutants versions of Polα and Polε led to changes in histone marks, resulting in elevated FT expression together with increased flower homeotic gene expression, which produced early flowering and curly leaf phenotypes [11,16–19]. However, which genes are the primary targets of the defective-polymerase induced epigenetic changes remains unclear.
Unlike Polα and Polε, the role of the Polδ in epigenetic inheritance has not been addressed so far. Besides lagging strand synthesis, Polδ also participates in many other processes that repair DNA lesions required to protect genome integrity [20,21].
Reports in yeast and plants have shown that a reduction in the amount of Polδ leads to genome instability and hyperrecombination phenotypes [20–22]. In mammals, mutations in the proofreading domain of Polδ produce predisposition to cancer. For instance, a POLD1 S478N variant in human populations predisposes to colorectal tumors and endometrial cancer [23].
Although the fact that replicative polymerases are essential in most studied organisms, viable mutant alleles of pola1 (At5g67100, encoding the catalytic subunit of Polα and pole1 (At1g08260, encoding the catalytic subunit of Polε were isolated in multiple genetic screenings [11,16,18,19,24,25]. However, viable hypomorphic alleles of POLD1 were not isolated so far, which may be due to its essential roles in DNA replication and repair. Here, we report the isolation of gigantea suppressor 5 (gis5), the first plant POLD1 viable mutant allele, which also proved to be thermosensitive. Under restrictive temperatures, the gis5 allele led to early flowering and curly leaf phenotypes which were dependent on the FT gene but caused by overexpression of SEP3, which showed a correlation with increased trimethylation of Lysine 4 (H3K4me3) at the SEP3 locus. These phenotypes mostly disappeared at permissive temperatures. Our findings reveal an unforeseen function of Polδ that may be linked to the correct establishment of transcriptionally active epigenetic marks during DNA replication.
We performed a genetic screen for induced mutations that suppressed the late flowering phenotype of gi-2 mutants, with the aim to isolate genes involved in the interaction between photoperiod and thermosensory pathways. One of the ethyl methanesulfonate (EMS)-induced mutations suppressed most of the gi-2 late flowering phenotype under long days (LD) conditions and was named gigantea suppressor 5 (gis5) (Fig. 1A). The gis5 mutation also accelerated flowering in a wild type (WT) background (Fig. 1A) so we used gis5 in this background in subsequent experiments. We also observed that in the WT background—but not in the gi-2 background–, gis5 displayed a curly leaf phenotype, reminiscent of curly leaf (clf) mutants [12]. Interestingly, the curly leaf phenotype depended on temperature, i.e.: it was strong at 24°C but disappeared at 18°C (Fig. 1B-C). We decided to evaluate whether the flowering phenotype of gis5 mutants was also temperature-dependent. Under LD, the flowering phenotype of gis5 mutants was relatively insensitive to temperature in the range 18–24°C (Fig. 1D). In stark contrast, the gis5 early flowering phenotype was mostly suppressed when plants were grown at 18°C under short days (SD) (Fig. 1E).
We mapped gis5 to a 120kb interval delimited by markers CER436434 and CER436454 (http://www.arabidopsis.org/browse/Cereon/index.jsp [26]) (Fig. 2A). We detected a C to T transition in the 18th exon of the gene encoding the catalytic subunit of Polδ (POLD1, AT5g63960) which led to a A707V substitution (Fig. 2A). To confirm that the gis5 mutation was the cause of the observed phenotypes, we complemented gis5 mutants with a WT genomic fragment containing the complete POLD1 gene with its own promoter and terminator sequences. Four independent transgenic lines, with single locus T-DNA insertions were evaluated and in all cases, the curly leaf phenotypes and the early flowering under both LD and SD were complemented to a high degree (Fig. 2B; S1A Fig.).
The A707 residue is perfectly conserved in δ DNA polymerases from other eukaryotes (S1B Fig.). We modeled WT and mutant POLD1 using the yeast crystal structure of Pol3 as a template (PDB: 3IAY) [27]. Both proteins share 50% identity, which suggests that this particular model could be as accurate as one obtained from low resolution X-ray crystallography [28]. The A707V mutation is located in a α-helix from the finger domain (Fig. 2C) which interacts with the nucleotide substrate during DNA polymerization. Interestingly, the A707 does not interact directly with the substrate and the A to V substitution does not affect the protein structure in any obvious manner (Fig. 2C, lower panel). Further, the A707 residue is inaccessible to the solvent, suggesting that it is not directly involved in protein to protein interactions (Fig. 2C; S1C Fig.). As the yeast Polδ conformation changes upon substrate binding [29], we modeled POLD1 in both substrate-free (4FVM) and bound (4FYD) conformations using X-Ray models from POLA1 (29% identity) as references (S1C Fig.). The α-helix bearing the A707 is greatly displaced when comparing both models, suggesting that the α-helix moves during catalysis (S1C Fig.) and that the A707V substitution may affect POLD1 activity. This was further supported by the fact that Val is known to destabilize α-helices, when replacing Ala residues [30]. Hence, the gis5 mutation might increase the finger instability at higher temperatures.
We reasoned that if the observed temperature-dependent phenotypes were due to a defect in the activity of Polδ, then we should observe similar temperature dependence on other phenotypes not related to flowering. It has been previously shown that suppression of Polδ by RNAi triggers a DNA replication stress response, including an increase in Homologous Recombination (HR) [31]. Hence, we tested if gis5 mutants displayed a DNA replication stress response and whether this response was temperature-dependent. Interestingly, the mRNA levels of BRCA1 and RAD51, two genes involved in HR, increased at 24°C in the gis5 mutant, but only relatively weak effects were observed at 18°C (Fig. 3A). To evaluate if these changes promoted HR, we used HR reporter lines that bear halved fragments of GUS reporter genes which are reconstituted after HR events [31]. We quantified HR events, seen as blue dots after X-Gluc staining, and observed that HR events were relatively few at 18°C and greatly increased in gis5 mutants (more than 100-fold) by growing plants at 24°C (Fig. 3B; S2 Fig.).
To further test if problems in DNA replication appeared at higher temperatures we introduced a pCYCB1–1:GUS reporter line in gis5 mutants. This reporter is used to reveal cells in late G2 phases [32]. The number of GUS stained cells increased in gis5 roots, but only at 24°C, thus suggesting that, at higher temperatures, defects in DNA replication accumulated in gis5 cells and demanded more time for HR-dependent repair during the G2/M transition (Fig. 3C).
We reasoned that if the gis5 allele were thermosensitive, further increasing temperature above 24°C would impair development. gis5 mutants grown at 28°C were severely affected, resembled dwarfed plants that did not set seeds and eventually died even in axenic culture (Fig. 3D).
To evaluate which flowering pathways are affected by the gis5 mutation, we investigated the epistatic relationships between gis5 and those mutations affecting the photoperiod, autonomous and vernalization pathways (Fig. 4A-B).
Mutations in the photoperiod pathway co-9 and gi-2 did not suppress gis5 early flowering phenotype under both LD and SD conditions, suggesting that gis5 is acting either downstream of these genes in the photoperiod pathway or in a parallel pathway (Fig. 4A-B).
The effects of autonomous and thermosensory pathways mutations [33], fve-3 and fca-9, were mostly additive to gis5 and the double mutants showed intermediate flowering phenotypes, suggesting that gis5 affects parallel pathways to those affected by fve-3 and fca-9 (Fig. 4A-B).
FLC acts downstream of the vernalization and autonomous pathways. The gis5 mutation produced a decrease in FLC mRNA levels in WT and fca-9 backgrounds which could indicate that gis5 might control flowering through the levels of FLC transcripts (S3A Fig.). Notwithstanding, gis5 mutants flowered much earlier than flc mutants, especially in SD (Fig. 4A-B), and the flc mutation did not further accelerate flowering in the gis5 background, suggesting that gis5 was acting mostly downstream of FLC. Hence, a role for FLC in gis5 early flowering seemed minor in the Col background used here, but may be more important in vernalization requiring accessions. Interestingly, the ft mutation suppressed most of the gis5 early flowering phenotype under both LD and SD (Fig. 4A-B), suggesting that FT acts downstream gis5. A mutation in the other flowering integrator gene, suppressor of overexpression of constans 1 (soc1), resulted in an intermediate effect when evaluated in the gis5 background, which is consistent with SOC1 being one of the downstream targets of FT [34]. A mutation in a third integrator gene, twin sister of ft (tsf), showed marginal effects in our conditions in gis5 and WT genetic backgrounds. Together, these results suggested that FT could be downstream of gis5 and prompted us to measure FT mRNA levels. FT was highly expressed in the gis5 background in both continuous light and SD (Fig. 4C-D). Further, the expression of FT was highly dependent on temperature in the gis5 mutant background.
To test if ectopic or tissue specific overexpression of FT was related to the gis5 phenotype, we evaluated the pattern of expression of the β-glucuronidase (GUS) reporter gene under the FT promoter [6]. GUS expression in gis5 P8.1kbFT:GUS plants was limited to the vascular tissue and was not detected in the apical meristem (S3B Fig.). These results, together with qRT-PCR data (Fig. 4C-D), show that FT is overexpressed in vascular tissue in gis5 mutants. To test if this overexpression is functional, we used artificial microRNAs directed against FT mRNA, as previously reported [3]. When artificial microRNAs were expressed under the companion-cell specific promoter SUC2, the early flowering of gis5 was greatly suppressed, while expression under the apical meristem specific FD promoter had no effect in the gis5 mutant background (Fig. 4E). These results taken together showed that FT overexpression within vascular tissue was necessary for the gis5 early flowering phenotype and were also consistent with the curly leaf phenotype observed in strong FT overexpressors [35].
Despite the fact that high expression of FT could explain both the temperature dependence and the leaf and early flowering phenotypes of gis5 mutants, the underlying mechanism was unclear.
The data presented above show that FT acts downstream gis5 and is also necessary for the expression of gis5 early flowering and curly leaf phenotypes. However, how a mutation in Polδ produced such effects was still unclear. To obtain an insight on the mechanisms, we performed a microarray analysis to study the transcriptome of gis5 mutants to investigate if other factors could be acting upstream FT and be direct targets of the gis5 allele. The genes were ordered based on the effect of gis5 on their expression. Only a few flowering genes were found among the upregulated and downregulated genes. SEP3 was at the top of the list, which also included SEP1 and SEP2 (S1 Table). Intriguingly, it has been reported that SEP3 overexpression accelerates flowering and leaf curling [36–38]. Further, a recent report has shown that SEP3 can act also upstream of FT and that both genes mutually regulate each other in a positive manner [38]. Hence, we decided to evaluate the expression of SEP genes under different photoperiod and temperature conditions in gis5 mutants. SEP1, SEP2 and SEP3 mRNAs were expressed at very high levels in the gis5 mutants under continuous light, and this effect was highly dependent on temperature (Fig. 5), which correlates well with the phenotype of gis5 mutants.
As FT was reported to act upstream of SEP3 in a thermosensory pathway [39], the temperature-dependence of gis5 phenotypes could be due to the amplification of an underlying response to temperature. We reasoned that under SD conditions and low temperatures (18°C), SEP3 expression should be independent of both the photoperiod and the thermosensory pathways; in contrast, under elevated temperatures (24°C) the thermosensory pathway would increase SEP3 expression in an FT-dependent manner. Hence, we also tested the expression of SEP genes in plants grown in SD in the ft mutant background. The three SEP genes were affected by the ft mutation in the gis5 background only at 24°C and their expression dropped by about 50% in the double gis5 ft mutants (Fig. 5B), which is consistent with SEP3 acting downstream FT in the thermosensory pathway [39]. However, the expression of SEP genes was highly elevated (30 to 100 fold) in the double gis5 ft mutants with respect to ft single mutants, revealing an FT-independent effect on SEP gene expression in gis5 plants. Interestingly, despite the high expression levels of SEP3 in gis5 plants, FT was still required for early flowering (Fig. 4). Among the SEP genes, SEP3 showed the strongest temperature response in the gis5 ft double mutants (4.4 fold), SEP1 showed a partial response (1.7 fold) and SEP2 showed no response to temperature. Taken together, these results suggested that at higher temperatures the gis5 mutation might cause increased FT expression and curly leaf as well as early-flowering phenotypes by elevating SEP3 expression.
Since the SEP3 locus was shown to be actively repressed by H3K27me3 marks [40], we evaluated if the V707A mutation in Polδ could affect histone marks in the SEP3 locus in a manner that is dependent on the temperature. We performed chromatin immunoprecipitation (ChIP) experiments to quantify H3K27me3, H3Ac and H3K4me3 enrichment at the SEP3 locus in WT and gis5 mutants grown at either 18 or 24°C. We did not find changes in H3K27me3 marks enrichment but a significant enrichment (about 5-fold) in H3K4me3 marks, which peaked in the first intron of SEP3 and decreased towards the 3´end (Fig. 6, top right panel, S4 Fig. and S7 Fig.). Importantly, this enrichment in H3K4me3 marks on the SEP3 locus was also dependent on temperature, showing a correlation with expression data (Fig. 5, lower left panel, Fig. 6, top panels, and S4 Fig.). H3Ac also increased in a temperature-dependent way and more likely reflects the increased transcriptional activity at the SEP3 locus [41].
Since mutations in Polα and Polε were proposed to affect histone mark deposition at FT and FLC loci [11,16,17,19], we decided to study the deposition of histone modifications at both loci, in the WT and gis5 mutants at both 18 and 24°C. Despite higher FT mRNA and lower FLC mRNA levels in gis5 mutants, ChIP against H3Ac, H3K4me3 and H3K27me3, followed by qPCR of locus-specific fragments, did not reveal temperature-dependent changes at both the FT and FLC loci, except for a marginal increase in H3K4m3 at the FT locus which was not statistically significant (S5 Fig., S6 Fig. and S7 Fig). These results supported the notion that the SEP3 locus was a primary target of a gis5-produced epigenetic modification.
To test if the elevated SEP3 expression could be the cause of the early flowering and curly leaf phenotypes of gis5 mutants, we suppressed SEP3 expression in the gis5 background using artificial microRNAs. Transgenic gis5 lines bearing artificial microRNAs against SEP3 showed low SEP3 expression, decreased FT expression, later flowering and plain leaves (Fig. 7; S5A Fig.). On the contrary, microRNAs against SEP1 did not show a significant effect on their own and subtle effects (if any) when coexpressed together with a microRNA against SEP3 (Fig. 7A). Further, expression of a microRNA against SEP3 under the phloem specific promoter SUC2 led to a suppression of both early flowering and curly leaf phenotypes (Fig. 7) while the same microRNA under the FD promoter did not show any effect (Fig. 7C).
These pieces of evidence strongly support the notion that increased H3K4me3 marks on the SEP3 locus and concomitant overexpression in phloem tissue are the main cause of the early flowering and leaf phenotypes of the gis5 mutants. Conversely, the DNA replication stress response in gis5 mutants was independent of SEP3 expression levels, since the expression of HR marker genes BRCA1 and RAD51 in gis5 background was not suppressed by suppressing SEP3 expression (S8B Fig.). These results strongly suggest that the DNA replication stress response is activated upstream SEP3 in gis5 mutants and it is not a byproduct of accelerated development.
The FLC expression levels were lower in gis5 mutants and interestingly, these effects were also independent of SEP3 (S8A Fig.) suggesting that FLC might have a minor SEP3-independent role in gis5 early flowering.
The results shown above are consistent with a model where a decrease in Polδ activity in gis5 mutants at higher temperatures leads to an increase in the expression of SEP3, resulting in a feedback loop with FT in vascular tissue to induce flowering and curly leaves phenotypes (Fig. 7D).
Despite SEP3 overexpression can account for the gis5 phenotypes tested here, it remains unclear whether the gis5 mutation leads to H3K4me3 increases in other loci. Our microarray results showed that the expression of only 81 genes changed by at least two-fold in the gis5 mutant with respect to the WT, suggesting that the effects of the gis5 allele are restricted at most, to a relatively small number of loci. To test if other loci displayed changes in H3K4me3 levels, we performed ChIP experiments with a subset of genes selected from the microarray, SEP1, PCC1, and ASN. Both SEP1 and PCC1 were highy expressed (S1 Table) and also displayed a temperature-dependent increase in H3K4me3 levels (S9 Fig.), whereas ASN was downregulated and did not display increased H3K4me3 levels (S9 Fig.). These results show that the gis5 allele might affect other loci diffetent from SEP3 which may account for other phenotypes not evaluated here.
Here we describe the isolation of a novel flowering mutant, gis5, which flowers early and displays curly leaves. These phenotypes are due to an A707V substitution in the catalytic subunit of the Arabidopsis Polδ encoded by the POLD1 gene. Null POLD1 alleles are embryo lethal (http://www.seedgenes.org/ [31]). To our knowledge, gis5 is the first hypomorphic and viable allele to be isolated and, interestingly, it is thermosensitive. This is supported by the weak phenotypes of gis5 mutants grown at 18°C, the strong early flowering and curly leaf phenotypes of plants grown at 24°C and the lethality of plants grown at 28°C. Further, the gis5 mutation also produced a DNA replication stress response and an increase in HR which were also temperature-dependent. It is likely that at 18°C both Polε and Polδ advance in the replication fork in a coordinated manner. At 24°C the lower activity of the gis5 allele would lead to larger single strand DNA stretches and eventually to DSB and increased HR, which is the mechanism to repair DSB [20]. Genetic instability was previously reported in Arabidopsis lines with low POLD1 expression levels [31]. However, defects in epigenetic inheritance were not observed in those RNAi lines and transcriptomic analysis in those lines did not reveal changes of importance. These results, together with our data suggest that the gis5 effects on epigenetic inheritance might result from a specific change in Polδ behavior rather than just decreased levels.
In a previous report, the incorporation of H2AZ to nucleosomes was proposed to be a mechanism of temperature perception, and its failure led to early flowering [42]. It is unlikely that a similar effect in temperature responsiveness is occurring in gis5 mutants because the genes misexpressed in gis5 mutants do not overlap with those misexpressed in mutants that fail to incorporate H2AZ [42]. Only one out of ten of the temperature responsive markers followed up by Kumar & Wigge (2010) showed a significant change in our microarray data of gis5 mutants.
The early flowering and curly leaves phenotypes of gis5 mutants are caused by high expression levels of SEP3 which activates FT in phloem tissue. This is supported by our expression and genetic data and also consistent with previous reports showing that SEP3 overexpression accelerates flowering and produces curly leaves [36–38]. SEP3 is well known to play roles in flower development downstream of FT [36,37], which is also consistent with SEP3 mRNA levels decreasing by about 50% in the ft gis5 double mutants with respect to gis5 single mutants. However, SEP3 mRNA levels were still very high (about 30-fold) in ft gis5 double mutants compared to either ft single mutants or WT plants indicating that gis5 increases SEP3 expression independently of FT. These results led us to propose that SEP3 and FT form a positive feedback loop in gis5 mutants, similar to the mutual activation of SEP3 and FT reported in clf mutants [38], although we do not have evidence to assume that this mutual regulation is direct. The SEP3-FT feedback loop also explains why gis5 early flowering is more dependent on temperature under SD compared to LD and continuous light. Under LD, FT mRNA levels increase in response to the photoperiod pathway contributing to the feedback loop and compensating for the decrease in SEP3 gene activation at lower temperatures.
The strong changes in SEP3 expression and the increases in H3K4me3 at the SEP3 locus were both temperature-dependent and correlated well with the DNA replication stress responses and the increase in HR, suggesting that changes in SEP3 epigenetic marks and expression were produced by the changes in the dynamics of DNA replication. Despite the fact that we cannot exclude the possibility that the effects on the SEP3 locus were indirect (i.e. by activating a SEP3 activator), we find this explanation rather unlikely because i) SEP3 was the top upregulated gene in gis5 mutants by our microarray data, ii) known SEP3 regulators or SEP3 corregulated genes such as FUL and SPL3 were not upregulated in our microarrays and iii) two direct transcriptional repressors of SEP3, SVP and AGL24 [43], were not downregulated in our microarrays. Hence, these pieces of evidence strengthen the idea that the effects of gis5 are direct on the SEP3 locus.
gis5 could affect SEP3 expression by changing its pattern of histone post-translational modifications. Increased H3Ac more likely reflects the increased transcriptional activity at the SEP3 locus [41,44]. H3K4me3 also correlates with increased transcriptional activity but it was also shown to function as a memory of recent transcriptional activity [41,44,45]. Two scenarios are then possible, that the gis5 mutation produces an increase in SEP3 expression and as a consequence, an increase in H3K4me3, or that an increase in H3K4me3 causes an increase in SEP3 transcription. We favor this second hypothesis. An increase H3K4me3 deposition at the SEP3 locus could result from of an interaction between Polδ and the local chromatin during the maturation of Okazaki fragments. In yeast, the ligation of Okazaki fragments take place at nucleosome midpoints, implying that nucleosomes are loaded immediately after the passage of the replication fork [46]. Further, some transcription factor binding sites are also preferentially sites of Okazaki fragment ligation. These DNA bound proteins restrain excessive strand displacement by Polδ during Okazaki fragment maturation, causing Polδ to dissociate from DNA. When Polδ processivity was perturbed, the site of Okazaki fragment ligation changed consistently with Polδ dissociating before the nucleosome mid-point [46]. These data raise the possibility that the gis5 allele of Polδ may be more sensitive to the local chromatin structure at the SEP3 locus dissociating earlier during local Okazaki fragment maturation. If strand-displacement synthesis by Polδ is required to remove H3K4me3 incorporated during the previous Okazaki fragment synthesis, Polδ premature dissociation would eventually lead to overaccumulation of H3K4me3, which could then be involved in a positive feedback loop with transcription.
Whether the mechanisms underlying the early flowering and curly leaf phenotypes of mutants affected in Polα and Polε are similar to the mechanisms underlying gis5 phenotypes is currently unclear. First of all, the Polδ has two chances of interacting with specific nucleosomes and DNA bound proteins, first during the extension of Okazaki fragments and then during their maturation, distinguishing Polδ from Polα and Polε. Interestingly, hypomorphic alleles of POLA1 and POLE1 also led to higher SEP3 expression levels [11,16,18,19], raising the possibility that common mechanisms may produce the phenotypes of all DNA polymerase mutants. However, the role of epigenetic modifications at the SEP3 locus were neither investigated further nor were the epistatic interactions between the SEP3 locus and the polymerases alleles. Further, for both Polα and Polε it was proposed that their mutations affected the interaction with LIKE HETEROCHROMATIN 1 (LHP1) [11,16–19], a protein with H3K27me3 binding activity which is required to repress target genes [47,48], although some level of controversy remained on whether the proposed interactions were direct [17]. Protein modeling of the gis5 allele suggests that the A707V substitution is unlikely to change the direct interactions with other proteins, given that this residue is not solvent accessible and very close to the nucleotide binding site, favoring the interpretation that gis5 changes the dynamics of DNA replication rather than the recruitment of histone methylation complexes, which is supported by the expected movements of the finger domain α-helix that contains the A707V substitution. Obtaining analogous mutations to gis5 in DNA Polα and Polε catalytic subunits will likely shed light on the possible common mechanisms that replicative polymerases may use to reestablish epigenetic marks during DNA replication.
The presence of the SEP3-FT feedback loop and the possible interaction of gis5 with the SEP3 locus accounts for the specificity of gis5 effects. However, one remaining important question is whether the gis5 effects are widespread all over the genome. Despite we found epigenetic changes in other loci different from SEP3 (S9 Fig.), the relatively low number of genes whose expression is altered in gis5 mutants (81 with a two-fold difference with WT, S1 Table), supports the notion that the effects of gis5 are specific for a relatively low number of loci. As discussed above for the SEP3 locus, these effects in some specific loci could result from interactions that may occur during the maturation of Okazaki fragments between Polδ and DNA bound proteins, which are loaded immediately after the passage of the replication fork [46]. Noteworthy, specific effects of mutations in the catalytic domain of Polδ are not exclusive of plants. A high specific effect was also observed in humans; a Ser605 deletion in the Polδ catalytic site is lethal in homozygosis but heterozygous individuals showed unexpected tissue specific phenotypes: mandibular hypoplasia, deafness and progeroid features (MDP) syndrome [49]. Other mutations which affect the proofreading domains of human Polδ and Polε were associated to cancer susceptibility, which is consistent with the mutator phenotype expected for these alleles [23]. In contrast, the molecular basis for the Ser605-deletion-triggered MDP is currently unknown and there is no evidence supporting a mutator phenotype or an increase in cancer susceptibility [49]. Our work raises the possibility that an epigenetic change on a small subset of master regulators could also explain the apparent specificity of the MDP produced by deletion of Ser605 in humans, as we show here for SEP3 in plants. In the same line of reasoning, epigenetic effects could also be part of the equation in the progression of tumors bearing defective alleles of replicative DNA polymerases, which could add to the mutator phenotypes of these defective alleles [23]. We think that the characteristics of the gis5 allele will be invaluable in future studies on the interplay between the replication of the lagging strand, DNA replication stress, epigenetic inheritance and development in multicellular organisms.
All the alleles and transgenic lines used were obtained in the Columbia background: ft-10 (GABI_290E08) [34], tsf-3 (SALK_087522) [50], soc1–2 (SALK_006054) [51], co-9 (SAIL_24_HO4) [52], gi-2 (CS3397) [53], flc-201 (SALK_003346) [54], fca-9 [55] and fve-3 [56], recombination GUS reporter lines 1406 (direct repeat line) and 1415 (inverted repeat line) [57,58], P8.1kbFT:GUS [6], PCycB1;1:GUS [32], and transgenic lines expressing amiRNA-FT (artificial microRNAs) under the 35S, SUC2 and FD promoters [3]. Seeds were sterilized with chlorine in the vapour phase and, depending on the experiments, plants were grown on a 1:1:1 soil, vermiculite and perlite mix and every two weeks plants were fertilized with a 0.1% solution of Hakaphos (Compo Agricultura, http://www.compo.es), or on plates with MS salts medium (DUCHEFA). Plants were grown at 16, 18, 23, 24 and 28°C under LDs (16-h light/8-h dark), SDs (8-h light/16-h dark) or continuous light, with a light intensity of 80 μmolm-2s-1 produced by cool white fluorescent tubes.
Seeds of the late flowering gi-2 background were mutagenized with ethyl methanesulfonate (EMS) in order to isolate early flowering, suppressors of gi-2 mutants. The gis5 gi-2 mutant was crossed with gi-5 (gi in an Arabidopsis Landsberg erecta accession) to generate the mapping population. About 600 F2 early flowering plants were used for fine mapping by analyzing recombination events using different molecular markers (InDels and dCAPS [26,59]). The gis5 mutation was mapped to chromosome 5 in a 120 kb interval between CER436434 and CER436454 markers. Sequencing revealed a C-to-T point mutation in the At5g63960 locus. For genetic analysis, the gis5 mutant was backcrossed to WT four times. A fragment of 10831bp, containing the full length of At5g63960, was released from the MBM17 BAC clone by using SalI and subcloned to the pPZP212 binary vector [60]. The construct was transformed into Agrobacterium tumefaciens strain GV3101, which was then used to transform gis5 mutants by floral-dipping as described [61]. The transformed seedlings were screened on MS salts plates, containing 50 mg L-1 kanamycin. Only homozygous, single-locus insertion lines were selected in the T3 generation and used for subsequent experiments. The primers used are described in S1 Text.
Protein structures 3IAY, that correspond to the crystal structure of the catalytic subunit of yeast Polδ in ternary complex with a template primer and an incoming nucleotide (closed conformation), 4FVM, that correspond to the catalytic subunit of yeast Polα in ternary complex with the template primer and the incoming nucleotide (closed conformation) and 4FYD, that correspond to the catalytic subunit of yeast Polα alone (open conformation), were used as templates and were obtained from the PDB website (http://www.rcsb.org/pdb/home/home.do).
Three-dimensional model of the GIS5 (wild-type and mutated) proteins were obtained by homology-modeling using Modeller V9.9 [62]. The stereochemical quality of the modelled structure was checked by assessment of the Ramachandran plot plot [63] with the rampage server (http://mordred.bioc.cam.ac.uk/~rapper/rampage.php), being 95.2% of the residues in the favoured regions. The analysis of the compatibility of the atomic model (3D) with its own amino acid sequence (1D) was performed with Verify3D [64]. Finally, the global superimposition between the template (3IAY) and the model has 881 equivalent positions with an rmsd of 0.43, without twists [65]. These parameters support the correctness of the model.
Double mutants were obtained by crossing gis5 mutants with ft-10, tsf-3, co-9, gi-2, fve-3, fca-9, soc1–201 and flc-201 mutants. After F1 selfing, F2 progeny was secreened by phenotyping and genotyping and verification by PCR-based methods. Primers are listed in S1 Text.
Seedlings were frozen in liquid nitrogen and total RNA was prepared using a Plant Total RNA Mini Kit (YRP50; Real Biotech Corporation, http://www.real-biotech.com), and 1 μg was used to synthesize cDNA with M-MLV reverse transcriptase (Invitrogen, http://invitrogen.com), and used to quantitate UBQ10, SEP1-3, FT, FLC, RAD51 and BRCA1 expression with the Mx3005P real-time PCR system (Stratagene, http://www.genomics.agilent.com) in conjunction with SyBR Green I (Invitrogen). UBQ10 was used as a housekeeping gene to normalize gene expression [66]. Relative expression levels were determined using the comparative cycle threshold (Ct) method [67]. The primers used are described in S1 Text.
ATH1 microarrays (Affymetrix) were used to compareWT and gis5 transcriptomes. Total RNA was isolated, as described above, from Col and gis5 10-d-old seedlings grown under continuous light at 24°C. The experimental design comprised three biological replicates of each genotype. Synthesis of cDNA, cRNA labelling and hybridizations were made according to Affymetrix protocols, and probe signal intensities were processed using the Affymetrix GeneChip operating software. The resulting cell intensity (CEL) files were analyzed for data quality control using the same software package. Subsequent normalization of the raw data and estimation of signal intensities were performed using ‘robust multi-chip average’ (RMA) [68] with the CARMAWeb web application (https://carmaweb.genome.tugraz.at). Genes with a P-value lower than 0.05 and fold changes representing log2 ratio≥1 (upregulated) or ≤−1 (down regulated) were considered to be differentially expressed. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [69].
Three 10 cm plates of 10 day-old plants (gis5 and Col) grown on MS agar under continuous light at 18 and 24°C were harvested and immersed in PBS supplemented with 1% formaldehyde. The seedlings were vacuum infiltrated for 20 min. Glycine was added to a final concentration of 0.1 M and incubated for 5 min. The seedlings were removed from the solution and frozen in liquid nitrogen. Approximately 2.0 g of seedlings were ground and resuspended in 25 ml NIB (50 mM HEPES [pH 7.4], 25 mM NaCl, 5 mM MgCl2, 5% sucrose, 30% glycerin, 0.25% Triton X-100, 0.1% beta-mercaptoethanol, 0.1% protease inhibitor cocktail (SIGMA P9599). After centrifugation at 2500 g for 20 min at 4°C, the nuclear pellet was resuspended and washed in NWB (17 mM HEPES [pH 7.4], 7 mM MgCl2, 33 mM NaCl, 13% sucrose, 13% glycerin, 0.25% Triton X-100, 0.1% beta-mercaptoethanol, 0.1% protease inhibitor cocktail). After centrifuging, the pellet was resuspended in 1mL TE buffer supplemented with 0.5% SDS and mixed on a rotator for 20 min at 4°C. The chromatin was diluted with TE buffer to a final SDS concentration of 0.25%. The DNA was sheared by sonication to approximately 500–1000 bp fragments. After centrifugation (10 min at 13,000 rpm, 4°C), approximately one tenth (4–6 μg of DNA) was mixed with RIPA dilution buffer (80 mM Tris-Hcl [pH 7.4], 230 mM NaCl, 1.7% NP40, 0.17% deoxycholate) in a 2:3 ratio and 1mM DTT, 0.5 μg/ml RNAse A, 0.2% proteinase inhibitor cocktail and 1,5 μl anti-H3K4me3 (Millipore, catalogue number: 07–473), anti-H3K27me3 (Millipore, catalogue number: 07–449) or anti-H3K9K14ac (Millipore, catalogue number: 06–599) were added. After overnight incubation with rotation at 4°C, the samples were cleared by centrifugation (14000rpm, 10 min, 4°C). A 30 μl aliquot of washed ProteinA-coupled agarose beads was added to the supernatant and the incubation continued on the rotating wheel for 1 hr at 4°C. The agarose beads were then washed with 5 times 1 mL of RIPA buffer (20 mM Tris-Hcl [pH 7.4], 140 mM NaCl, 1.0% NP40, 0.1% deoxycholate, 0.1% SDS). The immunocomplexes were eluted from the beads with two times 200 μl of glycine elution buffer (100 mM glycine, 500 mM NaCl, 0.05% Tween 20 [pH2.5]) and the combined elutes neutralized with 100 μl of 1 M Tris-HCl (pH 9.7). Crosslinks were reversed by incubation at 37°C for at least 6 hr in the presence of 60 μg/ml Proteinase K followed by at least 8 hr incubation at 65°C. The DNA was purified by two successive phenol/chloroform/isoamyl alcohol extractions and ethanol precipitation. Pellets were washed with 70% ethanol and resuspended in 100 μl of H2O; 4 μl were used for each q-PRC. All immunoprecipitations were quantified in comparison to an appropriate dilution of the input which was obtained by processing 10% of the supernatant of each NO-antibody precipitation (only beads) in parallel to the immunoprecipitated samples during the decrosslinking and DNA purification procedure. When indicated, data was relativized to a UBQ10 or FUS3 fragment. Each of the immunoprecipitations was performed 5–6 independent times. The primers used are described in S1 Text.
For HR frequency determination, we counted the number of GUS positive spots, each indicating a recombination event. The recombination reporter lines 1406 (direct repeat line) and 1415 (inverted repeat line) [57,58] were crossed with gis5. gis5 1406, gis5 1415, 1406 and 1415 plants were grown on MS salts plates under LD conditions at 18 and 24°C. At bolting, 15 plants for each genotype and condition were dissected, and the first true leaf of 12 individual plants was used for spot number determination. A picture of each first leaf was obtained under microscope and further analyzed with the ImageJ software for spot number determination. For FT tissue expression studies, the P8.1kbFTpro:GUS transgenic line [6] was crossed with gis5, and GUS assays were performed on 10 day-old gis5 P8.1kbFT:GUS and P8.1kbFT:GUS seedlings grown on MS salts plates. For CycB1 expression analysis gis5 was crossed with PCycB1;1-GUS reporter lines and GUS assays were performed as previously described [32].
The constructs directed against SEP1 and SEP3 genes were designed using WMD2 Web Micro RNA designer (http://wmd2.weigelworld.org/cgibin/mirnatools.pl; [70]). Overlapping PCR was used to replace MIR319a precursor by each microRNA and finally subcloned into pBI19 derived binary vectors for plant transformation. amiRNAs expression was driven by 35S (ectopic expression), SUC2 (expression in phloem companion cells) and FD (expression in the meristematic cells of the shoot apex). Transgenic lines were selected on medium supplemented with 50 μg mL-1 kanamycin. The primers used are described in S1 Text.
Sequence data from this article can be found in GenBank/EMBL data libraries under accession numbers: At5g63960 (POLD1), At1g65480 (FT), At4g20370 (TSF), At2g45660 (SOC1), At5g15840 (CO), At1g22770 (GI), At5g10140 (FLC), At4g16280 (FCA) and At2g19520 (FVE). The transcriptome data can be found in GenBank (http://www.ncbi.nlm.nih.gov/geo) under Gene Expression Omnibus accession number GSE58036.
|
10.1371/journal.pgen.1001065 | Genome-Wide Profiling of p63 DNA–Binding Sites Identifies an Element that Regulates Gene Expression during Limb Development in the 7q21 SHFM1 Locus | Heterozygous mutations in p63 are associated with split hand/foot malformations (SHFM), orofacial clefting, and ectodermal abnormalities. Elucidation of the p63 gene network that includes target genes and regulatory elements may reveal new genes for other malformation disorders. We performed genome-wide DNA–binding profiling by chromatin immunoprecipitation (ChIP), followed by deep sequencing (ChIP–seq) in primary human keratinocytes, and identified potential target genes and regulatory elements controlled by p63. We show that p63 binds to an enhancer element in the SHFM1 locus on chromosome 7q and that this element controls expression of DLX6 and possibly DLX5, both of which are important for limb development. A unique micro-deletion including this enhancer element, but not the DLX5/DLX6 genes, was identified in a patient with SHFM. Our study strongly indicates disruption of a non-coding cis-regulatory element located more than 250 kb from the DLX5/DLX6 genes as a novel disease mechanism in SHFM1. These data provide a proof-of-concept that the catalogue of p63 binding sites identified in this study may be of relevance to the studies of SHFM and other congenital malformations that resemble the p63-associated phenotypes.
| Mammalian embryonic development requires precise control of gene expression in the right place at the right time. One level of control of gene expression is through cis-regulatory elements controlled by transcription factors. Deregulation of gene expression by mutations in such cis-regulatory elements has been described in developmental disorders. Heterozygous mutations in the transcription factor p63 are found in patients with limb malformations, cleft lip/palate, and defects in skin and other epidermal appendages, through disruption of normal ectodermal development during embryogenesis. We reasoned that the identification of target genes and cis-regulatory elements controlled by p63 would provide candidate genes for defects arising from abnormally regulated ectodermal development. To test our hypothesis, we carried out a genome-wide binding site analysis and identified a large number of target genes and regulatory elements regulated by p63. We further showed that one of these regulatory elements controls expression of DLX6 and possibly DLX5 in the apical ectodermal ridge in the developing limbs. Loss of this element through a micro-deletion was associated with split hand foot malformation (SHFM1). The list of p63 binding sites provides a resource for the identification of mutations that cause ectodermal dysplasias and malformations in humans.
| The p63 protein encoded by the TP63 gene is a transcription factor of the p53 family and functions as a master regulator of ectodermal development. The key function of p63 during ectodermal development is underscored by phenotypic features in p63 knockout mice [1], [2] and in p63 knock-down zebrafish [3], [4]. The developmental abnormalities in animal models are reminiscent of those in p63-associated human disorders. Heterozygous mutations in p63 give rise to at least seven dominantly inherited clinical conditions with three major characteristics, ectrodactyly (also known as split hand/foot malformation, SHFM), orofacial clefting and ectodermal dysplasia with defects in skin, hair, teeth, nails and exocrine glands [5], [6]. There is a clear genotype-phenotype correlation in p63-associated disorders [7]. The most prominent of these disorders is the Ectrodactyly Ectodermal dysplasia and Cleft lip/palate syndrome (EEC, OMIM 604292) which combines all of the three phenotypic hallmarks and is almost invariably caused by missense mutations in the DNA binding domain of p63. Ankyloblepharon Ectodermal defects Cleft lip/palate syndrome (AEC, OMIM 106260) is caused by mutations in the SAM domain of the p63 that is involved in protein interaction. Nonetheless, mutations of p63 can explain only a minority of patients with only one of the three cardinal features, such as in patients with isolated SHFM (∼10%) and in patients with isolated cleft lip/palate (∼0.1%) [7]. There remains a large group of ectodermal dysplasia syndromes with phenotypes that resemble p63-associated syndromes [8]. The genetic basis of many of these clinically related conditions, referred to as the p63 phenotype network, is presently unknown.
There is ample evidence that diseases clustering within such phenotype networks are caused by mutations in functionally related genes that constitute a gene network [9]–[11]. Elucidation of functional interactions among genes within the p63 gene network, their encoded proteins and regulatory elements which control expression of these genes will therefore provide new candidate genes for genetic disorders from the p63 phenotype network. Identifying target genes and cis-regulatory elements controlled by p63 is an important step in dissecting the p63 gene network. Previous studies have focused on transcriptional target genes of p63 identified through individual candidate gene approaches [12]–[14] or through genome-wide approaches [15]–[19]. However, the role of regulatory elements controlled by p63 in transcription has not yet been addressed so far.
Split hand/split foot malformation (SHFM, OMIM 183600) is characterized by a deficiency of the central rays of the hands and feet, resulting in missing or malformed digits. SHFM may be isolated (non-syndromic) or be associated with other developmental anomalies (syndromic). Six distinct chromosomal loci for non-syndromic SHFM have been reported. Specific gene mutations have been identified in SHFM6 and SHFM4. SHFM6 (OMIM 225300, chromosome 12q13) is caused by a homozygous WNT10B mutation and it is the only autosomal recessive form of SHFM [20]. SHFM4 (OMIM 605289, chromosome 3q27) is caused by p63 mutations [21]. Chromosomal aberrations underlie three other types of isolated SHFM: 7q21 deletions and re-arrangements in SHFM1 (OMIM 183600) [22], 10q24 duplications encompassing the Dactylin gene (FBXW4) in SHFM3 (OMIM 600095) [23], and 2q31 deletions encompassing the HOXD gene cluster in SHFM5 (OMIM 606708) [24], [25]. In addition, linkage analysis has mapped SHFM2 (OMIM 313350) to chromosome Xq26 [26].
The SHFM1 locus on chromosome 7q21 has been delineated by various translocations, inversions, deletions and duplications [27]. The smallest region of overlapping deletions in SHFM1 patients [28] encompasses several genes: DYNC1I1, SLC25A13, DSS1, DLX5 and DLX6, of which only DLX5 and DLX6 have been shown to clearly play a role in early limb development. Dlx5 and Dlx6 are highly expressed in the apical ectodermal ridge (AER) of the developing limbs of mice [29]–[31] and in the fins of zebrafish [3], [4]. The AER is critical for limb outgrowth and patterning [32] and there is strong evidence that a failure to maintain the AER signaling is the main pathogenic mechanism in ectrodactyly [33]. The importance of the DLX5/DLX6 genes in limb development has been highlighted in mouse models. Dlx5 deficient mice do not show any limb defects [30]. However, an SHFM-like phenotype has been observed when both Dlx5 and Dlx6 were simultaneously deleted (Dlx5/Dlx6−/−). The limb developmental phenotype in Dlx5/Dlx6−/− mice could be fully rescued by overexpression of Dlx5 in the AER [29], [34]. These observations suggest that the DLX5 and DLX6 genes cooperate in limb development by controlling a common developmental program. DLX5 and DLX6 are further expressed in the craniofacial prominence, the otic vesicle and in the brain [29]–[31], which correlates well with the hearing loss and mental retardation that are present in 30% of the SHFM1 patients [27]. While DLX5 and DLX6 are obvious candidate genes for SHFM1, mutations have not been found in either of the two genes.
Here, we used a genome-wide DNA-binding profiling approach using Chromatin Immunoprecipitation (ChIP) followed by deep sequencing (ChIP-seq) in human primary keratinocytes to generate a catalogue of highly informative target genes and regulatory elements controlled by p63. One cis-regulatory element identified by DNA-binding profiling is located in the SHFM1 critical region and acts as an enhancer element for gene expression mediated by p63 during embryonic limb development. Our data indicate that loss of this element leads to SHFM1. This example illustrates that our catalogue of p63 binding sites can identify candidate genes and loci for the elucidation of disorders from the p63 phenotype network.
The most common isoform of p63, ΔNp63α, is highly expressed in the basal layer of the epidermis that consists mainly of keratinocytes. We therefore established human primary keratinocyte cultures (HKCs) from adult skin as our model system to elucidate p63 gene networks under physiological conditions. To identify target genes and regulatory elements controlled by p63, high-resolution global binding profiles of p63 were obtained from HKC cell lines established from two unrelated control individuals (wt1 and wt2) by ChIP-seq analysis using two antibodies recognizing different epitopes in p63 (4A4 and H129). Analysis of the sequenced reads using the peak recognition algorithm of Model-based Analysis of ChIP-Seq (MACS) [35] gave a highly significant overlap of 11,369 peaks from three profiles (P<1E-300) (Figure 1A). Overlapping peaks were therefore considered as a collection of high-fidelity p63 binding sites in HKCs. Indeed, a set of 17 representative binding sites of various peak heights, conservation scores and consensus motif scores (see below) were tested with independent ChIP followed by qPCR analysis (ChIP-qPCR) with two antibodies (4A4 and H129) and all of them could be validated (Table S1, Figure S1). This confirmed that the obtained p63 binding profile is highly reliable.
To determine the specific p63-binding sequences in the detected binding sites, a de novo consensus motif prediction pipeline was applied to generate a Position Weight Matrix (PWM) (see Materials and Methods for details). A highly significant consensus sequence was identified that is similar to the previously reported p63 and p53 consensus motifs (P<1E-250) [17], [18], [36], [37] (Figure 1B, Table S2). An additional significant AP1-like motif that can be bound by c-Fos and c-Jun proteins [38] was also identified in the detected binding sites (P<1E-50, Table S2). We combined our previously developed p53scan algorithm [37] with the newly identified p63 PWM, hereafter referred to as p63scan. Comparison of p63scan with the previously described motif algorithms p63MH [36] and p53scan [37] showed that p63scan had clearly higher sensitivity for motif recognition without compromising the specificity (Figure 1C). A slight increase of motif scores correlated with an increase of peak heights of the binding sites (Figure S2), suggesting a stronger binding of p63 to binding sites with higher motif scores. Using p63scan, 10,702 out of a total 11,369 p63 binding sites (94%) were found to contain at least one p63 motif (motif score 4.74, False Discovery Rate, FDR 10%). The high percentage of motif-containing binding sites indicates that most binding sites identified in this study participate in direct binding of p63. The de novo consensus motif prediction pipeline was applied to the subgroup of binding sites without p63 binding motifs to search for consensus motifs of other transcription factors or novel binding motifs of p63. A degenerate p63 binding motif was identified, and interestingly, the AP1 motif was also found more strongly enriched as compared to the motif analysis of all binding sites (Table S2). An alternative approach also was taken to examine all known motifs of transcription factors in the TRANSFAC Professional database (version 2009.3) [39] for their significant over-representation (P<1E-10, after Bonferroni correction) in the p63 motif-less binding sites relative to the p63 motif-containing sites. Consistent with the de novo consensus motif search, the AP1 motif as well as the BACH1 and BACH2 motifs that are similar to AP1 was found (Table S3). These data suggest that p63 can bind to DNA by collaborating with other transcription factors such as c-Jun or c-Fos. Interestingly, a previous report showed that p63 binds to an AP1 responsive element to regulate Keratin 1 in keratinocytes in a c-Jun dependent manner [40]. No other novel consensus binding motifs were detected.
Out of 11,369 binding sites, 1460 lie between 5 kb upstream of the transcription start site (TSS) and the end of the first intron of genes, referred to as TSS flanking regions, and 1908 binding sites are located within 25 kb distance to a gene (<25kb region) (Figure 1D). Statistical analysis showed that binding sites at these two chromosomal regions are enriched compared to genomic distributions of all binding sites (P<0.001). Genomic distribution of binding sites with or without a p63 binding motif was similar to that of all binding sites (Figure S3). In total, 10,895 genes had one or more p63 binding sites within 25 kb up- and down-stream of the gene, and they were considered as potential target genes (PTGs) of p63. GO annotation of these 10,895 PTGs using DAVID Bioinformatic Resources 6.7 (NIAID, NIH) [41] showed a non-random distribution with enrichment in functional categories of biological processes, such as development, adhesion, cell communication and intracellular signaling cascade (Table S4). Binding sites with or without a p63 motif were also mapped to genes as separate subgroups, and 10,438 and 944 genes, respectively, have p63 binding sites within 25 kb of the gene. GO annotation of genes mapped by binding sites with motifs resulted in very similar GO terms as annotation of all PTGs (Table S2). However, 944 genes mapped by binding sites without motifs were seemingly involved in slightly different biological processes (Table S4).
The p63 gene arose from two sequential gene duplications at the root of the vertebrates and has unambiguous orthologues only in that taxon [42]. We therefore assessed the evolutionary conservation of the identified binding sites and the p63 consensus motifs therein in aligned vertebrate genomes (PhastCons). The identified binding sites had higher average PhastCons Conservation Scores (PCCS) and were significantly more conserved than random sequences of the same size (Figure 2A). Moreover, PCCS of motifs identified in the p63 binding sites were also compared to that in the random genomic regions. By p63scan, 10,702 motifs were identified in 11,369 p63 binding sites and they were more conserved than 4,003 motifs identified in 100,000 random genomic regions (Figure 2B). These data support the functionality of the identified p63 binding sites. We did not observe a correlation between PCCS and peak height (Figure S4A) or a clear difference in PCCS of binding sites with and without p63 binding motifs (Figure S4B).
To validate whether the identified binding sites represented target genes and regulatory elements relevant to the p63-associated and other diseases with clinical similarities, the OMIM database was searched for diseases associated with the 10,895 potential target genes in this study. We found 904 OMIM disease entries associated with these genes (Table 1, Table S5), referred to as p63 potential target gene-associated diseases (PTG-associated diseases). To assess the relationship amongst PTG-associated diseases, their clinical features were analysed by text mining (Table S6) and evaluated with a similarity algorithm [43] (Table 1). The potential target genes of p63 do not have strong tendency to associate with diseases (P = 1). However, the feature terms in PTG-associated diseases are similar, as the similarity score of these diseases (0.284) is significantly higher than for a random distribution (0.200) (P<1E-6). This shows that PTGs are associated with diseases that have similar clinical phenotypes. Features associated with p63 syndromes are enriched in the top 10% of overrepresented feature terms for the PTG-associated syndromes (P<1E-28). Many of these terms such as stem cell and epithelium reflect p63 functioning (Table S6). This suggests that identified PTGs tend to cause similar disease phenotypes to p63-associated diseases. We did not observe a significant difference between terms derived from motif-containing binding sites or those from motif-less binding sites (Table S6). The significant similarity of disease features of PTGs suggests that these binding sites are relevant to p63-related developmental disorders.
To assess whether p63 binding sites can function as regulatory elements in the p63-related disease network, we focused on SHFM. From the human malformation disease database POSSUM [44] and the Jackson Laboratory's Mouse Genome Database [45], 20 genes were selected based on their localization in the human SHFM loci (Table S7). In addition, these genes are known either to associate with SHFM in human or to have similar phenotypes in mice. These genes are further referred to as SHFM-associated genes (Table S7). As regulatory elements can function over a large distance but might be blocked by insulator elements that are defined by CTCF binding sites [46], p63 binding sites were searched in broad chromosomal regions containing the SHFM-associated genes (up to 300kb from the genes) provided that no known CTCF binding sites are located between the binding site and the gene. With these criteria, p63 binding sites were identified near 12 SHFM-associated genes (Table 2). We propose that these p63 binding sites are potential regulatory elements that might contribute to SHFM.
In the SHFM1 locus on chromosome 7, several deletions have been identified which invariably contain the DLX5 and DLX6 genes as well as DYNC1I1, SLC25A13 and DSS1 (Figure S5) [28], [47]–[51]. We identified a new patient with non-syndromic SHFM (for clinical phenotype, see Materials and Methods) and a novel microdeletion of chromosome 7q21 by a targeted 385K chromosome 7-specific microarray. Surprisingly, the 880kb chromosomal deletion at 7q21.3 encompassed DSS1, SLC25A13 and part of DYNC1I1 but left the SHFM1 candidate genes DLX5 and DLX6 intact (Figure 3). The deletion was confirmed by genomic qPCR analysis (Figure S6). Compared with the previously reported minimal chromosomal deletion (Figure S5) [28], [47]–[51], the protein-coding genes in the overlapping region are DYNC1I1, SLC25A13 and DSS1 but these are not likely to contribute to the phenotype [48], [52], [53]. We therefore hypothesized that disruption of one or more regulatory elements caused the SHFM1 phenotype. To test this hypothesis, p63 binding sites were searched in the chromosomal region spanning the DLX5/DLX6 genes, taking into account the published CTCF binding sites to define the borders of enhancer activity [46]. Consistent with our hypothesis that DLX5/DLX6 are controlled by long distance regulatory elements, DLX5/DLX6 are located in a broad chromosomal region between two CTCF binding sites (chr7: 95882240–95882467 and chr7: 96495007–96495206) spanning approximately 600kb (green arrows in Figure 3). This region contains nine putative p63 binding sites that were identified by our ChIP-seq analysis. These include three high peaks (SHFM1-BS1, -BS2 and -BS3) and six lower ones (a–f) (Figure 3, Table 3).
To identify the binding sites potentially important for limb development, the average PhastCons conservation score (PCCS) [54] of each of the nine binding sites was examined. We found that SHFM1-BS1 had the highest PCCS (0.456) (Table 3) that belongs to the top-ranking 11.6% of all 11,369 binding sites (Figure 2 and Figure S3). To test the functionality of p63 binding sites, the three high p63 binding peaks, SHFM1-BS1, -BS2 and -BS3, were cloned directly in front of a luciferase reporter that is followed by the SV40 enhancer to test whether they are responsive to p63 transactivation. Transient transfection assays showed that only SHFM1-BS1 was highly responsive to p63 (Figure 4A). Transactivation activity was completely abolished by mutations in the p63 binding motif present in SHFM1-BS1 (Figure 4B, motif shown in Figure 3), indicating that the observed transactivation is p63-specific. Mutations in the DNA-binding domain of p63, R204W, R279H and R304W, that are found in EEC syndrome disrupted transactivation, whereas mutations found in non-syndromic SHFM4, K194E, and in AEC syndrome, L517F, reduced the transactivation activity not more than 2-fold (Figure 4C). Based on the structure of the DNA-binding domain in p53 that is highly homologous to that in p63, lysine 194 (Q165 in p53) is located in the DNA-binding domain but does not have direct contact with DNA [5], [55]. The AEC syndrome mutation L517F is located in the SAM domain of p63. Therefore these mutations are unlikely to have major effect on p63 DNA-binding. To examine the enhancer activity of SHFM1-BS1, -BS2 and -BS3, these elements were cloned in front of the SV40 promoter or endogenous mouse Dlx5 and Dlx6 promoters that drive the luciferase gene, but no clear additional activation upon co-transfection of p63 was observed (Figure 4D and 4E and data not shown). Furthermore, in the absence of the enhancer, we did not detect p63 activation on the Dlx5 promoter (Figure 4E, no BS) that was previously reported [14]. This discrepancy is probably due to different cells used in transient transfection assays. These results indicate that enhancer activity controlling expression of DLX5 and DLX6 genes may not be correctly recapitulated in a cellular system irrelevant to limb development.
To understand gene expression controlled by enhancer elements in embryonic limb development, we tested SHFM1-BS1, -BS2 and -BS3 in a transgenic reporter assay in zebrafish. A specific expression pattern of the GFP reporter controlled by SHFM1-BS1 but not by SHFM1-BS2 and -BS3 (data not shown) was observed in the AER and weakly in the ear and in the forebrain (Figure 5A). Expression of p63 detected by in-situ hybridization was only clearly localized to the AER (Figure 5B). The reporter expression promoted by the SHFM1-BS1 in the AER that directs growth and patterning of limbs and fins correlated perfectly with the expression of p63, Dlx5 and Dlx6 during embryonic fin or limb development (Figure 5C) [14]. To further determine whether gene expression regulated by SHFM1-BS1 depends on p63 in zebrafish, we examined the enhancer activity of SHFM1-BS1 in p63-knockdown embryos injected with a specific p63 morpholino [4]. In p63-morphant embryos at 48 hours post fertilization (hpf), the fin buds were severely reduced (mild) or absent (severe) (Figure 5C), as reported previously [3], [4]. In the mild phenotypes, the expression of GFP induced by the enhancer was strongly reduced, as was the expression of the zdlx5a and zdlx6a genes. No fin defects were observed in embryos injected with a control morpholino (data not shown). Enhancer activity of SHFM1-BS1 was also tested in transgenic reporter assays in mice. Consistent with the zebrafish data, specific expression was observed in the AER in mouse embryos (E9.5 and E15), and the expression was lost when the p63-binding motif was mutated in SHFM1-BS1 (Figure 5D). These data showed that the specific expression in AER is dependent on p63. Taken together, our data obtained from animal models clearly demonstrated that SHFM1-BS1 can function as an enhancer element to control gene expression during embryogenesis and its activity is dependent on p63.
Having shown that gene expression regulated by SHFM1-BS1 correlates with that of Dlx5 and Dlx6 in zebrafish and mice, we tested whether SHFM1-BS1 physically interacts with the Dlx5 and Dlx6 promoters. To do that, we used the Chromosome Conformation Capture technique (3C) [56] that allows detecting the three-dimensional proximity of two chromosomal locations (Figure 5E and 5F and Figure S7). In mouse embryonic limb tissues (E10), the interaction frequencies of SHFM1-BS1 with the promoter of Dlx6 and with the intergenic region between Dlx5 and Dlx6 were clearly higher than with the surrounding regions. This indicates that SHFM1-BS1 indeed strongly interacts with Dlx6. A weaker interaction of SHFM1-BS1 with Dlx6 was also detectable in E15 limbs. In addition, SHFM1-BS1 appeared to interact with the intergenic region between Dlx5 and Dlx6 that contains highly conserved enhancer elements [57], [58]. We did not observe clear interaction of SHFM1-BS1 with the promoter of Dlx5. Taken together, our data show that p63 binding sites identified in HKCs can function as regulatory elements to control gene expression in embryonic limb development. We further conclude that disruption of regulation of DLX5 and DLX6 controlled by p63 likely causes SHFM1.
In this study, we established the DNA-binding profiles of p63 in a physiologically relevant human cellular system to identify target genes and regulatory elements controlled by p63. We show that one of the identified p63 binding sites acts as a cis-regulatory element to control gene expression in the AER that correlates perfectly with the expression pattern of DLX6 and DLX5 during embryonic development. A novel microdeletion that includes this binding site but leaves DLX5 and DLX6 intact leads to SHFM.
With a prevalence of 2–6% in humans, congenital malformations represent a major medical problem [59]. Elucidation of the genetic basis of this heterogeneous group of disorders is important for genetic counseling and for basic research. Although current main stream genetic studies still focus on mutations in the coding regions of genes, disease mechanisms associated with genetic variants in short- or long-range regulatory elements are increasingly recognized. Consistent with regulatory elements being required for correct spatio-temporal expression of developmental genes [60], mutations in non-coding cis-regulatory elements have been reported to cause congenital defects and have emerged as a disease mechanism [61]–[64]. Evolutionary conservation can be a powerful tool in the identification of regulatory elements [65]–[67]. A recent study identified a number of highly-conserved elements surrounding the IRF6 gene which is known to be involved in several types of syndromic and non-syndromic cleft lip/palate [68]. In one of these elements a SNP that affects an AP-2alpha binding site was identified to associate with increased risk of cleft lip. This conserved element was able to drive the expression of a reporter gene during mouse orofacial development. Interestingly, in our ChIP-seq study we identified the very same cis-regulatory element as a strong p63 binding site that functions as an enhancer element to control expression of IRF6 [69]. However, as only ultra-conserved elements are the focus of the evolutionary conservation approach, not all important regulatory elements can be identified. For example, conservation analysis in vertebrates of the enhancer element SHFM1-BS1 in our study was not found as an ultra-conserved element (Figure S8 and data not shown), even though it is well conserved. In addition, the identity of the transcription factors controlling regulatory elements may not always be derived from the genomic sequences. Our functional strategy of genome-wide p63 binding profiling does not depend on motif prediction or evolutionary conservation and reveals a large number of potential cis-regulatory elements controlled by p63.
We used human keratinocytes for our studies as recent work on transcription factor p53 revealed that responsive elements are not always conserved across species [70]. Moreover, primary HKC cell lines represent a cell type that is highly significant for p63-associated disorders. As many as 43% of the binding sites (2510 out of 5807) from a p63 binding dataset using the ChIP-on-chip technique in a cervical carcinoma cell line [17] were also present in our ChIP-seq dataset. Given that different cell types and techniques were used in these studies, the overlap of these two datasets is remarkable. Nevertheless, our data from HKCs are highly reliable (Figure S1, Table S1) and appear to represent functional p63 binding sites more accurately [17].
Similar to recent reports on DNA-binding profiles of other gene-specific transcription factors [71]–[74], the number of identified p63 binding sites is large, which was not predicted by the classical paradigm of gene transactivation. Our extensive bioinformatic analyses suggest that the majority of the identified p63 binding sites are biologically functional, as 94% of the binding sites contain a p63 consensus motif, and evolutionary conservation (Figure 2) and phenotypic similarity of PTG-associated diseases (Table 2) are significantly higher than random expectation. The binding sites are frequently located in intronic regions or at a distance from promoters. Thus gene-specific transcription factors may not only activate transcription at proximal promoters but also regulate gene expression at a distance perhaps by looping mechanisms. A recent report on the chromatin interaction map of the Oestrogen-Receptor-α (ERα) [75] also found long-range interaction of ERα binding sites and their target genes. This proposed looping mechanism is consistent with the notion that SHFM1-BS1 physically interacts with the DLX5/DLX6 genes that are located more than 250kb downstream from SHFM1-BS1 (Figure 5E and 5F). Furthermore, binding sites identified in a certain cell type may also represent target genes and regulatory elements that can be regulated at different developmental stages in other cells and tissues. For example, SHFM1-BS1 was identified in human adult skin keratinocytes where DLX5/DLX6 are moderately expressed and their expression is not altered in EEC patient keratinocytes (our unpublished data). Nevertheless, SHFM1-BS1 can drive gene expression in the AER during early embryonic limb development.
It has been well established that p63 plays an important role in limb development, as mutations in p63 give rise to limb defects in complex syndromes as well as to isolated SHFM (SHFM4) [21]. In this report, our data strongly indicate that p63 plays a role in SHFM1 by regulating DLX5/DLX6 through SHFM1-BS1 that physically interacts with the Dlx6 promoter in the AER (Figure 5F). DLX5 and DLX6 were previously reported as target genes of p63 as p63 binds to Dlx5/Dlx6 promoters and activates these genes [14]. However, we did not detect p63 binding sites at the promoter regions of these two genes in our HKCs (Figure 3). It is plausible that looping of SHFM1-BS1 to the promoters may result in a binding signal in a ChIP experiment. We also did not observe p63 activation on the Dlx5/Dlx6 promoters in transient transfection assays (Figure 4E and data not shown). In addition, the SHFM4 mutations only affected transactivation mediated by SHFM1-BS1 moderately in our transfection assay using Saos2 cells which do not express any endogenous p53 and p63 (Figure 4C). The disruption of activation on Dlx5/Dlx6 promoter was previously reported in transfection assays using U2OS cells where endogenous wild type p53 is expressed [14]. The use of different cells in transfection assays may be responsible for the variable results of transactivation assays. Moreover, our observations suggest that SHFM4 mutations that do not directly affect DNA-binding might disrupt protein-protein interaction or DNA looping to the Dlx5/Dlx6 promoters to abolish transactivation. Importantly, we showed that the enhancer element SHFM1-BS1 activates gene expression in the AER during embryogenesis and that this activation is p63 dependent.
In addition to the functional data in model systems, we provide genetic data that support an important role for the enhancer element SHFM1-BS1 in limb development by the identification of a novel microdeletion 7q21 in an SHFM patient. This is a unique microdeletion, as the reported deletions in the SHFM1 patients so far all contain SHFM1-BS1 and DLX5/DLX6 (Figure S5) [22], [28], [47]–[51], [76]–[81]. Within the novel deletion, DYNC1I1 and SLC25A13 are unlikely to play a role in limb development [52], [53]. The other gene within the minimal deletion is DSS1. DSS1 is expressed in the mesenchyme of the developing mouse limb [48], and the causative role of DSS1 in SHFM1 has not been demonstrated. Moreover, the expression of DSS1 in limb bud mesenchyme remains normal in DLX5/DLX6−/− mice displaying typical SHFM phenotypes [34]. Our functional analyses support the notion that the enhancer element SHFM1-BS1 regulates expression of DLX6 and possibly DLX5, and that loss of this gene regulation gives rise to SHFM1. This model is in agreement with recent reports on genomic aberrations in 7q21 that were associated with SHFM1. In one report, a human breakpoint located at 38 kb telomeric to DSS1 and at 258 kb centromeric to DLX6 is associated with SHFM and hearing loss phenotype (Figure S5) [82]. This breakpoint leaves the SHFM1-BS1 association with DSS1 intact, but disconnects it from DLX5 and DLX6. Interestingly in this translocation, the chromosomal context between the p63 binding sites SHFM1-BS2 and -BS3 with DLX5/DLX6 is not affected, which suggests that SHFM1-BS2 and -BS3 do not play a role in SHFM1. Another report identified a familial paracentric inversion-deletion 7q21 that affected a potential enhancer element (Figure S5) [83]. However, the spatio-temporal expression mediated by the identified element in this report did not support a role in limb development. Therefore, it is more likely that the SHFM phenotype in this family is due to the dissociation of the DLX5/DLX6 genes from SHFM1-BS1 by the inversion. It should be noted that in the same report, a 5,115 bp deletion (chr7:96,402,577–96,407,691, hg18) was identified at the breakpoint. We did not observe a p63 binding site in this deletion (Figure 5S). Our results and those from others thus support the hypothesis that SHFM1-BS1 plays an essential role in the regulation of DLX5/DLX6. A genetic approach to delete SHFM1-BS1 in mice can give an unambiguous demonstration of its role to control expression of Dlx5 and Dlx6. Intriguingly, whereas Dlx5/Dlx6 are expressed in the craniofacial region at later stages of development (E14–17 in mice) [29], [34], absence of specific expression controlled by SHFM1-BS1 in craniofacial regions indicates that SHFM1-BS1 is not a regulatory element for orofacial development (data not shown). Different enhancer elements may regulate DLX5 and DLX6 in these tissues. It will be of interest to test other less conserved p63 binding sites within the CTCF boundaries for a role in craniofacial development.
In summary, we have identified binding sites of p63 and taken the first step to build a gene network regulated by p63 with ChIP-seq analysis in human primary keratinocytes. Our study provides potential target genes as well as high-resolution regulatory elements relevant to p63-related diseases. Reporter assays in a large scale to test p63 binding sites in the animal models will provide valuable information on functions of p63 target genes in ectodermal development. Our findings strongly indicate that loss of the regulatory element SHFM1-BS1 identified by a p63 binding site constitutes a novel disease mechanism responsible for SHFM1. Identified target genes and regulatory elements of p63 can therefore be analysed for mutations and microdeletions to understand the disease mechanisms of unresolved diseases that resemble p63-associated syndromes.
All procedures regarding establishing human primary keratinocytes were approved by the ethical committee of the Radboud University Nijmegen Medical Centre (“Commissie Mensgebonden Onderzoek Arnhem-Nijmegen”). Informed consent was obtained. All animal work has been conducted according to relevant national and international guidelines.
The patient was born with bilateral foot anomalies and had no other dysmorphic features, in particular no hand anomalies, evidence of ectodermal dysplasia, scalp defects, oral cleft, bifid uvula, tear duct anomalies, eyelid adhesions or abnormal nails. On review at age 2 years and 7 months of age, she was healthy and was well grown and development was within normal limits.
Skin biopsies were taken from the trunk of healthy volunteers to set up the primary keratinocyte culture [84]. Keratinocyte cultures in Keratinocyte Growth Medium (KGM) under undifferentiated condition were previously described [85].
Human primary keratinocytes under proliferating condition where p63 is expressed at the highest level were used for ChIP and ChIP-seq analysis. Cells were crosslinked with 1% formaldehyde for 10 minutes and chromatin was collected as described [86]. Chromatin was sonicated using a Bioruptor sonicator (Diagenode) for 2 times of 8 minutes at high power, 30s ON, 30s OFF. p63 antibodies 4A4 (Abcam) and H129 (Santa Cruz) were used in ChIP-qPCR and ChIP-seq analyses. ChIP experiments were performed as previously described [37]. ChIP-seq analysis was performed on a Solexa Genome Analyzer (Illumina) as described previously [71].
All 32-bp sequence reads were uniquely mapped to the human genome NCBI build 36.1 (hg18) with zero or one mismatch using ELAND (Illumina), resulting in 3.2, 6 and 20 million unique reads for the three analyzed samples, wt1 with 4A4 ChIP, wt2 with 4A4 ChIP and wt2 with H129 ChIP, respectively. Peak recognition was performed using MACS [35] with default settings and a P value threshold of 1E-9, giving 18,133, 14,963 and 29,166 peaks in ChIP-seq tracks of wt1 with 4A4 ChIP, wt2 with 4A4 ChIP and wt2 with H129 ChIP, respectively. Peaks were mapped to RefSeq genes, downloaded from the UCSC Genome Browser (hg18), to determine genomic location. The ChIP-seq data and associated peaks have been deposited in NCBI's Gene Expression Omnibus [87] and are accessible through GEO Series accession number GSE17611 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE17611).
To determine the p63 motif, a de novo motif prediction pipeline combining three motif prediction tools, MotifSampler [88], Weeder [89] and MDmodule [90], was run on 2273 (20%) randomly selected 200-bp peak sequences (centered at the peak summit as reported by MACS) and PWMs were generated. We used the ‘large’ analysis setting for Weeder. MDmodule and MotifSampler were each used to predict 10 motifs for each of the widths between 6 and 20. The significance of the predicted motifs was determined by scanning the remaining 80% of the peak sequences and two different backgrounds: a set of random genomic sequences with a similar genomic distribution as the peak sequences and a set of random sequences generated according to a 1st order Markov model, matching the dinucleotide frequency of the peak sequences. P values were calculated using the hypergeometric distribution with the Benjamin-Hochberg multiple testing correction. All motifs with a P value<0.001 and an absolute enrichment of at least >1.5-fold compared to both backgrounds were determined as significant. We calculated the ROC AUC for all significant motifs and chose the best performing motif based on the ROC AUC (See Table S2 for the results). The PWM of this motif was combined with the p53scan algorithm to generate p63scan, using an optimal threshold, determined by the maximum f-measure as described previously [37]. The p63scan algorithm can be downloaded from http://www.ncmls.eu/bioinfo/p63scan/. To examine the correlation of motif score and peak height, all peaks were divided in quartiles according to peak height (the number of reads per peak). For each quartile the distribution of the motifs score as determined by p63scan is depicted as a boxplot.
To detect putative transcription factor motifs reported in the TRANSFAC Professional database version 2009.3 [39], the MotifScanner program [91] was used. The search was performed on both strands using a 3rd-order Markov model calculated from the human promoter set of the Eukaryotic Promoter Database (EPD) as a background model. The parameter p (a prior probability of finding one instance of the motif in a sequence) was set to a value of 0.5. To identify motifs that are overrepresented in the p63 motif-less binding sites, the binomial test was used. The obtained P values were corrected for multiple testing (631 motifs for which sites were found in the p63-binding regions) using a Bonferroni correction.
Quantitative PCR primers were designed using Primer3 (http://frodo.wi.mit.edu) [92], and qPCR reactions were performed in the 7500 Fast Real Time PCR System apparatus (Applied Biosystems) by using iQ SYBR Green Supermix (Biorad) according to the manufacturer's protocol. For qPCR of ChIP analysis, one primer set was used for each tested binding region (Table S8) and ChIP efficiency of certain binding sites was calculated using percentage of ChIPped DNA against input chromatin.
Text mining-based [43] feature overrepresentation and gene to disease mapping were determined using the Online Mendelian Inheritance in Man (OMIM) disease database [93], [94]. Detailed information can be found in supplementary information. Human diseases associated with SHFM were taken from the Pictures Of Standard Syndromes and Undiagnosed Malformations (POSSUM) database [44], current as of August 2007, and mapped to genes through their OMIM IDs. Mouse SHFM-associated phenotypes and associated genes were taken from the Jackson Laboratory's Mouse Genome Database (http://www.informatics.jax.org/) [45].
To assess the evolutionary conservation of the 11,369 sites bound by p63, the PhastCons [54] conservation track from the UCSC Genome Browser was used to calculate PhastCons Conservation Score (PCCS). Conservation based on 44 vertebrate genomes was chosen because the p63 gene has 1-1 orthologs throughout the vertebrates [42]. The conservation for a region was calculated as the average conservation of each nucleotide therein. To analyse the correlation of PCCS and peak height, all peaks were divided in quartiles according to peak height (the number of reads per peak). For each quartile the distribution of the PhastCons Conservation Scores (PCCS) is depicted as a boxplot.
For detailed detection of chromosome 7 aberration, high resolution NimbleGen HG18 chromosome 7 specific 385K arrays were used (B3738001-00-01; Roche NimbleGen Systems, Madison, Wisconsin, USA). The 385K average probe distance was 365bp. DNA labeling, array hybridization, post-hybridization washes and scanning were performed according to the manufacturer's instructions (Roche NimbleGen). The acquired images were analyzed using NimbleScan V2.4 extraction software (Roche NimbleGen). For each spot on the array, the log2 Cy3/Cy5 ratio (relative intensity of the Cy3 labeled patient DNA vs. the Cy5 labeled male DNA reference pool of 5 healthy male individuals) was calculated using the segMNT algorithm, which also applied an automatic segment detection. A 50× averaging window was generated, resulting in 20kb segments for this array. Breakpoints were determined with SignalMap V1.9 software (Roche NimbleGen) and 20kb averaged log2 ratios were visualized in the UCSC genome browser.
The genomic regions of p63 binding site peaks were amplified by PCR with gateway cloning primers and cloned into a modified ccdB-containing pGL3-Enhancer Vector, or a ccdB-containing pGL3-Promoter Vector, or a ccdB-containing pGL3-Dlx5 Vector. The ccdB-containing pGL3-Dlx5 Vector was generated by amplification of mouse genomic DNA using primers described in Table S8 to obtain the mouse Dlx5 promoter to replace the SV40 promoter with BglII and HindIII sites in the ccdB-containing pGL3-Promoter Vector. Point mutations were introduced into p63-binding motifs of SHFM1-BS1 to generate mutant p63 binding sites, where the essential cytosine and guanine bases were mutated to adenosine. The ΔNp63α wild-type (pcDNA-mM_ΔNp63α) expression plasmid has been described previously [85]. Point mutations were introduced into this plasmid to generate R204W, R279H, R304W, K194E and L517F mutations. Transfection and luciferase assays were described previously [85]. All cloning and mutagenesis primers are described in Table S8.
Human genomic fragments containing the SHFM1-BS1, -BS2 and -BS3 were amplified with primers described in Table S8. The PCR fragments were subcloned in PCR8/GW/TOPO vector and then transferred, through recombination using Gateway technology, to the ZED destination vector for zebrafish transgenesis [95]. This vector contains the Xenopus Cardiac actin promoter driving DsRed as a positive control for transgenesis. To generate the zebrafish transgenic embryos, we used Tol2 transposon/transposase method [96] with minor modifications. Volume of 2–5nl of mixture containing 25ng/ul of transposase mRNA, 20ng/ul of phenol/chloroform purified ZED constructs and 0.05% phenol red was injected in the cell of 1 cell stage embryos. Three or more independent stable transgenic lines were generated for each construct. For the generation of transgenic mice, the genomic fragments with and without point mutations in p63 consensus motif were transferred into a vector containing the human minimal beta-globin promoter, lacZ and a SV40 polyadenylation signal. Constructs were linearized and the vector backbone removed prior to microinjection into the pronucleus of one-cell mouse embryos. F0 embryos of 9.5–13 dpc stages were harvested and stained for lacZ activity.
Once cell stage embryos were injected with 3ng of ΔNp63 MO II (TCCACAGGCTCCAGGATTCTTACCC) as described previously [4]. Injected embryos were raised at 28°C in standard E3 medium and fixed at 48 hours post fertilization in 4% paraformaldehyde overnight at 4°C. In situ hybridizations were carried out as described [97]. As a control, we injected a similar amount of a MO directed against the Xenopus tropicalis olig2 gene that shows no match in the zebrafish genome [98].
Chromosome Conformation Capture (3C) assay was performed as referred in Hagege et al., 2007 [56]. Limbs of E10- and E15-stage mouse embryos were dissected and processed to get single cells preparations. Ten million isolated cells were first fixated with 2% formaldehyde, and then cells were lysed and nuclei were digested with HindIII endonuclease (Roche). After that, DNA was ligated with T4 DNA ligase (Promega) in low concentration conditions to favour intramolecular ligations. A set of locus specific primers close to a HindIII site (Table S8) was designed with Primer3 v. 0.4.0 [92]. These primers were used to make semi-quantitative PCRs to measure the relative enrichment in each ligation product. The primer near to the BS1 enhancer was taken as the fixed primer, and the different interactions were tested using primers close to the promoters of DLX5 and DLX6 genes. For each interaction two negative control primers were designed about 30 kb upstream and downstream the promoter specific primer. PCR products were run in agarose gels and measured using a Typhoon scanner. Product values were related to a control composed of two BACs that encompass our region of interest.
|
10.1371/journal.ppat.1005035 | Characterization of a Prefusion-Specific Antibody That Recognizes a Quaternary, Cleavage-Dependent Epitope on the RSV Fusion Glycoprotein | Prevention efforts for respiratory syncytial virus (RSV) have been advanced due to the recent isolation and characterization of antibodies that specifically recognize the prefusion conformation of the RSV fusion (F) glycoprotein. These potently neutralizing antibodies are in clinical development for passive prophylaxis and have also aided the design of vaccine antigens that display prefusion-specific epitopes. To date, prefusion-specific antibodies have been shown to target two antigenic sites on RSV F, but both of these sites are also present on monomeric forms of F. Here we present a structural and functional characterization of human antibody AM14, which potently neutralized laboratory strains and clinical isolates of RSV from both A and B subtypes. The crystal structure and location of escape mutations revealed that AM14 recognizes a quaternary epitope that spans two protomers and includes a region that undergoes extensive conformational changes in the pre- to postfusion F transition. Binding assays demonstrated that AM14 is unique in its specific recognition of trimeric furin-cleaved prefusion F, which is the mature form of F on infectious virions. These results demonstrate that the prefusion F trimer contains potent neutralizing epitopes not present on monomers and that AM14 should be particularly useful for characterizing the conformational state of RSV F-based vaccine antigens.
| Respiratory syncytial virus (RSV) causes significant morbidity and mortality in children, yet an efficacious vaccine remains unavailable. Antibodies that preferentially recognize the prefusion conformation of the fusion (F) glycoprotein, particularly those that bind antigenic site Ø at the membrane-distal apex, potently neutralize infection and have aided vaccine design. Here we characterize AM14, a potent human antibody, which we show recognizes a novel epitope midway between the membrane-proximal region and the apex of the prefusion F trimer. The epitope is evenly distributed across two protomers, causing AM14 to be uniquely trimer-specific and, surprisingly, cleavage-dependent. These results indicate that the prefusion trimer is antigenically distinct from the monomer. Our findings also demonstrate that epitopes other than site Ø can be the target of extremely potent neutralizing antibodies and thus provide a new target for structure-based vaccine design. Recognition of this novel epitope could make AM14 an ideal candidate for strategies that combine passive prophylaxis with vaccination, since binding of AM14 would not block elicitation of antibodies against site Ø. Due to its unique specificity, AM14 will also be valuable for probing the conformation of RSV F-based vaccine antigens designed to be in the furin-cleaved trimeric prefusion conformation.
| Respiratory syncytial virus (RSV) is a ubiquitous paramyxovirus that infects nearly all children in the U.S. by two years of age [1]. In infants and young children RSV can cause acute lower respiratory tract infections, leading to bronchiolitis and pneumonia. In 2010, RSV was estimated to cause the deaths of more than 200,000 children, accounting for 2.3% of neonatal and 6.7% of infant deaths worldwide [2]. Although infant mortality in the U.S. due to RSV is low, the cost of hospital care for infected infants is estimated to be as high as $750 million per year [3, 4]. Prophylaxis with the humanized monoclonal antibody palivizumab (Synagis) is the only viable intervention for RSV but is limited to use in high-risk infants due to its cost and modest efficacy [5–7]. The development of very potent antibodies or an efficacious vaccine would bring protection to more children and reduce the financial burden of RSV.
Most RSV vaccine candidates contain at least one of the two viral surface glycoproteins: the fusion protein (F) and the attachment protein (G). Of these, only F is absolutely required for infection [8], and it is the target of palivizumab as well as the majority of neutralizing activity in human sera [9–12]. RSV F is a class I fusion protein that is initially synthesized as an inactive precursor. Proteolysis by a furin-like protease at two sites liberates a 27-amino-acid glycopeptide [13–15]. The N- and C-terminal polypeptides, F2 and F1, respectively, are connected by two disulfide bonds to form a single protomer [16–18]. Although three protomers eventually associate to form the mature trimeric F protein, the order and timing of F protein cleavage and trimerization are unknown.
To facilitate virus entry, the mature F protein is triggered to undergo a transition from a metastable prefusion conformation to a stable postfusion conformation, resulting in fusion of the viral and host-cell membranes. Crystal structures of both prefusion and postfusion F have recently been solved, providing molecular insight into this dramatic structural rearrangement [17–19]. Upon triggering, multiple secondary structure elements in the F1 N-terminus assemble into a long alpha helix that extends toward the target cell. This reorganization pulls the hydrophobic fusion peptide from the interior of the trimer and thrusts it into the cellular membrane, resulting in a prehairpin intermediate. The F1 C-terminus then migrates toward the N-terminus, irreversibly forming a six-helix bundle and driving fusion of the two membranes [20]. This rearrangement of F can also occur spontaneously, resulting in an abundance of postfusion F on the viral membrane [21].
RSV-neutralizing antibodies that bind both the pre- and postfusion conformations of F were the first to be isolated. These include antibodies 101F and the murine precursor of palivizumab, antibody 1129 [9, 10, 18, 22]. More recently, extremely potent antibodies that specifically target prefusion F have been characterized [19, 23, 24]. Three such antibodies—D25, AM22 and 5C4—all bind to the apex of the prefusion trimer at antigenic site Ø, which is dramatically reorganized during fusion [19]. Antibody MPE8, which cross-neutralizes four pneumoviruses, competes with palivizumab, yet preferentially recognizes the prefusion conformation [25]. In addition to their potential for passive prophylaxis, these antibodies were critical to the design of vaccine antigens stabilized in the prefusion conformation, which were shown to elicit much higher neutralizing activity than postfusion F in mice and rhesus macaques [26].
Although site Ø antibodies were originally thought to recognize quaternary epitopes on the prefusion trimer, they have subsequently been shown to react with monomers [27]. Here we characterize AM14, a potent RSV neutralizing antibody that was previously isolated from human PBMCs [23], and show that it recognizes a novel quaternary epitope on the native trimeric prefusion conformation of RSV F.
Microneutralization assays were performed to test the ability of AM14 to neutralize infection of HEp-2 cells by various RSV strains. AM14 potently neutralized all RSV strains tested, with IC50s of 13.6 ng/ml for strain A Long, 12.4 ng/ml for strain A2, 30.8 ng/ml for subtype B strain 18537 and 4.6 ng/ml for subtype B strain 9320 (Fig 1A). For comparison, palivizumab neutralized these strains with IC50s of 300 ng/ml, 320 ng/ml, 380 ng/ml and 120 ng/ml, respectively (S1A Fig). Clinical RSV isolates were also tested in the microneutralization assay (Fig 1A). AM14 neutralized subtype A clinical strains with a geometric mean IC50 of 15.1 ng/ml and a range of 4.7–56.9 ng/ml. Neutralization of subtype B clinical strains was similar, with a geometric mean IC50 of 11.3 ng/ml and a range of 1.5–89.2 ng/ml. The results obtained in this HEp-2 cell-based assay were similar to those previously reported for neutralization of strain A2 in a Vero cell-based assay (IC50 values of 2.1 ng/ml for AM14 and D25, and 209 ng/ml for palivizumab) [23, 28]. Similar to other prefusion-specific neutralizing antibodies tested previously, AM14 did not inhibit attachment of RSV to the surface of HEp-2 cells (S1B Fig) [19], suggesting that it prevents entry by blocking a step downstream of attachment. Collectively, these results demonstrate that AM14 is a potent neutralizer of RSV infection, capable of neutralizing both A and B subtypes equally well by a mechanism independent of viral attachment to the cell surface.
Since the neutralization potency of AM14 was similar to that of the prefusion-specific antibody D25, we hypothesized that AM14 might also exclusively recognize the prefusion conformation. To test this possibility, we performed a Luminex-based binding assay using furin-cleaved RSV F ectodomains stabilized in the prefusion (DS-Cav1) or postfusion (F ΔFP) conformation [18, 26]. In this experiment, AM14 bound tightly to prefusion RSV F derived from A and B subtypes with EC50s of 0.63 nM and 0.18 nM, respectively (Fig 1B). In contrast, no binding to furin-cleaved postfusion F was detected. Motavizumab, which binds equally well to both RSV F conformations [18, 19], recognized all proteins tested in this assay, confirming the immobilization of RSV F proteins to the beads (Fig 1B).
To further characterize the binding of AM14 to prefusion RSV F, surface plasmon resonance experiments were performed (Fig 1C). AM14 Fab bound to immobilized prefusion RSV F with an equilibrium dissociation constant (KD) of 0.18 nM, with rapid association and dissociation rate constants of 1.87 x 107 M-1s-1 and 3.4 x 10-3 s-1, respectively. This is in contrast to D25, which bound prefusion RSV F with association and dissociation rate constants more than 10- and 30-fold slower, respectively (1.35 x 106 M-1s-1 and 9.65 x 10-5 s-1) (S2A Fig). Therefore, although AM14 binds specifically to prefusion RSV F with sub-nanomolar affinity, its kinetics are much faster than those of other potent antibodies such as D25.
To identify the epitope on prefusion RSV F recognized by AM14, the crystal structure of AM14 alone and in complex with prefusion F was determined. Crystals of AM14 Fab in space group P212121 diffracted X-rays to 2.0 Å, and after a molecular replacement solution was obtained, the structure was built and refined to an Rwork/Rfree of 18.6%/22.6% (S3 Fig). To obtain a structure of AM14 Fab bound to RSV F, a matrix of F proteins and Fabs was screened, including ten different protein complexes, eight of which formed crystals. Only four of those diffracted past 8 Å (S1 Table). Prefusion RSV F complexed with both AM14 and motavizumab Fabs formed rod-like crystals in spacegroup P21 that diffracted X-rays to 5.5 Å. The asymmetric unit was composed of one prefusion F trimer, three AM14 Fabs and three motavizumab Fabs (S4 Fig). The structure was refined in Phenix with NCS torsion restraints and reference-model restraints to an Rwork/Rfree of 21.1%/27.7%. Secondary structures and connecting loops fit the electron density well, particularly at the antibody–F protein interfaces (S5 Fig). Additionally, the electrostatic potential of the interface showed substantial charge complementarity, with a positively charged region on RSV F interacting with negatively charged residues in the complementarity-determining regions (CDRs) of the AM14 Fab (S6 Fig).
The orientation of the three AM14 Fabs was largely similar to that observed in negative-stain EM images of prefusion RSV F in complex with AM14, with the exception of the side-view, in which the apex of the trimer is not visible (Fig 2A). This difference is likely due to a combination of the trimer apex protruding from the stain and averaging of slightly different tilts of the complex. The crystal structure shows that AM14 binds at the junction of two protomers within the RSV F trimer (Fig 2B), with approximately 330 Å2 buried on the membrane-distal protomer and 520 Å2 buried on the membrane-proximal protomer. The heavy chain CDRs 1 and 3 make contact with RSV F, as does the light chain CDR 3 (Fig 2C). The AM14 contacts are localized to three regions of the RSV F primary sequence that fold together in the quaternary structure of the prefusion trimer. The first two regions map to the ends of two loops connecting α2 with α3 and β3 with β4, both of which undergo dramatic conformational changes in the pre- to postfusion transition, moving by nearly 100 Å [19]. These two regions together with α4 of antigenic site Ø form the continuous α5 helix of postfusion F, which forms the inner heptad repeat (HRA) of the six-helix bundle, explaining the prefusion-specificity observed for AM14 [19]. The third region of the epitope, located on an adjacent protomer, maps to the loop connecting β17 with β18, which partially overlaps with antigenic site IV and is in a similar conformation in pre- and postfusion structures [18, 19, 22]. Collectively, these data provide a structural basis for the prefusion specificity of AM14, and predict that AM14 is trimer-specific.
Due to the low resolution of the X-ray crystal structure, we sought to verify the AM14 epitope and identify critical interactions by generating monoclonal antibody-resistant mutants (MARMs). After three rounds of selection in HEp-2 cells, four unique AM14-escape viruses were isolated and sequenced. Three of the viruses each contained a single mutation in RSV F (L160S, N183K, or N426D), whereas the fourth virus contained three mutations in RSV F (I79M/R429S/H515N) (Table 1). Resistance of viruses to neutralization by AM14 was confirmed by microneutralization assays (S7 Fig). For one MARM, N426D, a full-length prefusion F variant was generated and expressed on the surface of HEK293 cells. Binding of AM14 IgG to these cells was reduced approximately four-fold compared to cells expressing wild-type prefusion F (Fig 3). Consistent with this result, binding of AM14 IgG to purified prefusion F N426D was reduced by approximately 100-fold in an ELISA (S8 Fig). Binding of antibody 101F to prefusion F N426D was also slightly reduced when measured by ELISA, which is not surprising given that the 101F epitope (residues 427–437) is close to this region [22]. The binding of motavizumab, MPE8 and D25 was not affected by the N426D mutation, consistent with the known locations of their epitopes.
Mapping of the MARMs on the prefusion F structure revealed that L160S and N183K are located on the loops connecting α2 with α3 and β3 with β4, respectively, in agreement with the crystal structure (Fig 2B). This verifies that AM14 makes a substantial interaction with the prefusion-specific region of the membrane-distal protomer. Two other MARMS, N426D and R429S, are located on the loop connecting β17 with β18 in the adjacent protomer, near antigenic site IV. The isolation of these MARMs further supports the low-resolution crystal structure and indicates that the membrane-proximal protomer is a critical component of the AM14 epitope. The final two MARMS, I79M and H515N, were part of the same RSV F variant harboring R429S, and are likely compensatory mutations since they are located more than 30 and 80 Å away from the AM14 epitope, respectively.
The crystal structure and MARMs suggested that AM14 is a quaternary-specific antibody with an epitope spanning two protomers. To test the quaternary specificity, we measured binding of immobilized AM14 to monomeric F and trimeric prefusion F by ELISA (Fig 4). The monomeric F was composed of RSV F residues 1–524 and lacked the foldon trimerization motif. Consistent with the hypothesized quaternary epitope, AM14 binding to monomeric F was reduced by nearly 100-fold compared to the prefusion F trimer (Fig 4A). The residual monomer-binding activity may be due to transient trimerization at high concentrations, which would be indistinguishable from monomeric F binding in this assay.
Since previously described quaternary-specific antibodies against other class I fusion proteins have in some cases shown preference for the mature protease-cleaved fusion protein over the uncleaved protein [29], we sought to determine if cleavage of RSV F was required for AM14 binding. In these assays, AM14 failed to bind prefusion F with mutated furin sites (Fig 4A). In contrast, both D25 and MPE8 (Fig 4B and 4C, respectively) bound to cleaved and uncleaved monomeric and trimeric F proteins with profiles similar to palivizumab, an antibody not expected to have preference for the conformation, cleavage or trimerization of RSV F (Fig 4D). Although D25 was originally described as a quaternary-specific antibody, 90% of its epitope on prefusion F is located on a single protomer, explaining its ability to bind monomeric F (Fig 4B and [19]) and a peptide spanning RSV F residues 153–211 (S2B Fig). The approximately 30-fold decrease in affinity observed for D25 binding to peptide is likely due to the absence of contact residues in the F2 subunit and the neighboring protomer, as well as differences in peptide structure compared to the complete prefusion F. Collectively, these results demonstrate that AM14 is unique in its ability to discriminate cleaved, trimeric, prefusion F from the other forms tested.
Production of soluble prefusion RSV F has thus far relied on the presence of the foldon motif at the C-terminus of F1 to stabilize the weak interprotomer interactions that are normally formed when RSV F is localized in the membrane [19, 26]. Having demonstrated that AM14 is a trimer-specific antibody, we sought to determine if the binding of AM14 was sufficient to stabilize trimeric prefusion F in the absence of a trimerization motif or stabilizing mutations. Since D25 exhibits some degree of interprotomer binding, we also tested this antibody. For this experiment, we first determined the gel filtration elution volumes for purified prefusion, trimeric RSV F (DS-Cav1) incubated with excess Fab. These were then compared to complexes formed by co-expression of each Fab with RSV F ectodomain lacking foldon and stabilizing mutations. Co-expression of AM14 Fab and the F ectodomain resulted in a complex with an elution profile very similar to that of AM14 complexed with DS-Cav1 (Fig 5, compare red and black traces). In contrast, co-expression of D25 and F ectodomain resulted in a complex that eluted later (i.e., smaller) than the corresponding complex of D25 with DS-Cav1 (Fig 5, compare blue and grey traces). Thus, D25 co-expression was not sufficient to stabilize the formation of prefusion F trimer in the absence of a trimerization motif. This property appears to be unique to AM14 due to its epitope being split evenly between two protomers.
AM14 neutralization of RSV is similar in potency to that of the antigenic site Ø antibody D25 [19, 23]. It was originally hypothesized that site Ø-directed antibodies would be the most potent neutralizers due to the location of the epitope on the accessible apex of the prefusion F trimer [19]. Consistent with this hypothesis, the prefusion-specific antibody MPE8, which binds to a lower region on prefusion F, has decreased neutralization potency compared to D25 [25]. In contrast, the potency of AM14 was similar to D25, despite the equatorial binding observed in both the crystal structure and negative stain EM. This demonstrates that antibodies targeting other regions of prefusion F can be as potent as those binding to the apex, which may be an important consideration when optimizing RSV vaccine antigens. In addition, the location of this epitope could make AM14 a candidate for passive prophylaxis. AM14 would have the advantage of high potency and would not block site Ø, leaving this antigenic supersite accessible for inducing protective antibody responses induced by vaccination or infection of the upper airway.
In our ELISAs, prefusion-specific antibodies D25 and MPE8 both bound to uncleaved postfusion F, albeit with low affinity. One possible explanation is that the presence of pep27 in uncleaved postfusion F leads to a more flexible state in which the association of HRA and HRB is weaker and portions of these epitopes that are normally occluded in the cleaved postfusion state are accessible. This hypothesis is supported by the finding that the six-helix bundle-directed antibody 114F bound tighter to cleaved postfusion F (S9 Fig). It is further supported by the reduced SDS-stability of uncleaved compared to cleaved postfusion F trimers. Although uncleaved postfusion F is not biologically relevant, there are a number of subunit vaccines in development that are based upon uncleaved F. Our data suggest that for these proteins in particular, D25 and MPE8 may not be good indicators of prefusion F.
Besides AM14, other antibodies have been identified that recognize higher-order protein structures in both class I and class II viral fusion proteins. The specificity of these antibodies can be grouped into at least three categories. In the first, antibodies contact only one protomer, but recognize a conformation that exists only in the assembled fusion protein, as is the case for the HIV antibody 35O22, which binds to both gp120 and gp41 subunits [30]. This type of antibody has also been proposed for influenza HA antibodies [31]. In the second category, antibodies make direct contact with more than one protomer. This quaternary-specific binding is observed for AM14 and has previously been demonstrated for the influenza antibody HC63 and the HIV antibody PGT151 [29, 32–34]. Additionally, dengue virus antibodies have been identified that recognize both protomers within one dimer of the class II fusion protein E [35]. Interestingly, these antibodies have been reported to shift the monomer–dimer equilibrium, similar to the stabilization of trimeric RSV F that we observed here for AM14 [35]. A third category of higher-order specificity can exist due to the ordered array of glycoproteins on the surface of a virus, as has been described for class II fusion proteins of West Nile virus [36]. Antibody CR4354 binds across two E protein dimers, preventing the rearrangement of E into trimers after exposure to low pH [36]. Thus, the mode of AM14 binding is but one way in which antibodies are able to recognize higher-order structures in class I and II viral fusion proteins.
In addition to quaternary-specific binding, AM14 also exhibited a unique dependence on furin cleavage. RSV F is distinct from the F proteins of other paramyxoviruses due to the presence of two furin sites separated by a glycosylated 27 amino acid spacer, pep27, which is released from the protein after cleavage [15]. There are two likely explanations for the cleavage-dependence of AM14 binding. The first is that pep27 sterically inhibits binding of AM14 to the prefusion trimer. An alternative is that uncleaved F does not adopt the native trimeric state. This would be in contrast to the related paramyxovirus PIV5 F protein, which contains only one cleavage site, since the structures of the cleaved and uncleaved PIV5 prefusion F proteins are nearly identical [37, 38]. Future work on the structure of uncleaved RSV F will be needed to resolve these two possibilities.
The specificity of the potent RSV neutralizing antibody AM14 makes it a useful reagent for probing or isolating the cleaved trimeric state of prefusion F. AM14 may also allow an unparalleled view of prefusion F in its native state, as this antibody can be used to capture trimeric prefusion F without the use of a trimerization motif or stabilizing mutations, as was done for HIV Env using the recently identified trimer-specific antibody PGT151 [29]. AM14 can also be used to help unravel the role of furin cleavage in prefusion F trimerization and to track the order of these events in the secretory pathway, similar to what has been done using antibodies specific for trimeric influenza HA [39, 40]. Further, the high potency and properties of AM14 described here could make it well suited for passive prophylaxis.
Plasmids encoding RSV F prefusion (DS-Cav1) and postfusion (F ΔFP) proteins based on strain A2 [18, 26] were transfected into FreeStyle 293-F cells (Invitrogen). Uncleaved versions of prefusion and postfusion F proteins were produced by changing the basic residues of the two furin-cleavage sites to asparagine residues using site-directed mutagenesis. Proteins were purified from the media using Ni-NTA Superflow resin (Qiagen) and Strep-Tactin resin (IBA). Tags were removed by digestion with thrombin or HRV3C protease, followed by gel filtration using a Superose 6 column (GE Healthcare Biosciences). For crystallization, RSV F proteins were expressed in the presence of kifunensine (5 μM), digested with Endo H (10% w/w), mixed with a 1.5-fold molar excess of Fab, and purified using the Superose 6 column. AviTagged F proteins were biotinylated with biotin ligase BirA (Avidity) and separated from excess biotin by gel filtration with a Superdex 200 column (GE).
Plasmids encoding antibody heavy and light chains were transfected into Expi293 cells (Invitrogen). IgGs and Fabs were purified using Protein A agarose (Fisher) or CaptureSelect IgG-CH1 affinity matrix (Life Technologies), respectively.
Biotinylated proteins were coupled to avidin-coated MagPlex beads at 1 μg per 50,000 beads (Radix). Approximately 1000 beads per well were incubated with 10-fold serial dilutions (1 μM to 0.1 pM) of each antibody in a 384-well plate, washed with PBS plus 0.1% BSA with 0.05% Tween 20 and incubated with 0.33 μg/ml phycoerythrin (PE)-conjugated mouse anti-human IgG Fc secondary antibody (Southern Biotech). Beads were washed and emission at 575 nm was measured using the FLEXMAP 3D flow cytometer (Luminex).
Biotinylated DS-Cav1 was immobilized on an SA sensor chip to a total of 292 response units using a Biacore X100 (GE). A buffer-only sample was injected over the DS-Cav1 and reference flow cells, followed by AM14 Fab 2-fold serially diluted from 5 nM to 19.5 pM in HBS-EP+, with a duplication of the 156 pM concentration. The data were double-reference subtracted and fit to a 1:1 binding model using the Biacore X100 analysis software.
Samples were diluted to approximately 0.03 mg/ml, adsorbed to a freshly glow-discharged carbon-film grid for 15 sec, and stained with 0.7% uranyl formate. Images were collected semi-automatically using SerialEM [41] on a FEI Tecnai T20 with a 2k x 2k Eagle CCD camera at a pixel size of 0.22 nm/px. Particles were picked automatically and reference-free 2D classification was performed in EMAN2 [42].
AM14 Fab crystals were produced by hanging-drop vapor diffusion by mixing 1 μl of AM14 Fab (8.7 mg/ml) with 1 μl of reservoir solution containing 0.1 M sodium acetate pH 4.5, 0.2 M ammonium sulfate and 25% (w/v) PEG 4000. Crystals were soaked in reservoir solution supplemented with 30% (v/v) ethylene glycol and frozen in liquid nitrogen. Data to 2.0 Å were collected at the MacCHESS beamline (Cornell High Energy Synchrotron Source, Cornell University).
The ternary complex was produced by mixing Endo H-treated DS-Cav1 with a 1.5-fold molar excess each of AM14 Fab and motavizumab Fab before separation of the complex from excess Fab by gel filtration. Crystals were produced by hanging-drop vapor diffusion by mixing 0.67 μl of protein (5.6 mg/ml) with 1.33 μl of reservoir solution containing 11.4% (w/v) PEG 8000, 1.7% (v/v) 2-methyl-2,4-pentanediol and 0.1 M imidazole pH 6.5. Many cryopreservation solutions were tested, but diffraction was highest when the crystal was directly plunged into liquid nitrogen after removal of the cold gas layer [43]. Diffraction data were collected to 5.5 Å at the SBC beamline 19-ID (Advanced Photon Source, Argonne National Laboratory).
Diffraction data were processed using the CCP4 software suite: data were indexed and integrated in iMOSFLM [44] and scaled and merged with AIMLESS [45]. A molecular replacement solution for the 2.0 Å AM14 Fab dataset was found by PHASER [46] using the heavy and light chains of PDB ID: 4ERS [47] and PDB ID: 4JHA [19], respectively, as search models. The structure was built manually in COOT [48] and refined using PHENIX [49].
A molecular replacement solution for the 5.5 Å ternary complex was obtained using PHASER with prefusion RSV F (PDB ID: 4JHW [19]), motavizumab Fab (PDB ID: 3IXT [11]) and the 2.0 Å AM14 Fab structures as search models. The asymmetric unit contained the prefusion trimer bound by three motavizumab Fabs and three AM14 Fabs. Rigid-body refinement was then performed in PHENIX, and several of the Fab constant domains were manually placed into the electron density using COOT, followed by another round of rigid-body refinement in PHENIX. Group B-factors and coordinates were refined in PHENIX with NCS torsion restraints and reference-model restraints. The reference models were the 2.4 Å prefusion F structure (PDB ID: 4MMS [26]), the 2.75 Å motavizumab Fab structure (PDB ID: 3IXT [11]), and the 2.0 Å AM14 Fab structure determined here. Data collection and refinement statistics for both structures are presented in S2 Table.
RSV strain Long was incubated at 5 x 106 pfu/ml with 3 μg/ml of AM14 for 1 hr prior to infection of confluent HEp2 cells at an MOI of 1.0. Following a 2–3 hour infection, viral inoculum was removed and medium containing 3 μg/ml of antibody was added and incubated at 37°C for 5–7 days. Virus was harvested from wells containing cytopathic effect during the first round of selection and subjected to an additional 2 rounds of selection at 10 μg/ml. Following each round of selection, RNA was isolated from virally infected cells and analyzed by sequencing to determine F protein sequences. After the third round of selection, viruses were plaque purified and microneutralization assays were performed in HEp-2 cells as previously described [50].
A plasmid encoding RSV F residues 1–574 and harboring the DS-Cav1 stabilizing mutations was created, along with an N426D variant, for expression on the cell surface. Expi293 cells were transfected, harvested 48 hours post-transfection, and washed twice with PBS, followed by incubation with RSV F-specific antibodies (1 μg/ml) for 1 hour at 4°C and Alexa488-conjugated goat anti-human secondary antibody (Invitrogen) (5 μg/ml) for 30 minutes at 4°C. Cells were washed, fixed with 0.5% paraformaldehyde, and evaluated by flow cytometry (LSR II instrument, Becton Dickinson). Data were analyzed using FlowJo software (Tree Star) and GraphPad Prism.
96-well plates were coated with purified monoclonal antibody (AM14, D25, MPE8, palivizumab or 114F) at 6 μg/ml in PBS. Plates were blocked with 2% pig serum in PBS with 0.05% Tween 20 and washed with water. Purified F proteins were serially diluted 4-fold (8 μg/ml to 0.49 ng/ml) and added to plates, which were then washed before incubation with biotinylated anti-HisTag antibody (0.3 μg/ml) (Bio-Rad) and streptavidin-HRP (1:2000) (GE). After addition of o-phenylenediamine dihydrochloride (Sigma), reactions were stopped with 2 N sulfuric acid and absorbance was read at 490 nm.
RSV F ectodomain consisting of residues 1–513 with a C-terminal thrombin cleavage site, 6x His-tag and Strep-tag II was co-expressed with either AM14 or D25 Fab in FreeStyle 293-F cells. Complexes were purified using Ni-NTA and Strep-Tactin resins. Separately, DS-Cav1 was produced as described above and mixed with a 1.5-fold molar excess of either AM14 or D25 Fab prior to analysis. Tags were removed from both complexes by thrombin digestion before gel filtration using a Superose 6 XK 16/70 column (GE).
|
10.1371/journal.pgen.1003443 | Analysis of Rare, Exonic Variation amongst Subjects with Autism Spectrum Disorders and Population Controls | We report on results from whole-exome sequencing (WES) of 1,039 subjects diagnosed with autism spectrum disorders (ASD) and 870 controls selected from the NIMH repository to be of similar ancestry to cases. The WES data came from two centers using different methods to produce sequence and to call variants from it. Therefore, an initial goal was to ensure the distribution of rare variation was similar for data from different centers. This proved straightforward by filtering called variants by fraction of missing data, read depth, and balance of alternative to reference reads. Results were evaluated using seven samples sequenced at both centers and by results from the association study. Next we addressed how the data and/or results from the centers should be combined. Gene-based analyses of association was an obvious choice, but should statistics for association be combined across centers (meta-analysis) or should data be combined and then analyzed (mega-analysis)? Because of the nature of many gene-based tests, we showed by theory and simulations that mega-analysis has better power than meta-analysis. Finally, before analyzing the data for association, we explored the impact of population structure on rare variant analysis in these data. Like other recent studies, we found evidence that population structure can confound case-control studies by the clustering of rare variants in ancestry space; yet, unlike some recent studies, for these data we found that principal component-based analyses were sufficient to control for ancestry and produce test statistics with appropriate distributions. After using a variety of gene-based tests and both meta- and mega-analysis, we found no new risk genes for ASD in this sample. Our results suggest that standard gene-based tests will require much larger samples of cases and controls before being effective for gene discovery, even for a disorder like ASD.
| This study evaluates association of rare variants and autism spectrum disorders (ASD) in case and control samples sequenced by two centers. Before doing association analyses, we studied how to combine information across studies. We first harmonized the whole-exome sequence (WES) data, across centers, in terms of the distribution of rare variation. Key features included filtering called variants by fraction of missing data, read depth, and balance of alternative to reference reads. After filtering, the vast majority of variants calls from seven samples sequenced at both centers matched. We also evaluated whether one should combine summary statistics from data from each center (meta-analysis) or combine data and analyze it together (mega-analysis). For many gene-based tests, we showed that mega-analysis yields more power. After quality control of data from 1,039 ASD cases and 870 controls and a range of analyses, no gene showed exome-wide evidence of significant association. Our results comport with recent results demonstrating that hundreds of genes affect risk for ASD; they suggest that rare risk variants are scattered across these many genes, and thus larger samples will be required to identify those genes.
| Common and rare variants are important constituents of the genetic architecture of Autism Spectrum Disorders (ASD) [1]–[12]. Nonetheless analysis of rare variants has produced the vast majority of findings that implicate certain genes as playing a role in liability for ASD (i.e., ASD genes). Because of the promise of identifying novel ASD genes via rare variants, and the potential downstream implications regarding treatment, an ambitious exome sequencing study has been implemented including nearly 2000 case and control subjects sequenced at two genomic centers. Exome sequencing studies of complex traits have shown success in candidate gene studies [13]–[18]; however, most published candidate gene studies have not reported a p-value small enough to attain exome-wide significance [19].
For rare variants, even if effects are strong, single variant tests typically have little power. Rare variants have to be combined in some way, such as within a gene or across genes, for an association test to reach sufficient power. Hence statistical tests examine the cumulative effects over the observed rare variants in the target set. A number of statistical methods to test for association with rare variants are now available. Several of these tests fall into the category of burden tests in that they assess association with a “super-variant” [20]–[24]. Each of these burden methods assumes variants impact the phenotype in a common direction. Rather than aggregating variants, another class of methods, including C-alpha [25] and SKAT [26], look for an unusual distribution of rare variation among cases and controls.
Power of the test is determined by the number of causal variants in the gene, the size of the corresponding effects, and the sample size. Assuming that the rarest variants are likely to have the largest effects, it is challenging to amass substantial evidence for association without a large sample size. Based on extrapolation of effect sizes and frequencies from published studies [19], the results indicate that thousands of individuals are required to obtain genome wide significance.
In this ARRA autism sequencing consortium (AASC) study, data have been produced by two sequencing centers (Baylor College of Medicine and Broad Institute) and by different exome capture methods, different sequencing platforms and different pre-processing alignment and variant calling methods. Therefore the coverage and quality of these data sets varies. Nonetheless, as we show in the sequel, these data can be harmonized using standard filtering criteria. Given the distinct data sources, the most effective way of testing for association is unclear. Following in the tradition of association studies, meta-analysis is a natural option [27]. With this approach we can perform the analysis on each data set separately and then combine p-values using the weighted Z-score method. Alternatively, after filtering to homogenize data, we can combine the two data sets directly and perform mega-analysis. Meta-analysis has the advantage of permitting and adjusting for heterogeneity between samples [28]. All other things being equal, this is the preferred choice. On the other hand, if the power of mega-analysis is better, then this option is worth pursuing. In this report we show that mega-analysis is the more powerful procedure for gene-based tests, such as SKAT [26], a result that might be counter-intuitive given the well-known efficiency of meta-analysis for tests of linear form such as logistic regression. For these data we also find that population structure appears to be corrected for by using principal components analysis [29].
After quality control and controlling for ancestry, analysis of AASC data reveals no clear-cut associations, including associations in known ASD genes. We conclude that rare variants affecting risk are not clustering in a small number of genes, supporting recent results from de novo single nucleotide and copy number studies showing that hundreds of genes in the genome affect risk for ASD [4]–[6], [8].
The AASC whole-exome sequencing data included 1039 ASD subjects of European ancestry and 870 controls of similar ancestry. Approximately half of the samples were sequenced using the Solid platform and called with AtlasSNP 2 [30] (Baylor: 505 cases, 491 controls) and the remainder were sequenced using the Illumina platform and called with GATK [31] (Broad: 534 cases, 379 controls).
We considered 6 filters to make these data sets more similar in terms of the distribution of variants in the exome. Filters were sequential in their stringency for including a variant: Filter PASS included variants that pass the baseline filter of GATK; Filter MISS excluded any variant with more than 10% missingness; Four additional filters placed increasingly stringent requirements on depth and balance of reference and alternative allele calls (see Methods). If not otherwise stated, results for analyses were based on the least stringent of these: Filter DpBal, which filters by missingness , depth , balance for Broad and for Baylor.
Seven control samples were sequenced by both centers, facilitating an independent comparison of cross platform calls and an evaluation of the filtering process. To do so, we identified all rare (), non-synonymous variants located in at least one of the two data sets. Using Filter PASS, in total, these seven samples had 337,478 calls and only .039% of them were mismatched. With Filter DpBal, 290,426 calls remained and .017% of them were mismatched (Table S1). Of the heterozygotes called by one center, but not the other, the mismatch rate was not symmetric: 9 heterozygotes were called by Baylor, but not by Broad, while 42 heterozygotes were called by Broad, but not by Baylor. On closer inspection, many of these heterozygotes did appear to be present; however, one of the variant callers was not confident enough to make the call. Application of the stricter filters (B–D) led to the removal of many of the heterozygous calls for which the callers matched without further improvement in the mismatch rate. For instance, with Filter D only 65% of the matching heterozygous calls from Filter PASS were preserved compared to 85% for Filter DpBal.
Post filtering, the Broad and Baylor data sets had similar numbers of minor allele calls per sample per gene (Figure 1A). The Baylor variant count was slightly greater than the Broad count (Figure 1B), due in part to the larger number of samples in the Baylor data set. The average count of rare variants per gene was 9.24 for Baylor and 8.82 for Broad. Association analysis was limited to non-synonymous variants that had minor allele frequency (MAF) less than . A total of 156,636 and 152,851 variants were retained in the Baylor and Broad samples, respectively. After filtering 9,738 and 5,808 indels were retained in the Baylor and Broad samples, respectively.
Information from two or more datasets can be combined via meta-analysis with the weighted Z-score approach [32]. In the context of the SKAT test this approach assimilates gene-level information without consideration of the directionality of any single variant effects. Alternatively, if the data are combined after careful filtering and harmonization, it is possible to analyze all data simultaneously using a mega-analysis approach.
For a theoretical comparison of these approaches, see the Methods; here we provide empirical analysis. To compare analytically the power of meta- and mega-analysis we assume two data sets have the same sample size and rare variants at the same locations. Results of this analysis show that, regardless of the number of variants, mega-analysis has greater power than meta-analysis, unless the signal is so strong that both have power close to one (Figure 2).
More realistic power comparisons can be made based on the observed Baylor and Broad variant calls directly in simulation. We focus on the 1090 genes with the largest number of variants to obtain the greatest flexibility for configurations of causal variants. From the combined list of variants, some of which are observed only in Baylor or Broad, but not both, and some of which are shared, we randomly pick a fraction as causal variants. We use causal variants to generate the phenotype based on the model in Eqn. 1 with odds ratio inversely proportional to allele frequency. The fraction of rare variants that are causal varies from 20% to 50%. In the analysis we upweight variants inversely proportional to allele frequency using SKAT's default setting. We also use SKAT to calculate the p-values for Baylor, Broad and the merged data sets based on its standard approximation technique. For this simulation analysis and for all our other data analysis, we combine all singleton variants as a super-variant. For meta-analysis the weighted Z-score method combines the two p-values from Baylor and Broad for each gene. Notice that in this analysis, mega-analysis performs better than meta-analysis under a variety of different distributions of causal variants and different log odds ratios (Figure 3).
To gain intuition into the comparison between meta- and mega-analysis, consider combining information across two dataset of approximately equal size. If, in the combined sample and for a particular variant, we observe all of the rare alleles in cases and none in controls, then the evidence for association is higher than if we combine statistics in which half of the rare alleles are observed in cases from each of two sub-samples. For example, for a variant observed 4 times, twice in cases from both subsamples, the mega and meta p-values are .06 versus .17, respectively. The difference in evidence occurs because there are five ways 4 alleles can be partitioned between cases and controls in the mega dataset (4∶0, 3∶1, 2∶2, 1∶3 and 0∶4); however, there are only three ways that 2 alleles can be partitioned between cases and controls. Thus with a larger sample, it is possible for rare alleles to obtain more unusual configurations. As variants become extremely rare the situation becomes more unfavorable to meta-analysis. Unless the sample is very large, most samples will draw only one copy of the rare allele and in this scenario neither of the two case-control configurations is unusual. With singleton variants SKAT can only gain information about association if the rare variants are grouped to form a super-allele.
Alternatively, mega-analysis also has advantages when considering rare alleles with no effect. If, for a particular variant, we observe half of the rare alleles in cases and half in controls in the combined sample, but all of the alleles are in cases in the first sample and all are in controls in the second sample, then the evidence for association is appropriately diminished by considering the full sample simultaneously (for 6 variants, mega versus meta). If there were only one variant per gene, it would be possible to adjust the meta-analysis to capture the sign of the association and overcome this weaknesses; however, gene-based statistics rely on having multiple variants per gene to gain power. With multiple variants, the power differential in mega versus meta occurs because mega-analysis assimilates information variant by variant, cancelling out false signals that differ in direction of association across data sets and capitalizing on true signals that match in direction. By construction, meta-analysis is restricted to combining information at the gene level post hoc, rather than at the variant level. In total, these comparisons explain why mega-analysis has greater power than meta-analysis for statistical tests such as C-alpha and SKAT, that are based on the distribution of rare variants across cases and controls.
To evaluate how sensitive the test statistic is to linkage disequilibrium typical of rare variants, we select 144 genes that have exactly d = 20 variants in the Broad data set. Using these data we randomly assigned case-control status to generate a null distribution for test statistics. With no linkage disequilibrium structure among rare variants, and appropriately chosen weights, the score test statistics is known to follow a distributions under the null hypothesis. Alternatively, notable dependencies among rare variants result in a statistic that follows a mixture of distributions, with degrees of freedom less than . Results from simulations under the null in the form of a Q-Q plot (Figure 4), show that the independence assumption is a reasonable approximation for these data.
For association analysis of common variants (CVs, MAF) it is common practice to control for ancestry by regressing out the most predictive eigen-vectors for ancestry derived from a representative sample of CVs [29]. To determine if the distribution of rare variants varied in ancestry space similarly to CVs, we plot individuals based on their ancestry coordinates [33] using three sets of single nucleotide variants (SNVs): CVs, low frequency variants (LFVs, ), and both types of variants (CVs+LFVs). The ancestry coordinates are the eigen-vectors obtained by applying principle components analysis to CVs (14,702 CVs used in Baylor and 56,607 CVs used in Broad), LFVs (8783 LFVs used in Baylor and 29,509 LFVs used in Broad) and CVs+LFVs respectively. The variants used for PCA have no missing genotypes. We find that individuals cluster fairly similarly for CVs versus LFVs in eigen-vector 1, but less so for eigen-vector 2; and individuals cluster almost identically for CVs and CVs+LFVs (Figure 5 for Broad and Figure S1 for Baylor; notice that the similarity of clusters observed in CVs is apparent using EVs 1 and 3 for CVs+LFs). In the subsequent data analysis we explore the effect of using eigen-vectors from CVs and LFVs to control for confounding due to population structure.
Cases and controls included in the AASC sample have been chosen to have matching ancestry based on eigen-vectors derived from CVs obtained from GWAS genotyping platforms [10]. Examining the distribution of cases (orange) and controls (blue) from Baylor and Broad plotted versus the top 2 eigen-vectors calculated from CVs in the exome shows that the samples are fairly evenly distributed in ancestry space but many of the subjects on the boundary of the eigenspace are cases (Figure 6). When combining Baylor and Broad samples into a common eigen-space, it is evident that the two samples overlap substantially (Figure S2). The Baylor sample, however, includes greater diversity.
As a first step to investigate the distribution of rare variants, we identify all pairs of individuals who share doubleton variants, i.e., each had one copy of an SNV seen only twice in the entire sample. Doubletons are of interest because they are the rarest variants in our sample for which we have strong confidence in the variant calls. When we tally the total number of doubleton variants possessed by each individual in the Baylor case sample, the distribution of the doubleton-count varies widely, with some individuals having a far greater share of these rare variants than expected due to chance. We examine the distribution of doubletons as a function of the eigen-map. Figure 7 displays the relative count of doubletons in the 2-dimensional eigen-map for the Baylor and Broad samples. Individuals with the largest number of doubletons tend to be clearly separated from the majority of the subjects in ancestry space by the top two eigen-vectors.
To compare the distribution of doubleton counts with the distribution of common variants, for each individual in the Baylor case sample we tally their count of minor alleles (MAC_c) over exonic CVs. From Figure 8A, 8B it is clear that individuals with a large count of doubletons also possess a disproportionate number of minor alleles, suggesting that these individuals are toward the boundary of the European ancestry space. Indeed all of these individuals are separated in eigenspace from the majority of the individuals (Figure 7A, orange points). Furthermore, sample records suggest that many of these individuals are from Portugal, a population whose individuals have a somewhat larger component of African ancestry. The same pattern exists in the Broad case sample (Figure 7B and Figure 8C); however the Broad sample does not include any individuals with very large numbers of doubleton variants.
These findings suggest that the distribution of common variants might function as a proxy for the distribution of rare variants. Next we look to see if these descriptive analyses support the use of an eigen-map to control for confounding in rare variant tests due to ancestry. To test for association between ASD and rare variants in the AASC sample, we apply burden tests and SKAT to the filtered version of the data sets and obtain the p-values of genes in the Baylor, Broad and combined datasets. We investigate the effects of population structure by calculating the genomic control inflation factor [34] when the test is performed with and without including 10 eigen-vectors for ancestry obtained from genotypes of CV [29].
Before comparing choices of eigenvectors, we investigate the behavior of the genomic control statistic, , when calculated based on rare variant test statistics. SKAT has been shown to provide accurate p-values in the tail of the distribution for moderate sized samples [26]. Indeed, for these data, we also find that the nominal p-values appear to be accurate in the tail of the distribution (see below). The distribution of the p-values across the genome, however, does not follow the expected uniform distribution (Figure S3A, S3B). Specifically, for those genes clearly not associated with the phenotype (p-values ) we find that SKAT tends to report p-values biased downward toward .5, causing an apparent, but uninteresting inflation in the GC factor. Notably, the algorithm for computing p-values seems to be accurate for smaller p-values; we do not find a bias in estimate of the first quantile (Figure S3A, S3B). A similar phenomenon holds true for the burden test, but to a much lesser extent (Figure S3C, S3D). This is likely due to the very small counts of rare variants. Using permutations to obtain p-values would remedy the situation, but at a substantial cost in computation.
These insights into the null distribution of the rare variant test statistics lead us to calculate , a variant on the GC principle based on the first quantile (rather than the median) of the p-value distribution. For a properly calibrated statistic has an expected value of 1 when there is no confounding due to population structure (see Text S1). To compare the behavior of these two genomic control factors we conduct the following experiment. We calculate and based on SKAT statistics computed for the 1000 largest genes. Then we permute case and control status 100 times, computing the genomic control factors for each permutation, to obtain the distribution of these statistics (Figure 9 and Figure S4). Notice that the observed value of is close to the mean of the simulated distribution for all 3 choices of eigen-vectors. In contrast shows much greater variability and the mean of the permutation distribution is shifted further above 1, supporting our conjecture that provides a positively biased estimate of the effect of confounding when using the SKAT statistic for samples like this one.
Next we examine the effect of adjusting for ancestry (using CVs) on the rare variant test statistics. Notice that while is inflated for all conditions, is controlled fairly well in the Baylor and Broad samples individually (Table 1); in the mega SKAT analysis there is a slight inflation (1.08). From Table 1 and from the -log10(observed p-values) versus -log10(expected p-values) plot (Figure 10) we see the distribution of the test statistics follows the null hypothesis quite closely. We conclude that adjusting for ancestry using CVs is sufficient to yield a substantial reduction in .
We explore this further by contrasting the results obtained when applying no correction versus correction based on eigen-vectors derived from CVs, LFVs and CVs+LFVs and find that the corrected results are nearly indistinguishable regardless of the scenario (both data sets individually, SKAT or burden test, meta- or mega-analysis; Table S2). For example, in the Broad sample and the SKAT statistic, using no eigenvectors yields compared to , and 1.03, derived using CVs, LFVs and CVs+LFVs, respectively.
As described previously most analyses of the data use Filter DpBal to screen called variants. Because one should always be concerned about the possibility of screening out risk variants by this filtering process, we first examine the number of genes exceeding a threshold (i.e. signals) for 3 filters ranging from lenient (Filter PASS) to stringent (Filter DpBal; Table 2). Applying the test statistic to the individual data sets we find no large excess of signals even for the most lenient filter. However, for mega-analysis, filtering is essential to avoid false positive signals. Consider the number of genes with p-values less than .001; with baseline filtering (PASS) we observe a significant excess of such genes (), but no excess with any other filters (Table 2). Next, considering the number of genes with p-values less than .01 the pattern continues; with baseline filtering (PASS) we observe a highly significant excess of such genes (), but this large excess is absent for Filter DpBal (Table 2). It is quite likely that the slight excess of genes with p-values less than .01 after filtering is due to real, but weak signals in a small set of genes. A candidate diagnostic for filtering is matching of minor allele count per person of rare variants (MAC) across platforms (Table 2). However total MAC is a crude measure of alignment. Diagnostic plots such as Figure 1B give a more insightful comparison across genes and we conjecture that a filter chosen to attain good alignment of MAC across genes is a candidate for successful data harmonization. MAC should also be similar across cases and controls for most genes; for Filter DpBal, MAC per person is 330 and 300 in cases and controls, respectively.
While filtering is beneficial to remove false positives, it has the potential to remove real signals as well. We explore the effect of filtering on a particular gene (SCN2A) that has been demonstrated to be an ASD gene based on 3 recurrent de novo loss of function mutations [4], [5]. In the Baylor sample, with Filter PASS we obtain a suggestive p-value of .009, but many of the observed variants have high missingness, very low depth and poor balance of alleles. With Filter MISS the p-value is .033. Finally, with additional filtering the signal is removed altogether. (Specifically, Filter DpBal removes 2 putative severe missense mutations [35] and 1 putative loss of function variant from cases.) There is no evidence of association in the Broad sample for this gene.
Prior to filtering, a sizable fraction of the loci in which a variant is called for one subject cannot be called – either heterozygous or homozygous – for other subjects; it is current practice to remove loci that have variant calls for some subjects, but of subjects have missing calls. After filtering (Filter DpBal), .3% of the values are missing, but the missingness is not evenly distributed across sites or case/control status (Table 3). Most notably this “missingness rate” in Baylor cases is twice as high as the missingness in Baylor controls and 90% of the missingness arises from the Baylor site. Although differential missingness has the potential to cause false positive associations, differences between cases and controls within each data set are not so high as to induce an excess of false positive associations in meta-analysis even in the unfiltered data; however, if we apply mega-analysis to the unfiltered data, we obtain a significant excess of genes with p-values (; Table 4). This problem is remedied by applying Filter DpBal: after filtering, which removes loci with high rates of missingness, we obtain no excess of small p-values for the SKAT mega-analysis test statistics. When evaluating this issue at a finer scale after filtering by looking at the effect of differential missingness at the gene level, we find no association between the test statistic and differential missingness (Figure S5).
Neither SKAT nor burden gene-based tests produce a test statistic exceeding the threshold for exome-wide significance (). Genes with p-values are reported in Table S3. Note that nearly half of these genes have more rare variants in controls than cases, suggesting a protective effect, but we view this as unlikely. Moreover, the evidence is also not sufficiently compelling to replicate any known ASD gene. To explore this last issue in more detail we compile a list of genes with at least two functional de novo mutations identified in the recent ASD studies [4]–[6], [8] (Table S4), and we examine the 114 ASD genes cited by [36] as ASD genes (Table S5). For all genes in these lists we obtain the p-values of SKAT and the burden tests applied to Broad and Baylor samples separately and jointly by mega-analysis. None of the genes yield compelling signals, arguing strongly that our power is insufficient to detect associations with rare variants without further information to guide our analysis.
Studies of the distribution of de novo copy number and sequence variants in ASD and control subjects invariably find elevated rates of damaging de novo events in ASD subjects [1]–[8]. These studies also invariably find relatively little convergence of de novo events on particular loci in the human genome. These results are consistent with only one conclusion about the genetic architecture of ASD, namely that there are hundreds of genes in the genome that can affect liability, possibly more. Indeed various statistical analyses of the data support this conclusion [5], [8].
Another common theme of ASD studies is that while de novo events are rare, they can successfully identify ASD liability genes, and in general the distribution of rare variation has been a key tool for gene discovery [37]. By contrast common variation has not yet proven an effective tool for discovering replicable ASD genes, although there are tantalizing findings [10].
With these observations in mind the AASC has implemented a study of rare variation in ASD based on WES [38]. Here we report on data from almost 2000 ASD subjects and controls. We find the distribution of rare variation between cases and controls is remarkably similar, showing that ASD risk genes cannot be identified in a case-control sample of this size. Indeed, even known ASD genes showed little association in this study. This finding is in keeping with other studies of rare variants, but with quite different phenotypes, supporting the conjecture that rare variant association studies require large samples [19], [39], [40]. With respect to the genetics of ASD, the results are also consistent with the inference from de novo studies that there must be hundreds genes affecting liability to ASD [3]–[6], [8]. These results underscore the scale of the challenges ahead in our effort to discover ASD genes. Large samples must be amassed and assessed and effective study designs implemented [41].
To gain insight into the limited power of this study, consider three scenarios: (A) the gene has 15 variants, each with MAF, for which all have odds ratio of 4; (B) the gene has 20 variants, each with MAF, for which 10 have odds ratio of 3; and (C) the gene has 40 variants, each with MAF, for which 30 have odds ratio of 2. We list the required samples size of each scenario in Table S6 to achieve a power of 50% and 80% per gene (with a p-value threshold of ). Even though the power of mega-analysis is only 0.31, 0.11 and 0.06 for our study, assuming these scenarios were realistic, power would have been sufficient to discover a fraction of the large number of ASD genes present in the genome. We conclude that these scenarios do not describe likely models for risk genes in ASD.
As with GWAS, to assimilate large samples and gain power, multiple studies must be combined. In the analysis of samples from multiple studies, meta-analysis, based on Z-scores, has become the norm for most genetic investigations. This form of meta-analysis has power equal to mega-analysis for single variant tests [42], hence it is reasonable to assume that meta-analysis is generally superior to mega-analysis because the former more easily accommodates heterogeneity across studies. A notable result from our study is that these results do not carry over to gene-based tests such as SKAT. In that setting mega-analysis has considerably more power than meta-analysis because mega-analysis assesses the concordance of association for a variant across all sites and then combines information across all variants within a gene. In this way, the method separates true signals from false ones and attains a greater signal to noise ratio. In contrast, meta-analysis combines information across studies at the gene level and hence can not assess the pattern of signals at the variant level across sites.
A drawback of mega-analysis is that we encounter challenges when combining datasets collected across multiple studies, which can differ in many respects due to the use of different sequencing platforms and protocols. For instance, these differences lead to differential coverage by exon and different alignment errors. Even the best laboratory process has measurement error and these errors are exacerbated when they differ across batches of samples, particularly if they differ between cases and controls. For these reasons caution must be exercised if one is to reap the benefits of mega-analysis. Indeed, even after careful filtering, heterogeneity between sites could account for the modest inflation in the associate test statistics and the genomic control factor after combining sites via meta- and mega-analysis.
In this study we construct extra filters to ensure that the distribution of rare variation of the WES data is similar for the two centers. We find good results filtering called variants by fraction of missing data, read depth, and balance of alternative to reference reads. Ideally a filter is tuned by measuring some individuals on multiple platforms. We tune our filters using subjects measured twice. If such data are unavailable, however, we find that another promising approach is to compare minor allele counts (based on rare variants) per gene. A good filter is one that aims to equilibrate these quantities.
Even with the most minimal filtering we observe no excess of positive signals for association within the individual data sets, but for mega-analysis we observe a great number of positive associations. These false discoveries are diminished, however, after filtering. Likewise mega-analysis is more susceptible than meta-analysis to the impact of differential missingness across platforms and across case/control status. Indeed, without filtering, mega-analysis has many false discoveries but meta-analysis did not. However, using filtered data we find that mega-analysis is quite robust to differences in missingness rates across platforms and case/control status, although we recognize that this robustness could fail for more extreme heterogeneity of missingness. Still our study has some differences in missingness and yet does not produce detectable false discoveries. From our analyses we conjecture that filtering that removes variants with missingness (per data set) is largely effective.
When combining data sets the effects of population substructure on association is also a concern due to clustering of rare variants in ancestry space [40], [43]. Even though our case-control samples are approximately pair-matched by ancestry in the study design, we find weak evidence of population structure confounding the test of association. In our data these effects could be mitigated by regressing out principal components of ancestry using common variants or low frequency variants. This result supports findings of [44], but is contrary to other predictions [43]. Thus, although rare variants tend to be younger, and therefore distinctly clustered in populations, in our sample estimates of ancestry derived from common variants capture the major features of the distribution of rare variants in ancestry space.
In conclusion we find that WES data on nearly 2000 samples collected for a case-control study are insufficient to discover novel liability genes for ASD, even after applying efficient methods like mega-analysis and controlling for ancestry effectively. These results demonstrate that much larger samples will be required for effective gene discovery and lend further support to the prediction that there are hundreds of genes that impact ASD liability in the human genome.
The AASC whole-exome sequencing data includes 1039 independent subjects diagnosed with autism spectrum disorders (ASD). Subjects were selected to be of European ancestry, based on genetic (eigen-vector) analysis and European origin. Samples were selected from the Autism Genetic Resource Exchange (AGRE, research.agree.org), the Autism Simplex Collection (TASC [45]), National Database for Autism Research (NDAR, ndar.nih.gov) and the Boston's Autism Consortium (autism.consortium.org). 870 independent controls were selected from the NIMH repository (www.nimhgenetics.org) to be of similar ancestry to cases (Baylor cases: 440 males, 65 females; Baylor controls: 240 males, 251 females: Broad cases: 429 males, 105 females, largely from the autism Consortium; Broad controls: 177 males, 202 females.) The Broad cases included probands only from trios. These trios were previously analyzed for de novo variants [5]. De novo variants were included in these analysis.
To evaluate sequence quality, 7 controls were sequenced at both centers. The capture/enrichment assays used were Nimblegen (Baylor) and Agilent (Broad). The Baylor samples were sequenced using the Solid platform and called with AtlasSNP 2 [30]. The Broad samples were sequenced using the Illumina platform and called with GATK [31]. Standard filters were used as part of both pipelines to produce calls for SNVs and indels. For details see Text S1.
In general, the MAF of SNVs matched well for the majority of the SNVs in the two data sets, but some differed considerably (Figure S6). One source of differences was the read depth: Broad reads had greater mean depth and also greater variability than Baylor reads (Figure S7). Overall counts of variants differed by platform (Table 5). We utilized additional filters to make these data sets more compatible. Relying on the validated de novo variants [5] and 7 overlapping samples we constructed an additional 3-round filter (see Text S1 and Table S7). First, for each data set, we excluded the variants that had or more missing calls. Second, we discarded the variants that had average depth less than . Third, we filtered the variants by the quality of the minor allele call. We defined the balance of depth for each minor allele call as the reference depth divided by the total depth. If more than half of the minor allele calls had a balance larger than or depth smaller than , we discarded this variant. Based on these features we constructed 6 filters denoted by PASS, MISS, DpBal, B, C and D of increasing stringency. For Filter DpBal, , for the Broad data set and for the Baylor data set; for Filter B, , ; for Filter C, , for the Broad data set and , for the Baylor data set; for Filter D, , . If less than half of the minor allele calls had a balance larger than , we kept this variant but changed the specific calls that did not pass the quality threshold from heterozygote to the common homozygote call.
Two rounds of filtering were performed on called indels. First, for each data set, we excluded indels with MAF greater than or more than missing calls. Second, we excluded indels that had more than six calls in one data set and none in the other data set.
For subjects sequenced, let denote the vector of phenotypes. For a gene with rare variants let be the -dimensional genotype vector. For dichotomous phenotypes we consider a logistic model:(1)where is the intercept, is a vector of regression coefficients for fixed covariates such as sex and ancestry, and is the vector of log odds ratios for the genetic variants. For analytical purposes only we also discuss the corresponding linear model for continuous phenotypes:(2)where . Without loss of generality, we assume .
We want to test the null hypothesis . One way to increase the power of the test is to assume that and test if [46]. Tests of this hypothesis are often called burden tests. To add prior information to this test, the weighted sum test has been proposed [22]. The idea of weighted sum test is to use rather than in model (1) so that biologically more plausible risk variants have larger weights in the test statistic. In our study, we use the weighted sum test with weights , where is the MAF of th variant. To implement the test, the genotypes in model (Eqn. 1) are replaced by a single composite term , which is the weighted sum of the genotype values of all rare variants . To assess significance of as a predictor, we use the score test.
There are drawbacks to a burden test. It assumes that all rare variants in the gene have the same direction and magnitude of association. In reality, variants can be damaging, protective, or have no effect, potentially reducing the power of the test. To overcome these drawbacks, the C-alpha test [25] has been proposed. The test is sensitive to unusual patterns in the distribution of rare variants across cases and controls. It has good power if most of the copies of a rare variant occur in cases (or controls), yet unlike the burden test, this pattern can vary across SNVs. SKAT [26] is a generalization of the C-alpha test. It has the advantage of readily incorporating covariates, but without covariates it reduces to the same form as C-alpha. This statistic is based on the generalized linear model (Eqn. 1 or 2), with random effects for the 's, which are assumed to follow an arbitrary distribution with mean zero and variance [47]. The test statistic is the score test for , which is of the formwhere K = GWG' is the kernel matrix, is a weight matrix, and for the logistic model (1) and for the linear model (2). The SKAT statistic can also be expressed in terms of the individual score tests for evaluating for each of the variants; let , , thenThe null distribution of is approximately a linear combination of distributions,(3)The SKAT p-values can be obtained by applying Davies exact method [48] to the data and inverting the characteristic function of .
Suppose we have samples from two (or more) datasets. To fix ideas, consider two data sets, and where and are the sample sizes, respectively. To perform meta-analysis using the weighted Z-score approach, first compute , where the p-values are obtained for each data set independently, and is the standard normal distribution function. Then the meta-analysis p-value is computed from , whereWhen applied to the SKAT test, this statistic combines information at the gene level without consideration of the directionality of any single variant effects.
We formally consider the SKAT test statistics in meta- and mega-analysis by deriving a closed form expression for the power of meta- and mega-analysis under restricted conditions. In the Results we show via simulations that the results hold more generally. Analysis is greatly simplified by choosing weights , a choice suggested in [22]. This weight is equivalent to scaling aswhere is the MAF of the th variant. For the following calculations we also assume no linkage disequilibrium (LD) between rare variants [49], [50]. Consequently we haveIn the Results we show that this assumption appears to be reasonable in the AASC data. Under these conditions and assuming there are no covariates, we note that , with(4)for the linear model (Eqn. 2), and(5)for the logistic model (Eqn. 1; see Text S1).
It follows that the mega-SKAT statistic , where the experiment-wise non-centrality parameter is the sum of non-centrality parameters from the individual studies: . Hence, when combining 2 studies, with sample sizes and , in which the 'th variant has log odds ratio , the contribution to the signal is proportional toNotice that this term is approximately equal to the number of realizations of the variants in the pooled data ( in the example above) times the square of the log odds ratio. For rare variants the number of realizations tends to be very small, emphasizing that large samples are essential to gain good power.
In a comparison of the power of meta- and mega-analysis we assume data sets and have the same sample size and rare variants at the same locations. Furthermore, building on our analysis above, we assume the individual test statistics from the two samples are distributed as and . Under -level type I error, the power function of weighted z-score meta-analysis and the power function of mega-analysis can be approximated as given in Text S1 (Eqn. S3–S4). The derived expressions are complex, but from Figure 2 we see, regardless of the degrees of freedom, mega-analysis has greater power than meta-analysis.
To gain more analytical insight, consider a gene for which each sample has sufficient coverage to detect all rare variants and that a total of rare variants are observed. Let and be the corresponding phenotype vectors and and the genotype vectors for variants , . Furthermore, let and denote the th variant scores corresponding to and . Next let's look at the test statistics for mega-analysis Under the alternative hypothesis, the per-variant scores and corresponding to 'th causal variant tend to have the same sign; positive for risk variants and negative for protective variants. Under the null hypothesis these per-variant score statistics are uncorrelated and tend to cancel each other out, on average. Consequently the final term in the expansion above tends to be positive under the alternative hypothesis and close to zero under the null.
In the Text S1 we find that the information captured by the meta-analysis statistic is approximated by the two lead terms (). Thus this expansion reveals why mega-analysis is more powerful than meta-analysis for quadratic test statistics such as SKAT. Mega-analysis cancels out false signals that differ in sign. Meta-analysis is restricted to gene level information and hence cannot account for directionality. The strength of the signal over a gene is determined by two factors: the sum of the per-variant contributions to the signal, versus the number of degrees of freedom. Both meta and mega-analysis assimilate the same signal (), but the strength of the signal for meta-analysis is apportioned over more degrees of freedom, effectively diminishing the power. For mega-analysis, the degrees of freedom increase only if the rare variants occur at different locations in the separate studies. The power advantage of mega-analysis is most pronounced when the rare variants accumulate at common locations across data sets. meta-analysis is not able to assimilate information within a variant across data sets as efficiently.
|
10.1371/journal.pmed.1002611 | Sexual transmission of Zika virus and other flaviviruses: A living systematic review | Health authorities in the United States and Europe reported an increasing number of travel-associated episodes of sexual transmission of Zika virus (ZIKV) following the 2015–2017 ZIKV outbreak. This, and other scientific evidence, suggests that ZIKV is sexually transmissible in addition to having its primary mosquito-borne route. The objective of this systematic review and evidence synthesis was to clarify the epidemiology of sexually transmitted ZIKV.
We performed a living (i.e., continually updated) systematic review of evidence published up to 15 April 2018 about sexual transmission of ZIKV and other arthropod-borne flaviviruses in humans and other animals. We defined 7 key elements of ZIKV sexual transmission for which we extracted data: (1) rectal and vaginal susceptibility to infection, (2) incubation period following sexual transmission, (3) serial interval between the onset of symptoms in a primary and secondary infected individuals, (4) duration of infectiousness, (5) reproduction number, (6) probability of transmission per sex act, and (7) transmission rate. We identified 1,227 unique publications and included 128, of which 77 presented data on humans and 51 presented data on animals. Laboratory experiments confirm that rectal and vaginal mucosae are susceptible to infection with ZIKV and that the testis serves as a reservoir for the virus in animal models. Sexual transmission was reported in 36 human couples: 34/36 of these involved male-to-female sexual transmission. The median serial symptom onset interval in 15 couples was 12 days (interquartile range: 10–14.5); the maximum was 44 days. We found evidence from 2 prospective cohorts that ZIKV RNA is present in human semen with a median duration of 34 days (95% CI: 28–41 days) and 35 days (no CI given) (low certainty of evidence, according to GRADE). Aggregated data about detection of ZIKV RNA from 37 case reports and case series indicate a median duration of detection of ZIKV of 40 days (95% CI: 30–49 days) and maximum duration of 370 days in semen. In human vaginal fluid, median duration was 14 days (95% CI: 7–20 days) and maximum duration was 37 days (very low certainty). Infectious virus in human semen was detected for a median duration of 12 days (95% CI: 1–21 days) and maximum of 69 days. Modelling studies indicate that the reproduction number is below 1 (very low certainty). Evidence was lacking to estimate the incubation period or the transmission rate. Evidence on sexual transmission of other flaviviruses was scarce. The certainty of the evidence is limited because of uncontrolled residual bias.
The living systematic review and sexual transmission framework allowed us to assess evidence about the risk of sexual transmission of ZIKV. ZIKV is more likely transmitted from men to women than from women to men. For other flaviviruses, evidence of sexual transmissibility is still absent. Taking into account all available data about the duration of detection of ZIKV in culture and from the serial interval, our findings suggest that the infectious period for sexual transmission of ZIKV is shorter than estimates from the earliest post-outbreak studies, which were based on reverse transcription PCR alone.
| Sexual transmission of Zika virus (ZIKV) is now documented, but the risks of transmission are not well understood.
It is not known whether other flaviviruses can be transmitted through sexual intercourse.
We developed a sexual transmission framework for ZIKV infection that identified 7 key elements related to ZIKV sexual transmission, and we conducted a living systematic review through 15 April 2018 of available evidence about each element.
We found that, where documented, sexual transmission of ZIKV is much more common from men to women than from women to men. For sexual transmission of ZIKV, the median serial interval—the time between onset of symptoms in 2 sexual partners—is 12 days.
The median duration of ZIKV RNA persistence in semen is longer (34 days) than in the female genital tract (12 days). ZIKV can be detected for longer periods using reverse transcription polymerase chain reaction compared to viral culture.
We found no evidence of sexual transmission for any other arthropod-borne flaviviruses.
Studies about the duration of detection of ZIKV in bodily fluids and the serial interval suggest that the period ZIKV can be transmitted through sexual contact might be shorter than was estimated from the earliest studies in 2016.
| Zika virus (ZIKV) can be transmitted between humans through sexual contact, although it is most commonly transmitted by infected Aedes spp. mosquitoes [1,2]. Sexual transmission of ZIKV has important implications for public health, for people living in endemic regions, and for sexual partners of travellers returning to non-endemic regions from endemic regions because ZIKV infection during pregnancy can cause congenital infection of the foetus and because ZIKV infection can trigger the immune-mediated neurological condition Guillain-Barré syndrome [3,4]. ZIKV is an RNA flavivirus. Flaviviruses are a genus of viruses from the Flaviviridae family, of which the majority are transmitted to vertebrates by infected mosquito or tick vectors [5].
Scientists working in Senegal in 2008 were the first to report presumed sexual transmission of ZIKV in a case report that documented their own symptoms and serological findings [6]. One scientist developed symptoms after returning to the US, and his wife, who had not travelled outside the US, became unwell 4 days later. The large ZIKV outbreak (2015–2017) in the Americas resulted in additional reports of travel-associated ZIKV sexual transmission in the US and Europe, which Moreira and colleagues synthesised descriptively in a systematic review of the literature up to December 2016 [7]. In vivo and in vitro experimental studies have provided evidence of the biological plausibility of this route of infection [8].
While possible sexual transmission has been established, there are many unanswered questions about the transmissibility of ZIKV through sexual intercourse. For mosquito-borne ZIKV infection, the incubation period and duration of viral shedding in serum have been estimated, allowing implications for blood donation to be assessed [9]. Additional information about parameters related to person-to-person transmission of ZIKV has not been systematically collated or quantified, although several narrative reviews have been published [10,11]. Evidence about sexual transmission of other arthropod-borne flaviviruses in humans, including West Nile virus (WNV), yellow fever virus (YFV), Japanese encephalitis virus (JEV), and dengue virus (DENV) [12], has not been synthesised, but WNV and YFV have been detected in human semen [13,14]. The primary objective of this review was to systematically review evidence about defined aspects of the sexual transmission of ZIKV. Secondary objectives were to systematically review evidence about the sexual transmissibility of other arthropod-borne flaviviruses and to establish these reviews using a living systematic review approach [15].
In March 2017, we developed a sexual transmission framework for ZIKV [16,17], based on standard concepts about person-to-person transmission of infection [18]. The framework includes key elements in the course of an infection in an individual and transmission to a sexual partner, some of which can be measured and others that can only be determined indirectly or through modelling. Fig 1 shows these elements and the relationships between them: (1) susceptibility to infection, (2) incubation period after sexual transmission, (3) serial interval, (4) duration of infectiousness, (5) reproduction number, (6) probability of transmission per sex act, and (7) transmission rate. The framework does not include transmission from and to mosquitoes, which would be needed to estimate the proportion of all ZIKV infections due to sexual transmission. The sexual transmission framework defined the outcomes and informed the structure of the review.
We performed this review as a living systematic review [15] because research into many aspects of ZIKV is a new and fast-moving field. Several studies are ongoing [19] and have published interim results [20], and updated results could affect public health decisions. The protocol for this review was registered on 19 May 2017 in the database PROSPERO (CRD42017060338) [21]. We summarise the details that make the review a living systematic review in S1 Text. Future updates will be reported quarterly online (http://zika.ispm.unibe.ch/stf/) and in the online comments section of this publication. Reporting is in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (S1 PRISMA Checklist).
The search includes the electronic databases PubMed, Embase, bioRxiv, arXiv, PeerJ, and LILACS and online repositories from the US Centers for Disease Control and Prevention, the European Centre for Disease Prevention and Control, the Pan American Health Organization, and the World Health Organization from the earliest date of each database and without language restrictions. The searches include Medical Subject Headings (MeSH) terms and keywords for ZIKV and flaviviruses together with terms and keywords for viral persistence and sexual transmission (S1 Text). An automated search is run every day, with results deduplicated and imported into REDCap (Research Electronic Data Capture). We checked reference lists of included studies to identify additional relevant studies. For this report, we identified studies published before and up to 15 April 2018.
We included observational studies, in vitro and in vivo studies, and mathematical modelling studies that directly addressed any of the elements of the sexual transmission framework in either humans or animals for ZIKV or another arthropod-borne flavivirus. We included observational studies that reported 1 or more cases of sexual transmission, 1 or more measurements of presence of virus in bodily fluids, or both. As bodily fluids we included semen, cervical and vaginal secretions, and saliva; diagnostic methods included reverse transcription polymerase chain reaction (RT-PCR) and viral culture. We did not include reviews, editorials, or commentaries that did not report original data. Table 1 provides an overview of the eligibility criteria for each outcome. Primary outcomes can be directly estimated from observational studies, and secondary outcomes are calculated or inferred from indirect evidence.
One reviewer screened titles and abstracts of retrieved papers. If retained in the first step, the same reviewer screened the full text of the paper. One reviewer extracted data into piloted extraction forms in REDCap [22]. A second reviewer verified exclusion decisions and data entry.
We provide descriptive summaries of findings about the elements of the ZIKV sexual transmission framework for basic research studies (element 1), observational epidemiological studies (elements 2–4), and mathematical modelling studies (elements 5–7). In addition, we used data from included studies to calculate estimates for the serial interval (i.e., the period between the start of symptoms in the primary and the secondary individual) and the duration of the detection of ZIKV. We report the median serial interval and its interquartile range. To estimate the duration of detection of ZIKV positivity, we conducted interval-censored survival analysis and fitted Weibull distributions using the “straweib” package [23,24] in R (version 3.4.1), based on previous studies [20,23,24]. We assumed that all infected patients were RT-PCR or viral culture positive at symptom onset. We report median estimated durations and corresponding 95% confidence intervals. Additional information about the methods is provided in S2 Text. For other flaviviruses, we summarise findings from all study types descriptively.
We assessed the methodology of included studies using specific checklists for each study type. For observational studies, we used the National Institutes of Health (NIH) Quality Assessment Tool for Case Series Studies [25] and UK National Institute for Health and Care Excellence (NICE) checklists for case–control studies and cohort studies [26]. For in vivo studies we used the SYstematic Review Centre for Laboratory animal Experimentation (SYRCLE) risk of bias tool for animal studies [27], and for mathematical modelling studies, the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Questionnaire to Assess Relevance and Credibility of Modelling Studies [28]. We performed the assessment by a consensus-driven approach among multiple reviewers. We appraised the certainty of the key elements according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool [29–31] (S3 Table). In accordance with GRADE, assessments of the overall certainty of evidence from observational studies started at low certainty. We downgraded the level of certainty for small sample size and evidence from case reports or case series. We assessed outcomes of mathematical modelling studies as high, medium, low, or very low certainty.
We identified 1,227 unique citations and excluded 901 by title and abstract screening (Fig 2). Of the remaining 326 potentially eligible citations with relevant abstracts, 128 publications were eligible for inclusion. Table 2 summarises characteristics of the included studies.
We included 41 in vivo and in vitro studies of ZIKV [32–69] (Table 2). Of these 41 studies, 6 were in vitro studies and 35 were studies in in vivo animal models: 12 in nonhuman primates (NHPs) such as cynomolgus macaques (Macaca fascicularis), rhesus macaques (M. mulatta), and common marmosets (Callithrix jacchus) and 23 in mice. In 1 study, both guinea pigs and NHPs were used [64]. These studies provide insight into the underlying biological mechanisms of susceptibility to ZIKV infection through sexual transmission and substantiate the biological plausibility of this transmission route.
In mouse and NHP models, the vaginal and rectal mucosae were shown to be susceptible to infection with ZIKV [36,40,47,49,51,59]. When ZIKV-infected male mice were mated with uninfected female mice, the female mice became infected [47,53,69]. Female-to-male transmission of ZIKV in mice was unsuccessful [47]. In rhesus macaques, systemic infection through oropharyngeal mucosal inoculation with ZIKV was only successful after inoculation with a very high dose of virus, suggesting a very low risk of oral mucosal transmission [54]. Four rhesus macaques became viraemic after intranasal or intragastric inoculation with ZIKV [64]. In guinea pigs, direct transmission between animals infected subcutaneously with ZIKV and co-housed uninfected animals was seen [64].
Human prostate cells, testicular cells, and mature spermatozoa are susceptible in vitro to ZIKV infection [35,60–62]. Human Sertoli cells can support high levels of ZIKV replication and persistence [70]. In multiple mouse models, using different strains of ZIKV, the testes seem to be a preferred site for viral replication, able to sustain high viral loads for a longer duration than other organs [32,34,35,43,47,49,51,58,63]. In some of these models, ZIKV caused inflammation of the testes [35,39,41,47,49], reduced testicular size, and decreased levels of testosterone [49,56,68,71]. The testes of experimentally infected NHPs harboured high levels of ZIKV [37,38]. High titres of ZIKV RNA were detectable in semen until day 28 in rhesus and cynomolgus macaques [38]. However, 1 group of rhesus macaques had only low levels of viral RNA in the testes and no detectable virus in the prostate or epididymis [44]. In common marmosets, ZIKV RNA in semen was sporadically detected [65].
The female genital tract of macaques was able to sustain ZIKV replication for shorter durations and with lower viral loads than the male genital tract [33,38,46]. Although cervical and endometrial cells were susceptible in vitro [35], virus was not detected in the female genital tract in mice or NHPs for longer than 7 days after infection [33,46]; in 1 study, the ovary sustained higher titres up to 14 days post-infection [35]. Intravaginal infection of mice led to systemic infection [36,40,42,47,48,50,67,68,69] and to adverse congenital outcomes [47]. In pregnant female mice, sexual transmission led to more ZIKV dissemination to the female reproductive tract, compared with subcutaneous or intravaginal inoculation [69]. In NHPs the incubation time following infection was longer for intravaginal infection compared with subcutaneous infection [55].
In mice, viral titres were lower in the salivary glands than the testes and ovaries [35]. In NHPs, viral RNA was detected in saliva up to 28–42 days [33,37,38,52,57], and ZIKV could be cultured at day 7 and day 14 [38,65].
Most studies did not describe in detail the methods used to avoid bias. Detailed certainty assessment of the in vivo studies is provided in S3 Table.
We included 18 studies reporting on the sexual transmission potential of other arthropod-borne flaviviruses [14,141–157]. Ten of the 18 studies (56%) were in vitro experiments or observations in animals, and 8 studies (44%) were case reports or case series. JEV was demonstrated to be transmissible from male to female pigs via semen [141–144]. Persistence of virus was demonstrated for at least 17 days in boars [141]. JEV can be cultured from the seminal fluids of pigs [145]. In humans, we found 1 case report of male-to-female sexual transmission of WNV, although the secondary partner also lived in a mosquito endemic area [146]. WNV was found postmortem in the prostate and testis of a 43-year-old man on immunosuppressive therapy following a kidney transplant [147]. Intravaginal inoculation of WNV in mice led to local acute inflammation followed by systemic illness in a proportion of the animals [148]. The testes of 6 Japanese macaques (M. fuscata) showed low DENV neutralising antibody titres [149]. Experimentally DENV-infected pigtail macaques (M. nemestrina) showed dissemination of virus in the prostate gland and seminal vesicles [150]. DENV RNA could be detected in experimentally infected mice 3 days after infection [151]. In humans, 4 case reports describe the presence of DENV in saliva, diagnosed by either RT-PCR or viral culture, for up to 7 days [152–155]. DENV RNA was demonstrated in the vaginal secretion of 1 patient up to 18 days after onset of symptoms [156]. Female mice that were mated to male mice infected with tick-borne encephalitis virus had worse reproductive outcomes than ones mated to a group of non-infected males; in 1 female mouse, the virus was detected [157]. YFV was demonstrated in the urine and semen of a patient by RT-PCR 21 days after onset of symptoms [14].
This systematic review summarises published data related to sexual transmission of ZIKV and other arthropod-borne flaviviruses published on or before 15 April 2018. In animals, vaginal and rectal mucosae are susceptible to ZIKV, with the testis as a preferred site of replication. Male-to-female transmission was more frequent than female-to-male transmission in animal models and in humans. In humans, we estimated the serial interval for sexually transmitted infection to be 12 (interquartile range: 10–14.5) days. ZIKV was detectable in semen for a median of 34 (95% CI: 28–41) days by RT-PCR and 9.5 (95% CI: 1.2–20.3) days by viral culture. In mathematical modelling studies, the reproduction number for sexual transmission of ZIKV was below 1. The overall certainty of the evidence was low. We found no evidence that other arthropod-borne flaviviruses can be sexually transmitted.
The ZIKV sexual transmission framework allowed us to synthesise evidence from both animal and human studies in a structured way, taking into account the risks of bias in the included studies. Susceptibility of tissues to ZIKV could only be assessed in animal models. There were consistent findings in animal models that help to explain the overrepresentation of reported cases of human male-to-female transmission, even though mice are not a natural host for ZIKV and in vivo studies often use immunocompromised animals [8]. First, vaginal mucosae are more susceptible than urethral mucosae to infection [36,40,47,49,51,59]. Second, high levels of ZIKV replication in the testes in mice and sustained detection of viral RNA and of virus in tissue culture in mice and NHP models are consistent with the longer duration of detection in men than women. Rectal mucosa is also susceptible to ZIKV, so, although only observed once [85], unprotected anal intercourse is also a likely route of ZIKV transmission. The risk of bias of the included in vivo studies, as assessed with the SYRCLE tool, was high. Most of these studies explored the suitability of animal models or investigated pathophysiological pathways, and potential sources of bias were rarely reported.
Our analysis shows that, when assessed from case reports and case series, the duration of detection of ZIKV in semen by RT-PCR is overestimated; all reports are of people with ZIKV detected, and a small number of outliers influence the estimate. A prospective cohort study that consecutively enrolled people with symptomatic ZIKV infection estimated a shorter duration of persistence [22]. Case reports and case series are early sources of information about a new disease, but, by the nature of these studies, researchers report novel and unusual findings. Parameters and effect sizes estimated from aggregating data from these sources are likely to be overestimates, reflecting the so-called random high: extreme values in a distribution that are observed by chance and are more likely to be reported because they are noteworthy [158]. As evidence has accumulated in well-designed prospective studies, the estimated duration of persistence of ZIKV RNA has decreased. Notably, the prospective cohort study in Puerto Rico found ZIKV in semen in only half of men with symptomatic infection and vaginal fluid in only 1 of 50 women. Similarly, ZIKV was found in semen in only 60/183 (33%) ZIKV-infected men in the US [74]. Persistence of viral RNA in body fluids is often used as a proxy for the duration of ZIKV infectiousness, although it remains unclear whether the presence of viral RNA corresponds with infectious virus. ZIKV RNA positivity persists for longer than detection of ZIKV in viral culture in both mice [47] and human semen samples. However, viral cultures might underestimate the duration of infectiousness if low pH or other specimen-dependent factors produce false negative results [83,159]. The estimated serial interval was based on observations from only 15 couples, but was consistent with that of several respiratory infectious diseases [160]. The serial interval for sexual transmission was towards the lower end of estimates for mosquito-borne transmission (10–23 days) [161].
Some elements of the infection process, such as the incubation period, transmissibility of ZIKV per sex act, and transmission rate could not be observed. In the mathematical models published so far [139,140,162], the estimates were based on assumptions about the transmissibility of mosquito-borne infection. Estimates from our review might provide more reliable data for use in future modelling studies. The potential for sustained sexual transmission of ZIKV appears low, based on the reproduction number estimated in the mathematical modelling studies. The estimated reproduction number was higher for mosquito-borne transmission, 1.96 (95% CI: 0.45–6.23) than for sexual transmission [139], although this number is highly dependent on the geographical location [163].
This review did not find evidence supporting sexual transmission of other arthropod-borne flaviviruses. The continual updating of the literature search identified a finding of YFV in urine and semen [14]. However, it remains to be clarified for many viruses if detection in semen means that there is a risk of sexual transmission [13].
The strengths of our living systematic review are the high coverage of the body of published literature, the structured overview, and the reanalysis of individual patient data on persistence of ZIKV. The automation of search and deduplication processes makes it feasible to keep the review updated as new information becomes available. Updated analyses of the data from case reports show regression to the mean of the median estimate of the duration of RNA detection in semen (https://zika.ispm.unibe.ch/stf/). Future updates of this review will also allow for incorporation of techniques to synthesise mathematical modelling studies, such as multi-model ensembles. This study also has limitations. Screening and data extraction were not done by 2 independent reviewers because of time constraints, but we believe that we reduced errors by having a second reviewer check the decisions and data extracted. The statistical methods used to estimate the duration of persistence of ZIKV in bodily fluids assume that all samples are positive for ZIKV at time 0 [20], which might not be the case. Additionally, the sexual transmission framework might not include all factors that are required to investigate the risks of sexual transmission of ZIKV. The certainty of this body of evidence was assessed as being low or very low because of bias in the observational study designs and the indirectness of evidence from animal studies. The certainty of the evidence base could increase if the design and reporting of both animal and human studies improve and if their findings are consistent with, and increase the precision of, the evidence presented here.
The risk of sexual transmission of ZIKV is particularly relevant for women who are pregnant or planning a pregnancy and for people with high levels of sexual partner change such as some groups of men who have sex with men and women at high risk. An expert group has used the ZIKV sexual transmission framework to stimulate discussion about research priorities [17]. One important limitation to the generalisability of findings from our review is that the data that we analysed about sexual transmission of ZIKV in humans relied largely on information from travellers returning from endemic areas with symptomatic ZIKV infection and their sexual partners. This group probably differs from people in endemic regions in ways that could affect sexual transmission of ZIKV, such as previous exposure to other flaviviruses [164]. Additional studies in ZIKV endemic settings could enrol travellers who work in areas with mosquito-borne ZIKV transmission and who return to families living nearby but in areas, e.g., at high altitude, where the vector does not survive [17]. There are unanswered questions about the potential for asymptomatic ZIKV infection following sexual transmission, since ZIKV transmitted by mosquitoes often causes asymptomatic infection [165–167], about clinical differences between ZIKV infections acquired through sexual and mosquito-borne routes, and about the long-term consequences of ZIKV in the genital tract, such as its effects on the testes and on male infertility. Research about the potential for sexual transmission of other flaviviruses is needed, although these viruses often display different symptomatology or affinity for different species.
Clinicians and policymakers need information that helps to advise both opposite sex and same sex couples on how to reduce the risk of sexual transmission of ZIKV. The relationship between detectable RNA in semen and infectiousness therefore needs to be further investigated in both laboratory and epidemiological studies. Current guidelines for travellers returning from endemic areas advise 6 months of protected intercourse [1,2]. As more information becomes available, a revision of the duration of protection might be indicated.
This living systematic review gives an up-to-date synthesis of information about the sexual transmission of ZIKV with a structured framework. Planned regular updates will allow timely updating of relevant data from a rapidly expanding evidence base. We did not quantify the absolute risk of sexual transmission of ZIKV, but it appears small based on information about the proportion of people with symptomatic ZIKV who have ZIKV detected in genital secretions and the short median duration of detection of ZIKV in semen and vaginal fluid. Taking into account all available data about the duration of detection of ZIKV in culture and from the serial interval, our findings suggest that the infectious period for sexual transmission of ZIKV is shorter than estimates from the earliest post-outbreak studies, which were based on RT-PCR alone.
|
10.1371/journal.ppat.1007314 | A planarian nidovirus expands the limits of RNA genome size | RNA viruses are the only known RNA-protein (RNP) entities capable of autonomous replication (albeit within a permissive host environment). A 33.5 kilobase (kb) nidovirus has been considered close to the upper size limit for such entities; conversely, the minimal cellular DNA genome is in the 100–300 kb range. This large difference presents a daunting gap for the transition from primordial RNP to contemporary DNA-RNP-based life. Whether or not RNA viruses represent transitional steps towards DNA-based life, studies of larger RNA viruses advance our understanding of the size constraints on RNP entities and the role of genome size in virus adaptation. For example, emergence of the largest previously known RNA genomes (20–34 kb in positive-stranded nidoviruses, including coronaviruses) is associated with the acquisition of a proofreading exoribonuclease (ExoN) encoded in the open reading frame 1b (ORF1b) in a monophyletic subset of nidoviruses. However, apparent constraints on the size of ORF1b, which encodes this and other key replicative enzymes, have been hypothesized to limit further expansion of these viral RNA genomes. Here, we characterize a novel nidovirus (planarian secretory cell nidovirus; PSCNV) whose disproportionately large ORF1b-like region including unannotated domains, and overall 41.1-kb genome, substantially extend the presumed limits on RNA genome size. This genome encodes a predicted 13,556-aa polyprotein in an unconventional single ORF, yet retains canonical nidoviral genome organization and expression, as well as key replicative domains. These domains may include functionally relevant substitutions rarely or never before observed in highly conserved sites of RdRp, NiRAN, ExoN and 3CLpro. Our evolutionary analysis suggests that PSCNV diverged early from multi-ORF nidoviruses, and acquired additional genes, including those typical of large DNA viruses or hosts, e.g. Ankyrin and Fibronectin type II, which might modulate virus-host interactions. PSCNV's greatly expanded genome, proteomic complexity, and unique features–impressive in themselves–attest to the likelihood of still-larger RNA genomes awaiting discovery.
| RNA viruses are the only known RNA-protein (RNP) entities capable of autonomous replication. The upper genome size for such entities was assumed to be <35 kb; conversely, the minimal cellular DNA genome is in the 100–300 kilobase (kb) range. This large difference presents a daunting gap for the proposed evolution of contemporary DNA-RNP-based life from primordial RNP entities. Here, we describe a nidovirus from planarians, named planarian secretory cell nidovirus (PSCNV), whose 41.1 kb genome is 23% larger than any riboviral genome yet discovered. This increase is nearly equivalent in size to the entire poliovirus genome, and it equips PSCNV with an unprecedented extra coding capacity to adapt. PSCNV has broken apparent constraints on the size of the genomic subregion that encodes core replication machinery in other nidoviruses, including coronaviruses, and has acquired genes not previously observed in RNA viruses. This virus challenges and advances our understanding of the limits to RNA genome size.
| Radiation of primitive life as it took hold on earth was likely accompanied by genome expansion, which was associated with increased complexity and a proposed progression from RNA-based through RNA-protein to DNA-based life [1]. The feasibility of an autonomous ancient RNA genome, and the mechanisms underlying such fateful transitions, are challenging to reconstruct. It is especially unclear whether RNA entities ever evolved genomes close to the 100–300 kilobase (kb) range [2, 3] of the “minimal” reconstructed cellular DNA genome [4]. This range overlaps with the upper size limit of nuclear pre-mRNAs [5], which is likely the upper size limit for functional RNAs due to the relative chemical lability of RNA compared to DNA. However, pre-mRNAs are incapable of self-replication, the defining property of primordial genomic RNAs.
RNA viruses may uniquely illuminate the evolutionary constraints on RNA genome size [6–9], whether or not they descended directly from primitive RNA-based entities [10–13]. The same constraints may also inform research on the biology and pathogenesis of RNA virus infections, because they shape the diversity of viral proteomes and RNA elements. The causes and consequences of changes in genome size can be understood in the context of a relationship that locks replication fidelity, genome size, and complexity within a unidirectional triangle [14]. RNA viruses appear to be trapped in the low state of this relationship (Eigen trap) [15], which is characterized by low fidelity (high mutation rate), small genome size (10 kb average), and low complexity (few protein/RNA elements). Specifically, low-fidelity replication without proofreading constrains genome expansion [16], since accumulation of mutations [17] would lead to the meltdown of larger genomes during replication (error catastrophe hypothesis) [18, 19].
This constraining relationship is supported by evidence from nidoviruses (order Nidovirales): enveloped viruses with positive-stranded RNA genomes in the range of 12.7 to 33.5 kb–the largest known RNA genomes [20–23] (Fig 1A and 1B, S1 Table). The Nidovirales is composed of two vertebrate families, Arteriviridae and Coronaviridae (subfamilies Coronavirinae and Torovirinae), and two invertebrate families, Mesoniviridae and Roniviridae [24, 25], and includes important pathogens of humans (Severe acute respiratory syndrome coronavirus, SARS-CoV; Middle East respiratory syndrome coronavirus, MERS-CoV) and livestock (different arteriviruses, coronaviruses and roniviruses) [26–30]. All known nidoviruses with genomes larger than 20 kb also encode a proofreading exoribonuclease (ExoN) [14, 31–34] (Fig 1B), which, once acquired by an ancestral nidovirus, may have relieved the constraints on all three elements of the triangular relationship simultaneously, providing a solution to the Eigen trap [14].
In the last 20 years of virus discovery, however, despite the application of unbiased metagenomics to RNA virus discovery [35, 36], the largest-known RNA viral genome has only increased ~10% in size–a mere fraction of the nearly ten-fold increase observed for DNA viruses [37–39] (Fig 1A). Thus, other constraints have apparently limited genome size, even in RNA viruses equipped with proofreading capability. Further characterization of nidovirus molecular biology, variation, and evolution may provide insight into these other factors.
Nidovirus genomes are typically organized into many open reading frames (ORFs), which occupy >90% of genome and can be divided into three regions: overlapping ORF1a and ORF1b, and multiple ORFs at the 3’-end (3’ORFs) [14] (Fig 2). The products of these regions predominantly control genome expression/replication, and virus assembly/dissemination, respectively.
ORF1a and ORF1b are expressed by translation of the genomic RNA that involves a -1 programmed ribosomal frameshifting (PRF) at the ORF1a/ORF1b overlap [40, 41]. The two polyproteins produced without or with frameshifting, pp1a (ORF1a-encoded) and pp1ab (ORF1a/ORF1b-encoded), vary in size from 1,727 to 8,108 aa. They are processed to a dozen or more proteins by the virus’ main protease (3CLpro, encoded in ORF1a; Fig 2) with possible involvement of other protease(s) [42]. These and other proteins form a membrane-bound replication-transcription complex (RTC) [43, 44] that invariably includes two key ORF1b-encoded subunits: the Nidovirus RdRp-Associated Nucleotidyltransferase (NiRAN) fused to an RNA-dependent RNA polymerase (RdRp) [45, 46], and a zinc-binding domain (ZBD) fused to a superfamily 1 helicase (HEL1), respectively [47–50]. The RTC catalyzes the synthesis of genomic and 3’-coterminal subgenomic RNAs, the latter via discontinuous transcription that is regulated by leader and body transcription-regulating sequences (lTRS and bTRS) [51–53]. Subgenomic RNAs are translated to express virion and, in ExoN-positive viruses, accessory proteins encoded in the 3’ORFs [23, 54–59]. Most nidovirus proteins are multifunctional, but some released from the N-terminus of pp1a/pp1ab and/or encoded in the 3’ORFs are specialized in the modulation of virus-host interactions [26, 60–65].
Intriguingly, despite the large variation in genome size among extant nidoviruses, the size of ORF1b varies extremely little within either the ExoN-negative (12.7–15.7 kb genome range) or ExoN-positive (19.9–33.5 kb genome range) nidoviruses [66]. There is no overlap between these two groups of viruses in the size range of ORF1b: the smallest ORF1b of an ExoN-positive nidovirus is almost double the length of the largest ExoN-negative ORF1b. In contrast, the ORF1a and 3’ORFs regions exhibit considerable size variation, and their sizes overlap between the ExoN-positive and ExoN-negative clades.
A current theoretical model of nidoviral genome dynamics, the three-wave model, proposes that a genome expansion cycle is initiated by a bottleneck increase of ORF1b (the first wave) in a common ancestor of ExoN-positive nidoviruses, which then permits parallel expansion of ORF1a and, often, 3’ORFs in subsequent overlapping waves in separate lineages [66]. Extant nidovirus genomes of different sizes have reached particular points on this trajectory of genome size, apparently due to the lineage-specific interplay of poorly understood genetic and host-specific factors. A single cycle of this process can account for genome expansion from the lower end of genome sizes (12.7 kb) to the upper end (31.7 kb); expansion of genomes far beyond that size range has been hypothesized to require a second cycle, beginning with a new wave of ORF1b expansion [66]. In the absence of newly discovered RNA viruses with significantly larger genomes since the time of that analysis, and due to the unknown nature of the ORF1b size constraint(s), however, the feasibility of a second cycle has remained uncertain, and the notion that ~34 kb is close to the actual limit of RNA virus genome size [35] has seemed plausible.
To examine whether this limit applies beyond the currently recognized ~3000 RNA virus species (isolated from only a few hundred host species), further sampling of virus diversity is required, particularly from host species in which viruses have thus far remained virtually unknown. To this end, we analyzed de novo transcriptomes from both major reproductive biotypes (strains) of the planarian Schmidtea mediterranea [67]: a hermaphroditic sexual strain, and an asexual strain whose members reproduce via transverse fission [68]. We report the discovery and characterization of the first known planarian RNA virus, dubbed the planarian secretory cell nidovirus. PSCNV has the largest RNA genome by a considerable margin–a feat made more remarkable by the fact that its genome is organized as a single ORF. Concomitantly, it has adapted the nidoviral regulatory toolkit in novel ways, and acquired many features that revise the known limits of viral genomic and proteomic variation–some of these features being unique among nidoviruses, others among RNA viruses, and still others among all known viruses. Our results imply that viruses with the nidoviral genetic plan have the potential to expand RNA genomes further along the trajectory envisioned by the multi-cycle, three-wave model.
To identify potential nidovirus-like sequences in the planarian transcriptome, we queried two in-house de novo-assembled Schmidtea mediterranea transcriptomes [67] for sequences that significantly resembled a reference coronavirus genome. Two nearly identical (99.97%) nested transcripts, txv3.2-contig_1447 (originating from the sexual strain) and txv3.1-contig_12746 (from the asexual strain), showed a statistically significant similarity to known nidoviruses as reciprocal BLAST top hits. We hypothesized that these transcripts are genomic fragments of a new nidovirus species. We further identified several overlapping EST clones with >99% nucleotide identity to the transcriptome contigs, and assembled these into a putative partial genome (S1 Fig). Finally, with additional transcriptome search iterations and Sanger sequencing of the transcript 5’-end, we assembled a 41,103-nt transcript (excluding the polyA tail). Based on several criteria (see below), we assigned this RNA sequence to the genome of a virus we dubbed Planarian Secretory Cell Nidovirus (PSCNV) (S1 Fig). This sequence was the reference genome used for further analyses (see Materials and Methods for more detail).
The complete PSCNV genome encodes a single 40,671-nt ORF that is flanked by a 128-nt 5’-UTR and a 304-nt 3’-UTR (Figs 1B and 2). In addition, we found the main ORF overlapping multiple small ORFs in other reading frames, whose lengths exceeded 150 nt: 8 ORFs in the same strand as the large ORF (plus-strand), lengths ranging from 156 to 267 nt, 5 of which mapped to the 3’-terminal quarter of the genome; and 24 ORFs in the reverse complement strand (minus-strand), distributed throughout the genome, with lengths ranging from 153 to 681 nt. To further verify the presence of the viral genome in vivo, we amplified large overlapping genomic subregions by RT-PCR (S2 Table, S1 Fig) [69]. These sequences could not be amplified from S. mediterranea genomic DNA, nor could they be found in the reference planarian genome [70]; thus, they appear to derive from an exogenous source.
A survey of 16 S. mediterranea RNA-seq datasets from nine laboratories worldwide uncovered PSCNV reads in five datasets from three American locations. Of the positive datasets, three originated from the sexual strain, and two from the asexual strain. Overall, viral sequences were much more abundant in transcriptomes obtained from sexual strains (S3 Table).
The PSCNV sequences detected in these studies vary little from one another. The three most complete sequences (tentatively reconstructed from PRJNA319973, PRJNA79031, and PRJNA421285) are characterized by >99.9% identity across a nearly 13 kb span of the genome, where at least 2 reads (and at least 10 reads for >95% of positions) from each dataset mapped to each position of the reference genome. Indeed, sequences from PRJNA319973 and PRJNA79031 –the two datasets from the Newmark laboratory–exhibit only a single mutation relative to the reference genome, and the sequence from PRJNA421285 –from the Sanchez Alvarado laboratory–differs at only 9 positions (S4 Table). This low variation is notable, as two of the datasets analyzed (PRJNA79031 and PRJNA421285) are derived from sexual S. mediterranea, and the other one (PRJNA319973) from an asexual S. mediterranea lab strain. The source populations of these two (freshwater) strains are separated from each other by about 500 km of the Mediterranean Sea: the asexual laboratory strain was established from a population in Barcelona [71], and the sexual strain originates from a Sardinian population. A recent study of the evolutionary history of S. mediterranea suggests that these populations diverged from each other at least 4 million years ago [72].
Given the long-separate history of these two planarian strains prior to becoming research subjects and the relatively high mutation rate in characterized nidoviruses, the detection of nearly identical viral transcripts in both strains is strong evidence that the virus is transmissible. The absence of viral sequences from asexual strains in most labs, and their presence in all labs that have reported RNA-seq data from the sexual strain, strongly suggest that the virus first infected (or was endemic to) the sexual strain, and has subsequently spread to asexual laboratory stocks.
We examined PSCNV infection in planarian tissues by whole-mount in situ hybridization (ISH). PSCNV RNA was detected abundantly in cells of the secretory system in both sexuals and asexuals (Fig 3A). Fluorescent ISH revealed viral RNA in gland cell projections that form secretory canals (Fig 3B). Notably, viral RNA was detected largely in ventral cells (Fig 3C) whose localization corresponds to mucus-secreting cells that produce the slime planarians use for gliding locomotion, and to immobilize prey [73].
We then analyzed planarians by electron microscopy (EM) for the presence of viral structures. In one specimen, membrane-bound compartments containing 90–150 nm spherical-to-oblong particles resembling nidoviral nucleocapsids [74, 75] were found in the cytoplasm of mucus-secreting cells. These sub-epidermal gland cells are notable for their abundant rough endoplasmic reticulum and long projections into the ventral epithelium, through which they secrete mucus (S2 Fig). These cells provide an ideal environment for nidoviral replication, which co-opts host membranes to produce viral replication complexes [76, 77]. Putative viral particles were found both in deep regions of these cells, and in their trans-epidermal projections (Fig 4A–4C). The latter location suggests a route for viral transmission. Notably, particles in sub-epidermal layers have a “hazy” appearance and are embedded in a relatively electron-dense matrix (Fig 4D). In contrast, particles closer to the apical surface of the epidermis appear as relatively discrete structures, standing out against electron-lucent surrounding material (Fig 4E). The size, ultrastructure, and host-cell locations are all consistent with these structures being nidoviral nucleocapsids [74, 75].
In 280 images from the positive specimen, all other ultrastructural features were normal. Importantly, typical mucus vesicles were evident in this specimen, often immediately adjacent to vesicles containing putative virions (Fig 4C, see also S2 Fig). As such, we determined that these structures do not represent artefacts caused by atypical fixation of this specimen.
The genome and proteome of PSCNV are by far the largest yet reported for an RNA virus. Its RNA genome is ~25% larger than that of the next-largest known RNA virus (BPNV, [21]), which is separated by a comparable margin from the first nidovirus genome sequenced 30 years ago (IBV, [78]) (Fig 1A). The size of the predicted PSCNV polyprotein (13,556 amino acids, aa) is 58–67% larger than the largest known RNA virus proteins produced from a single ORF (8,572 aa; Gamboa mosquito virus, [79]) or multiple ORFs through frameshifting (8,108 aa; BPNV, [21]) (Fig 5).
Functional annotation of the PSCNV polyprotein by comparative genomics [14, 31, 80, 81] presented a distinct bioinformatics challenge, due to its weak similarity to other proteins and its extremely large size, which exceeds the average size of protein domains by approximately 75-fold. We delineated at least twenty domains in the PSCNV polyprotein, including twelve domains conserved in nidoviruses or other entities, using a multistage computational procedure that combined different analyses within a probabilistic framework (Fig 2; S3–S16 Fig; S5 Table; see Materials and Methods). We initially identified six regions highly enriched in hydrophobic residues characteristic of transmembrane domains, named TM1 to TM6 accordingly (Fig 2). The number and relative location of the TM domains resemble those found in the proteomes of nidoviruses, which commonly have five or more TM domains in non-structural and structural proteins [82–85]. We then identified fourteen regions enriched in individual amino acid residues (S4 Fig), with the strongest signal observed for Thr-rich region (residues 10429–10559, 44.3% Thr residues, up to 13.4 SD above the mean). Notably, the Thr-rich region overlaps with a Ser-rich region (10461–10501 aa, 19.5% Ser residues, up to 5.5 SD above the mean). Subsequently, two tandem repeats were identified toward the N-terminus of the polyprotein (residues 1616–1682 and 1686–1751, Probability 96.6%, S5 Fig), which showed no significant similarity to other proteins in the databases using HHsearch.
We used the domains described above to split the polyprotein into nine regions, which were analyzed by an iterative HHsearch-based procedure (outlined in S3 Fig and S1 Materials and Methods). Our approach identified eight domains that, together with TM2 and TM3, form a canonical synteny of replicative domains in the central part of the polyprotein (genome), which is characteristic of known invertebrate nidoviruses (Fig 2): 3CLpro, NiRAN, RdRp, ZBD, HEL1, ExoN, and S-adenosylmethionine (SAM)-dependent N7- and 2’-O-methyltransferases (N-MT and O-MT, respectively). Five of these domains (3CLpro, NiRAN, RdRp, HEL1, and O-MT) were identified by hits exceeding the 95% Probability threshold, while three others were based on weaker hits: 35.0% for ZBD, 39.1% for ExoN, and 80.8% for N-MT. Despite the lower Probability values obtained for the latter three domains, synteny and conservation of essential functional residues strongly suggest that they encode true homologs of canonical nidoviral proteins. Overall, the analysis demonstrates the existence of the three definitive nidoviral genomic subregions in the PSCNV single-ORF genome: ORF1a-, ORF1b-, and 3’ORFs-like. Within these regions, TM2, 3CLpro, and TM3 map to the ORF1a-like region, while NiRAN, RdRp, ZBD, HEL1, ExoN, N-MT, and O-MT map to the ORF1b-like region.
In addition to the canonical replicative domains present in the canonical order and location, we found four domains that are novel for nidoviruses: one upstream and three downstream of the array of the conserved replicative domains (S5 Table). These include a homolog of ribonuclease T2 (RNase T2, Probability 80.0%) upstream of the TM2, two fibronectin type II domains (FN2a and FN2b, 91.3% and 78.5%, respectively), and an ankyrin repeats domain (ANK, 98.9%) downstream of the O-MT. For the three domains identified with the under-threshold hits, additional support came from conservation of functionally important residues (see below).
We subsequently generated multiple sequence alignments (MSAs) of these domains for a representative set of established nidovirus species, followed by phylogenetic reconstruction to characterize PSCNV by revealing common and unique features of its conserved domains. The next three sections summarize the salient features of the replicative, novel, and structural domains of the polyprotein.
The 3’ORFs region of nidoviruses encodes components of the enveloped virion [23, 54], which define receptor specificity [55–57] and typically include the nucleocapsid protein (N), characterized by biased amino acid composition and structurally disordered region(s) [104, 105], spike glycoprotein(s) (S protein in corona- and toroviruses) and transmembrane matrix protein (M in corona- and toroviruses) enriched with TM regions [58, 59, 106]. As expected from the weak sequence conservation of this region in other nidoviruses [14, 107] and its weak similarity with other viruses [108], we were unable to find statistically significant similarity between the PSCNV polyprotein and structural proteins of the known nidoviruses. Nevertheless, important nidoviral themes are evident.
First we noted that the genome distribution of the TM-encoding regions in PSCNV conformed to that observed in other nidoviruses, with TM1 and TM2 located upstream of 3CLpro, TM3 C-terminal to 3CLpro, and TM4–TM6 downstream, in the 3’ORFs-like region (Fig 2). In nidoviruses, the TM domains encoded in the 3’-genome region are known to be part of the S and M proteins or their equivalents, and occasionally additional accessory proteins [14, 58, 59, 106, 109]. The extracellular portion of the S protein is supported by multiple disulfide bridges between conserved Cys residues [56]. In PSCNV, a Cys-rich region was observed downstream of TM5 (S4 Fig). In an approximately 650 aa region surrounding the TM6 domain (4.7% of the polyprotein length), we identified six areas enriched in Pro, Leu, Gly, Gln, Asn, or Arg, in close proximity to each other (S4 Fig). This region accounted for 43% of all residue-enriched areas in the polyprotein; such an exceptionally high concentration of sequences enriched with specific amino acids is indicative of unusual properties. Accordingly, this area was predicted to include the longest stretch of disordered regions. In nidoviruses, disordered hydrophilic-rich areas are characteristic of N proteins.
In PSCNV, the polyprotein region downstream of O-MT is ~4000 aa, more than twice as large as the largest known structural protein of nidoviruses [106]. We reasoned that this part of its polyprotein might be processed by cellular signal peptidase (SPase) and/or furin to produce several proteins, as documented for maturation of the structural proteins of many RNA viruses, including nidoviruses [110–114]. Indeed, our analysis of potential cleavage sites of these proteases revealed highly uneven distributions (S4 Fig), with sites predicted only in the N- and C-terminal parts of the polyprotein: 1400–3100 aa (one SPase and four furin sites) and 10200–13200 aa (three SPase and five furin sites). All of these are outside of the region that must be processed by 3CLpro. With the exception of the most C-terminal furin site, all predicted sites are in close vicinity to provisional borders of the domains described above, as would be expected if these domains function as distinct proteins. Specifically, if the predicted SPase and furin sites are cleaved, TM1, TM4, TM5, and TM6 would end up in separate proteins, with one protein including the TM4 and ANK domains. With predicted cleavage sites flanking it from both sides, TM5 may be released as a separate protein, most similar to M proteins in size and hydrophobicity. We also note that two putative proteins may combine a FN2 module with a disordered region: FN2a with a Thr/Ser-rich region and FN2b with the Pro/Leu/Gly/Gln/Asn/Arg-rich region, respectively. Based on the reasoning outlined above, the latter combination may constitute a region of the N protein.
Overall, our analysis of the predicted PSCNV proteins suggests that its genome is functionally organized in much the same manner as in the multi-ORF nidoviruses: with the non-structural and structural proteins encoded in the 5’- and 3’- regions, respectively.
Next we sought to determine when PSCNV’s lineage emerged, relative to other nidoviruses. The proteome analysis described above indicates that PSCNV shares the main features characteristic of invertebrate nidoviruses, although it also exhibits distinctive properties indicative of a distant relationship with previously characterized nidoviruses. To resolve very deep branching, we used an outgroup in our analysis, and selected astroviruses for this purpose [23]. Astroviruses [115] and nidoviruses share multi-ORF genome organization, a central role for 3CLpro in polyprotein processing, and similarities in the RdRp domain. Conversely, astroviruses do not encode a HEL1, NiRAN or ZBD, and their 3CLpro is highly divergent. Given the divergent 3CLpro of PSCNV, RdRp remained as the only domain most suitable for phylogeny reconstruction; this domain has been used in many studies on macroevolution of nidoviruses [21, 23, 35, 116].
We performed phylogenetic analysis of the RdRp core region by Bayesian inference (BEAST software, LG+I+G4 model, relaxed clock with uncorrelated log-normal rate distribution). Nidoviruses including PSCNV formed a monophyletic group in >90% of the trees in the analyzed Bayesian sample, with PSCNV being one of the basal branches in the cluster of invertebrate nidoviruses in 88.7% of the trees, basal to either mesoni- and roniviruses (54.7% of the trees), or roniviruses (20.6%), or mesoniviruses (13.4%) (Fig 7 and S17 Fig).
In addition, we built a nidovirus phylogeny without an outgroup (BEAST software, LG+I+G4 model, relaxed clock with uncorrelated log-normal rate distribution), based on a concatenated alignment of five domains conserved in all nidoviruses (3CLpro, NiRAN, RdRp, ZDB, HEL1). Again, PSCNV belonged to the cluster of invertebrate nidoviruses in the majority of trees and was basal to either mesoni- and roniviruses (11.8% of the trees), or roniviruses (83.0%), or mesoniviruses (3.6%).
Is the unique single-ORF genomic organization of PSCNV an ancestral characteristic of nidoviruses, or has it evolved from an ancestral multi-ORF organization? To choose between these alternative scenarios, we need to reconstruct a genomic ORF organization of the most recent common ancestor (MRCA) of nidoviruses. Such reconstruction by orthology, which was used for RdRp-based phylogeny, is not feasible with the current dataset, as none of the open reading frames or their overlaps (with the exception of the ORF1a/ORF1b junction) are conserved in all known multi-ORF nidoviruses.
To address this challenge, we noted that nidoviruses with multi-ORF organization, unlike PSCNV, recurrently use initiation and termination codons to delimit ORF-specific proteins in the 3’ORFs region, indicative of pervasive selection forces that operate in all recognized nidovirus species. Therefore, we reasoned that multi- and single-ORF organizations in nidoviruses could be treated as two alternative discrete states of a single trait (ORF organization), regardless of the complexity of their actual evolutionary relations in the 3’ORFs region and assuming the rate of transition between any two multi-ORF organizations to be extremely high compared to that between single- and multi-ORF organizations. This reasoning allows us to reformulate the question in the framework of ancestral state reconstruction analysis: if each extant nidovirus is characterized by one of the two states of a trait (ORF organization), which state of the trait existed in their MRCA?
To conduct this analysis, we applied the BayesTraits [117] program to the RdRp-based Bayesian sample of phylogenetic trees including the outgroup, which accounts for uncertainty in the phylogeny inference of nidoviruses. The results strongly favored multi-ORF organization of the ancestral nidovirus (Log Bayes Factor (BF) 6.06 and 6.16, when multi-ORF genome organization, or no information about genome organization, were specified as states of the trait for astroviruses, respectively) (S17 Fig). Similarly, strong support (Log BF 4.79) for multi-ORF ancestral organization was obtained when the analysis was conducted based on a phylogeny without an outgroup, reconstructed using five nidovirus-wide conserved domains.
Each of the three main regions of the PSCNV genome is larger than its counterparts in all other nidoviruses (Fig 8A, S1 and S6 Tables). However, the size differences between PSCNV and the next largest nidovirus in each of these regions are smaller than those observed for complete genomes (Fig 8A: 5.7%, 20.6% and 15.6% for ORF1a, ORF1b and 3’ORFs, respectively, vs 22.9% for the genome). This paradoxical observation is due to profound differences in regional size variation among nidoviruses [66] such that different nidoviruses are the next largest to PSCNV for each of the three main regions (S1 Table).
To account for these and other differences in sizes of the three regions while assessing the regional size increases of PSCNV, we employed two measures in addition to the percentage size increase between PSCNV and the next largest nidovirus (see Materials and Methods, formulas D2 and D3 versus formula D1). First, for each genome region, we normalized the size difference between PSCNV and the next largest virus against the difference between the latter and the median-sized virus for that region (formula D2). Second, we checked how much the deviation calculated with formula D2 differs from that expected under a hypothesis that size changes are uniform across the three genome regions, and therefore proportional to genome-wide changes (formula D3). These measures show that, relative to the size variation among known ExoN-positive nidoviruses, the size increase in the ORF1b region was extraordinarily large (D2 = 1270.5% and D3 = 968.1%), while the corresponding increases in the two other regions were modest and smaller than could be expected (18.9% and 14.4% for ORF1a, and 44.3% and 33.7% for 3’ORFs) (Fig 8B, S6 Table).
Virus reproduction requires different viral protein stoichiometries at distinct replicative cycle stages, a challenge for a single-ORF genome theoretically producing equimolar quantities of encoded polypeptides. To this end, all previously described nidoviruses employ -1 PRF to translate ORF1a+ORF1b, in addition to ORF1a alone, to produce two polyproteins from a genomic template: pp1ab and pp1a, respectively [40, 41]. The net result of this mechanism is relatively high expression of the ORF1a- compared to ORF1b-encoded proteins, since PRF occurs at the ORF1a/1b junction in 15–60% of ORF1a translation events. In contrast, proteins encoded in the 3’ORFs region are produced by translation of subgenomic (sg) mRNAs, synthesized on specific minus-strand templates [51–53], which are in turn produced by discontinuous RNA synthesis on genomic templates. Discontinuous minus-strand template synthesis relies on lTRS and bTRS, which are nearly identical, short repeats at sites where RNA synthesis pauses (upstream of 3’ORFs) and resumes (in the 5’-UTR), respectively. Templates of some sg mRNAs may be terminated at bTRS. Both transcription and translation of sg mRNAs provide a means to produce relatively large quantities of structural proteins, compared to non-structural (replicative) proteins, late in the replicative cycle, and to regulate production of accessory proteins. We analysed the PSCNV genome for evidence of such mechanisms.
Finally, we used the PSCNV polyprotein as a query sequence to survey several flatworm species’ transcriptomes in the PlanMine database [119] for the presence of other nidoviruses related to PSCNV. We identified six contig sequences with highly significant similarity to PSCNV indicative of at least two nidoviruses (S18 Fig). These contigs originate from transcriptomes of S. mediterranea (uc_Smed_v2 and ox_Smed_v2 assemblies, two and one contigs, respectively; the latter contig was excluded from consideration due to being almost identical to one of the former contigs) and another planarian species, Planaria torva (dd_Ptor_v3 assembly, three contigs). Translations of the two uc_Smed_v2 contigs of 814 nt and 1839 nt gave hits of >99% aa identity to the very C-terminus of PSCNV polyprotein, indicative of a variant of PSCNV circulating in the same host species (see section above). In contrast, the dd_Ptor_v3 transcriptome included two short contigs (283 nt and 289 nt) with hits to the PSCNV RdRp domain (38 and 48% aa identity) as well as an 8811-nt contig, whose translation in the +1 frame gave 3 discontinuous hits, one to the O-MT domain of the ORF1b-like region (37% aa identity) and two to the 3’ORFs-like region and its FN2b domain (25% and 37% aa identity). These domains are separated by different distances in PSCNV and the 8811-nt contig. It is notable that all three hits from the P. torva contig correspond to its translation in the same frame, uninterrupted by stop-codons, suggesting that ORF1b-like and 3’ORFs-like regions of this putative and divergent virus could also be expressed from a single ORF.
The advent of metagenomics and transcriptomics has greatly accelerated the pace of virus discovery, leading to studies reporting genome sequences of dozens to thousands of new RNA viruses in poorly characterized hosts [35, 36, 79, 120–126]. These developments have substantially advanced our appreciation of RNA virus diversity, and improved our understanding of the mechanisms of its generation [127, 128]. Notwithstanding that sea change, the largest known RNA genomes continue to belong to nidoviruses, as has been the case for 30 years, since the first coronavirus genome of 27 kb was sequenced [14, 21, 78] (Fig 1A).
This study’s transcriptomics-based discovery of PSCNV in planarians reinforces the status of nidoviruses as relative giants among RNA viruses, and also demonstrates that RNA genomes may be substantially larger than previously understood. The discovery of a virus with this large 41.1-kb RNA genome was unexpected in the context of accumulating genomic data on viruses and emerging concepts in the field. Below, we discuss the implications of PSCNV’s distinctive features, and future directions of research.
The PSCNV polyprotein includes distant homologs of all ten domains common to invertebrate nidoviruses, as well as the vertebrate Coronavirinae subfamily [14, 45]. These were identified with high statistical confidence, using an iterative bioinformatics procedure with profile searches at its core. These domains include the definitive nidovirus markers NiRAN and ZBD, and all ten are syntenic between PSCNV and other nidoviruses. Most are located in the ORF1b-like (replicase) region, which also includes four subregions left unannotated (Fig 2). Of these unannotated subregions, one flanked by ZBD and HEL1 may correspond to the regulatory domain 1B, which is uniformly present but poorly conserved in helicases of nidoviruses [48, 49], while the other three may represent domains uniquely acquired by a PSCNV ancestor. Like all characterized invertebrate nidoviruses, but unlike most vertebrate nidoviruses [14, 129], PSCNV does not encode a homolog of an uridylate-specific endonuclease (NendoU) [31]. Accordingly, our rooted RdRp-based phylogenetic analysis assigned PSCNV to a monophyletic clade of invertebrate nidoviruses. Another topologically similar tree was inferred using five nidovirus-wide conserved domains with a dataset that did not include an outgroup. The observed tree topology is also broadly compatible with other observations of this study (see below), and with RdRp-based trees of known nidoviruses produced in other studies [14, 21, 35]. Given that PSCNV infects planarian hosts, consistent placement of this virus in the invertebrate nidovirus clade by different analyses makes biological sense. On the other hand, the precise position of PSCNV in the invertebrate nidovirus clade remains poorly resolved for several reasons, including the highly skewed host representation in the analyzed small sample of 57 nidoviruses, and the large divergence of invertebrate nidoviruses from each other.
The dominant tree topology placed PSCNV in a very long and deeply rooted branch, which has been recognized as a suborder in the pending taxonomic proposal [130]. This is further supported by the presence of the GDD tripeptide in the RdRp C motif (S9 Fig), most common in ssRNA+ viruses other than nidoviruses, which typically (except for the arterivirus Wobbly possum disease virus, WPDV, [81]) have an SDD signature instead [131]. The pronounced divergence of PSCNV is also evident in other conserved protein domains, 3CLpro, NiRAN and ExoN, each of which carries substitutions not observed in other invertebrate or all nidoviruses.
Two prominent replacements in PSCNV 3CLpro are functionally meaningful (S7 Fig). The replacement of the otherwise invariant His by Val in the putative substrate pocket is indicative of a modified P1 substrate specificity for this enzyme, which exhibits a strong preference for Glu or Gln residues in P1 position in most other ssRNA+ viruses, including vertebrate nidoviruses [42, 88–91]. Accordingly, we were unable to identify typical 3CLpro cleavage sites at the expected inter-domain borders in the portion of the PSCNV polyprotein that must be processed by 3CLpro. Furthermore, the nucleophilic catalytic residue of PSCNV’s 3CLpro is Ser, while its counterpart in other characterized invertebrate nidoviruses is Cys. Similar variation of this residue has been described among vertebrate arteri- and toroviruses versus coronaviruses [42, 88–91], with distinct variants being associated with deeply separated virus lineages at the rank of (sub)family. Diversification of the nucleophile residue was also observed in other ssRNA+ viruses that employ 3C(L) proteases [132, 133]. This recurrent Ser-Cys toggling of the catalytic nucleophile in other well-established viral families argues against independent origins of 3CLpros in PSCNV and other nidoviruses, despite their weak sequence similarity.
Besides its exceptionally large genome size, the single-ORF organization of the PSCNV genome is unprecedented for nidoviruses. This single-ORF organization was unexpected, given that multi-ORF organization is conserved across the vast diversity of nidoviruses separated by large evolutionary distances, and infecting vertebrate or invertebrate hosts. In contrast, other large monophyletic groups of ssRNA+ viruses with comparable host ranges (e.g., the order Picornavirales or Flavi-like viruses), include many viruses with either single- or multi-ORF organizations, which intertwine phylogenetically [79, 132, 133].
The use of 3CLpro as the main protease responsible for the release of key RTC subunits from polyproteins would be anticipated to remain essential in the single-ORF PSCNV. In contrast, two other conserved mechanisms of genome expression, ORF1a/1b -1 PRF and discontinuous transcription, might not be expected to operate in this virus, since they are associated with the use of multiple ORFs in nidoviruses. We reasoned otherwise, however, on the grounds that these mechanisms allow differential expression of three functionally different regions of the nidovirus genome, which are also conserved in PSCNV. We located a potential -1 PRF signal in the PSCNV genome. This signal is located at the canonical position observed in other nidoviruses, and could potentially attenuate in-frame translation downstream of the ORF1a-like region in a manner different from a mechanism used by other characterized nidoviruses, but with similar end-products (Fig 9). Such a postulated mechanism is used by encephalomyocarditis virus to attenuate the expression of replicase components in favor of capsid proteins from its main long ORF [134].
Likewise, we obtained several lines of evidence for upregulated transcription of the 3’ORFs-like region as a subgenomic RNA (Fig 10). The products of this region may also be derived from the polyprotein, but are likely required in greater abundance toward the end of the viral replication cycle, and separate expression from sg mRNA would more efficiently address this need. Importantly, no evidence, either bioinformatic or experimental, was obtained for other sg mRNAs, although we cannot exclude their existence. PSCNV’s putative TRSs are exceptionally long for nidoviruses (59 and 57 nt versus typically a dozen nt), perhaps because smaller repeats might emerge in its extraordinarily long genome by chance, interfering with transcription accuracy. Other unknown factors may also contribute to this large TRS repeat size.
The putative leader TRS (lTRS) and body TRS (bTRS), along with their predicted RNA secondary structures, suggest a model for transcriptional regulation of the PSCNV genome. We postulate that during anti-genomic RNA synthesis, the virus RTC unwinds two bTRS hairpins (Fig 10C, top). As a result, the region immediately upstream of the bTRS (yellow in the figure) becomes available for base-pairing with the 5’-terminus of the lTRS (Fig 10C, middle). This interaction will bring the two distant regions of the genome in close proximity, facilitating translocation of the nascent minus-strand from body to leader TRS (Fig 10C, bottom). The latter step is considered routine in the current model of sg RNA synthesis in well-characterized arteriviruses and coronaviruses [51, 135]. However, its mechanistic details are poorly understood and may operate differently among nidovirus families.
Although we cannot exclude the possibility that smaller ORFs are expressed by PSCNV, it seems unlikely that they would contribute substantially to the virus proteome, in line with the apparent inverse relationship between genome size and gene overlap [136]. Rather, such ORFs could be used for regulatory purposes, as in the case of the very small ORF at the border of ORF1a- and ORF1b-like regions, through the PRF mechanism proposed above.
The combined genomic and proteomic characteristics of PSCNV defy the central role of multiple ORFs in the life cycle and evolution of nidoviruses, despite their universal presence in all other nidoviruses [26, 60]. Contrary to conventional wisdom, single-ORF genome expression can involve the synthesis of subgenomic mRNAs. Rather than multi-ORF genome organization, functional constraints linked to the synteny of key replicative enzymes may be the hallmark characteristic of nidoviruses [137].
Most of the domains that we annotated in the PSCNV giant polyprotein are homologs of canonical nidovirus domains. However, we also mapped several unique domains. Below, we discuss possible functions of five small domains, all of which plausibly modulate different aspects of virus-host interaction.
PSCNV encodes a ribonuclease T2 homolog upstream of the putative 3CLpro in the ORF1a-like region (Fig 2). Ribonucleases of the T2 family (RNase T2) are ubiquitous cellular enzymes that non-specifically cleave ssRNA in acidic environments [138]. DNA polydnaviruses and RNA pestiviruses are the only two other virus groups that are known to encode related enzymes [139, 140]. In pestiviruses, the RNase T2 homolog is a domain of secreted glycoprotein Erns found in virions, but dispensable for virus entry [141]. The Erns structure is supported by four disulfide bridges that are formed by eight conserved Cys residues [139]. None of these residues were found in the PSCNV RNase T2 homolog, consistent with its location in the polyprotein region that produces cytoplasmic proteins in other nidoviruses. In polydnaviruses and pestiviruses, the RNase T2 homolog modulates cell toxicity and immunity [139, 140], and a similar role could be considered for the PSCNV RNase T2 homolog. The origin of this domain in PSCNV remains uncertain due to the lack of close homologs in either its host, S. mediterranea, or other cellular and viral species.
Two other unique domains of PSCNV are fibronectin type II (FN2) homologs, protein modules of approximately 40 aa with two conserved disulfide bonds, which are ubiquitous in extracellular proteins of both vertebrates and invertebrates [142, 143]. Because of the low similarity of FN2a and FN2b to each other and other homologs, it is not clear whether they emerged by duplication or were acquired independently. No other known virus encodes an FN2 homolog (although the putative nidovirus identified in P. torva may include an ortholog of FN2b, S18 Fig), suggesting that PSCNV’s FN2 domains function in a unique aspect of its replication cycle. FN2 domains are known to possess collagen-binding activity, and are found in a variety of proteins that bind to and remodel the extracellular matrix [144, 145]. Thus, it is conceivable that these domains might play a role in the shedding or transmission of PSCNV virions. This hypothesis is compatible with the accumulation of PSCNV RNA and particles, presumably virions, in the planarian mucus-secreting cells. Besides FN2 domains, this process might also involve the Thr/Ser-rich region adjacent to FN2a in polyprotein, since Thr-rich and Thr/Ser-rich regions have been implicated in mediating adherence of fungal and bacterial extracellular (glyco) proteins to various substrates [146, 147].
The identification of the ankyrin repeats domain (ANK) in PSCNV is unprecedented and intriguing. In proteins of other origins, the ANK domain is a tandem array of ankyrin repeat motifs (~33 residues each) of variable number and divergence that fold together to form a protein-binding interface [148]. Ankyrin-containing proteins are involved in a wide range of functions in all three domains of cellular life. In viruses described to date, they have been identified exclusively in large DNA viruses with genome sizes ranging from ~100 kb to 2474 kb, the latter of Pandoravirus salinus, the largest viral genome described so far [38, 148–150]. Acquisition of this domain, likely from a planarian host, might have provided a PSCNV ancestor with a mechanism to evade host innate immunity. Notably, according to SmedGB [102] annotation, host proteins SMU15016868 and SMU15005918, whose C-terminal domains are the closest homologs of PSCNV ANK (Fig 6), contain a Rel homology domain (RHD) at their N-termini. This N-RHD-ANK-C domain architecture is typical of the NF-ĸB protein, a precursor of a cellular transcription factor that triggers inflammatory immune responses upon virus infection or other cell stimulation [151]. NF-ĸB is activated for translocation to the nucleus by degradation of its inhibitor, C-terminal ANK domain of NF-ĸB protein or its closely related paralog, IĸB protein [148, 152, 153]. Several large DNA viruses have been shown to encode IĸB-mimicking proteins that prevent NF-ĸB from entering the nucleus in response to the infection, and thus downregulate the host immune response [154, 155]. PSCNV ANK may represent the first example of an IĸB-mimicking protein in RNA viruses, although RNA viruses including nidoviruses can target NF-ĸB protein using other mechanisms [156]. This striking parallel between PSCNV and large DNA viruses blurs the distinction between these viruses regarding how they adapt to hosts [157]. It further highlights the exceptional coding capacity of PSCNV genome among RNA viruses.
The single-ORF organization of PSCNV’s exceptionally large genome is intriguing, but we cannot determine whether this association between genome size and organization is causal or coincidental from observation of a single species. In this respect, determining whether the putative nidovirus we identified in P. torva also employs a single-ORF organization could be illuminating. An evolutionary switch between multi- and single-ORF organizations, regardless of its direction, must be a multi-step process, since it affects many translation regulatory signals. In our study, we used a simple model of this process with two character states within a Bayesian phylogenetic framework, to obtain support for the single-ORF organization of PSCNV emerging from the multi-ORF organization. This approach is apparently not sensitive to the choice of domains used for phylogeny reconstruction, or inclusion of an outgroup. However, given the deep position of the PSCNV lineage in the nidovirus tree, the ambiguous rooting of PSCNV relative to other invertebrate nidovirus families, and PSCNV being the only single-ORF nidovirus known, further analysis of this transition using improved sampling of nidoviruses and their sister clades [35, 36], and more sophisticated models is warranted.
In the few experimentally characterized coronaviruses with genomes of 27–31 kb, the mutation rate is low by RNA virus standards, due to ExoN proofreading activity [34, 158, 159]. This observation is in line with the inverse relationship between genome size and mutation rate in viruses and prokaryotes [160, 161]. Accordingly, we may expect mutation rates to differ among ExoN-containing nidoviruses with different genome sizes, with PSCNV having a particularly low mutation rate. While characterization of mutation rates of PSCNV and other nidoviruses must await future studies, we already note a distinctive similarity between cellular proofreading exonucleases and ExoN of PSCNV, which separates it from its orthologs in other ExoN-positive nidoviruses. Specifically, there is a correlation between the presence of the Zn-finger motif in the exonuclease active site [33, 92] and the genome size of the biological entity encoding the exonuclease: non-PSCNV nidoviruses with genome sizes in the range of 20–34 kb include a Zn-finger embedding catalytic His, while PSCNV and DNA-based entities with genome sizes >41 kb do not (S12 Fig) [162]. Based on these observations, it is plausible that this Zn-finger might limit ExoN's capacity to improve replication fidelity while providing other benefits, and its loss in the PSCNV lineage could have been a factor promoting genome expansion.
Besides the lack of the Zn-finger in ExoN, the reported size increase of the ORF1b-like region in PSCNV relative to other nidoviruses (about 10-fold greater than expected under an assumption of uniform expansion in all genome subregions) is particularly notable in the context of the theoretical framework presented in the introduction. Briefly, expansion of RNA genomes requires escape from the so-called Eigen trap (or Eigen paradox): such genomes are confined to a low-size state, in which low replication fidelity prevents the evolution of larger genomes, which in turn prevents the evolution of greater complexity, which could introduce tools to increase replication fidelity [15]. The three-wave model of genome expansion in nidoviruses notes that the ORF1b region, which encodes the core replicative machinery, appears to play a central role in such constraints. It proposes that a wave of expansion in the ORF1b region of a common ancestor precedes and permits subsequent lineage-specific waves in the ORF1a and 3’ORFs subregions. The wave of expansion in ORF1b involved the acquisition of the ExoN proofreading exonuclease, which permitted further expansion of other subregions due to a reduced mutation rate. Until now, however, the genomes of large nidoviruses (the 20-to-34 kb size range) appeared to have reached a plateau at the low-30 kb range, associated with very little variability in the size of ORF1b among members of this group (6.9-to-8.2 kb). The three-wave model predicts that further genome expansion far beyond 34 kb would require a second cycle of waves, beginning again with ORF1b [66]. The disproportionate increase in PSCNV’s ORF1b-like region is consistent with this prediction. The acquisition of additional, still-uncharacterized domains in this region of the PSCNV genome, as well as the distinctive features of its ExoN domain, may help to explain this “second escape” from the Eigen trap. Further characterization of the PSCNV ExoN and novel ORF1b domains are required, to assess their contribution to replication fidelity and other characteristics that may be critical for faithful replication and expression of exceptionally large RNA genomes.
Our discovery of PSCNV, and analysis of its genome, show that nidoviruses can overcome the ORF1b-size barrier and adopt divergent ORF organizations. If the multi-cycle three-wave model of genome expansion in RNA viruses holds, one would expect that a large expansion of ORF1b, as evident in PSCNV, would permit yet greater expansion of the ORF1a and 3’ORFs regions in other viruses of the PSCNV lineage. Thus, nidoviruses of yet-to-be-sampled hosts might prove to have evolved even larger RNA genomes than that reported here, further decreasing the gap between virus RNA and host DNA genome sizes.
Bioinformatics Materials and Methods are described in S1 Materials and Methods in detail.
The genome sequence of human coronavirus OC43 (GenBank KY014282.1) was used to query two in-house de novo-assembled Schmidtea mediterranea transcriptomes (transcripts assembled from multiple asexual and sexual planarian stocks, designated with txv3.1 and txv3.2 prefixes, respectively) [67] using tblastx (BLAST+ v2.2.29 [163]). With E-value cut-off 10, 25 S. mediterranea transcripts were identified and used in reciprocal BLAST searches against the NCBI NR database. Two nested transcripts, txv3.2-contig_1447 (assembled from sexual planarians, GenBank BK010449) and txv3.1-contig_12746 (assembled from asexual planarians, GenBank BK010448), showed statistically significant similarity to other nidoviruses, which exceeded its similarity to other entries. Sequences of these two transcripts overlap by 23,529 nt with only 7 nt mismatches (0.03%). The larger transcript, txv3.1-contig_12746, was used to search in planarian EST clones [69, 164], which found the following overlapping clones showing >99% nucleotide identity: PL06016B2F06, PL06005B2C04. PL06007A2B12, PL06008B2B03 PL08002B1C07, and PL08001B2B04 (GenBank DN313906.1, DN309834.1, DN310382.1, DN310925.1, HO005314.1, and HO005110.1, respectively). Transcripts txv3.1-contig_12746 and txv3.2-contig_1447, and the six EST clones were assembled into an incomplete putative genome. Conflicts between overlapping sequences were always resolved in favor of the txv3.1-contig_12746 sequence. Fifteen 3’-terminal nt of the reverse complement of txv3.1-contig_12746 (“TATTATGTGATACAC”) and two 3’-terminal nt of HO005314.1 and HO005110.1 (“TG”) were discarded due to their likely technical origin. The assembled sequence contains a stop codon followed by a short untranslated region and a polyadenylated (polyA) tail. The planarian transcriptomes were surveyed again for transcripts with >50 nt overlap at the 5’-end of the incomplete genome by consecutive rounds of nucleotide BLAST. This identified txv3.1-contig_349344 (from asexual planarians; 11,647 nt; 100-nt overlap with txv3.1-contig_12746 with no mismatches; GenBank BK010447) upstream of the original transcripts, and no further extension was achieved with more BLAST iterations. The 5’-end of the genome was then extended using 5’-RACE followed by Sanger sequencing (primers in S2 Table).
Reads from planarian RNA-seq datasets (used to assemble the two transcriptomes described above, and those available from EBI ENA [165]) were mapped to the PSCNV genome sequence by either CLC Genomics Workbench 7, or Bowtie2 version 2.1.0 [166]. Read counts and coverage were estimated using SAMtools 0.1.19 [167], and genome sequence variants were called by BCFtools 1.4 [168].
Freshly prepared RNA from mature sexual planarians was used for cDNA synthesis (iScript, Bio-Rad) or 5’-RACE (RLM-RACE, Ambion) according to manufacturer instructions. Large overlapping amplicons across the PSCNV genome (primers in S2 Table) were amplified by standard Phusion® High-Fidelity DNA polymerase reactions, with 65°C primer annealing temperature and 10 min extension steps.
Colorimetric and fluorescent in situ hybridizations were done following published methods [169]. Digoxigenin (DIG)-labelled PSCNV probes were generated by antisense transcription of the planarian EST clone PL06016B2F06 (GenBank DN313906.1) [69]. Following color development, all samples were cleared in 80% (v/v) glycerol and imaged on a Leica M205A microscope (colorimetric) or a Carl Zeiss LSM710 confocal microscope (fluorescent).
Sexual and asexual planarians originating from the Newmark laboratory were fixed and processed for epoxy (Epon-Araldite) embedding as previously described [170]. For light-microscopic histology, 0.5 μm sections were stained with 1% (w/v) toluidine blue O in 1% (w/v) borax for 30 s at 100°C, and imaged on a Zeiss Axio Observer. For transmission electron microscopy, 50–70 nm sections were collected on copper grids, stained with lead citrate [171] and imaged with an AMT 1600 M CCD camera on a Hitachi H-7000 STEM at 75 kV. Putative virions were seen by TEM in sections from a single worm, which led us to re-examine a collection of 1697 electron micrographs, drawn from 16 additional worms (12 sexuals, four asexuals) from cultures known to harbor PSCNV. All images that included some portion of a mucus cell were chosen for further examination (n = 165); the total number of cells represented cannot be determined without three-dimensional reconstruction from serial sections, which is not practical for such large and irregularly shaped cells. No additional examples of putative viral structures were found among the specimens included in these samples.
For various analyses we used the following databases: PlanMine [119], Smed Unigene [102], scop70_1.75, pdb70_06Sep14 and pfamA_28.0 supplemented with profiles of conserved nidovirus domains [172–174], Uniprot [175], genome sequences representing the current 57 nidovirus species that were delineated by DEmARC [176] and recognized by ICTV on year 2016 [177], NCBI Viral Genomes Resource [178], GenBank [179] and RefSeq [180].
To predict RNA secondary structure and PRF sites we used Mfold web server [181] and Knot-InFrame [182], respectively. Blastn (BLAST+ v2.2.29) [163] was used to identify RNA repeats.
Virus protein sequences were analyzed to predict disordered regions (DisEMBL 1.5 [183]), transmembrane regions (TMHMM v.2.0), secondary structure (Jpred4 [184]), signal peptides (SignalP 4.1 [185]), N-glycosylation sites (NetNGlyc 1.0) and furin cleavage sites (ProP 1.0 [186]). Multiple sequence alignments of RNA virus proteins were generated by the Viralis platform [187]. Protein homology profile-based analyses were assisted with HMMER 3.1 [188], and HH-suite 2.0.16 [189]. To identify sites enriched with amino acid residue, distribution of each residue along polyprotein sequence was assessed using permutation test executed with a custom R script.
To establish homology for ZBD, ExoN, and N-MT, for which top HHsearch hits were under the 95% Probability threshold, we considered several criteria about the source hits: 1) being among the top three for the respective query of a database; 2) being similar to several homologous profiles in two or three databases; 3) residing in the polyprotein position conserved in nidoviruses for the respective domain (S3 Fig, S5 Table); and 4) including most residues that are critical for function of the respective domain. For ZBD, we also observed a statistically significant enrichment in cysteine (Cys) residues (S4 Fig), in line with the coordination of three Zn2+ ions by characterized ZBDs, which involves predominantly Cys and His residues [48, 49].
Size differences between genome regions of PSCNV and nidoviruses (S1 Table) were estimated using three measures, D1, D2, and D3, that accounted for: 1) the region size, D1(region) = (p-M)/M*100%; 2) the region size variation, D2(region) = (p-M)/(M-m)*100%; and 3) the region size variation and genome size increase, D3(region) = D2(region)/D2(genome)*100%, where m and M are median and maximum sizes of the region in ExoN-containing nidoviruses, respectively, and p is the region’s size in PSCNV.
Phylogeny was reconstructed by a Bayesian approach using a set of tools including BEAST 1.8.2 package [190] and ProtTest 3.4 [191] as described in [81]. BayesTraits V2 [117] was used to perform ancestral state reconstruction. Preference for a state at a node was considered statistically significant only if Log BF exceeded 2 [192].
Protein alignments were visualized with the help of ESPript 2.1 [193]. To visualize Bayesian samples of trees, DensiTree.v2.2.1 was used [194]. R was used for visualization [195].
|
10.1371/journal.pcbi.1006475 | Quasiperiodic rhythms of the inferior olive | Inferior olivary activity causes both short-term and long-term changes in cerebellar output underlying motor performance and motor learning. Many of its neurons engage in coherent subthreshold oscillations and are extensively coupled via gap junctions. Studies in reduced preparations suggest that these properties promote rhythmic, synchronized output. However, the interaction of these properties with torrential synaptic inputs in awake behaving animals is not well understood. Here we combine electrophysiological recordings in awake mice with a realistic tissue-scale computational model of the inferior olive to study the relative impact of intrinsic and extrinsic mechanisms governing its activity. Our data and model suggest that if subthreshold oscillations are present in the awake state, the period of these oscillations will be transient and variable. Accordingly, by using different temporal patterns of sensory stimulation, we found that complex spike rhythmicity was readily evoked but limited to short intervals of no more than a few hundred milliseconds and that the periodicity of this rhythmic activity was not fixed but dynamically related to the synaptic input to the inferior olive as well as to motor output. In contrast, in the long-term, the average olivary spiking activity was not affected by the strength and duration of the sensory stimulation, while the level of gap junctional coupling determined the stiffness of the rhythmic activity in the olivary network during its dynamic response to sensory modulation. Thus, interactions between intrinsic properties and extrinsic inputs can explain the variations of spiking activity of olivary neurons, providing a temporal framework for the creation of both the short-term and long-term changes in cerebellar output.
| Activity of the inferior olive, transmitted via climbing fibers to the cerebellum, regulates initiation and amplitude of movements, signals unexpected sensory feedback, and directs cerebellar learning. It is characterized by widespread subthreshold oscillations and synchronization promoted by strong electrotonic coupling. In brain slices, subthreshold oscillations gate which inputs can be transmitted by inferior olivary neurons and which will not—dependent on the phase of the oscillation. We tested whether the subthreshold oscillations had a measurable impact on temporal patterning of climbing fiber activity in intact, awake mice. We did so by recording neural activity of the postsynaptic Purkinje cells, in which complex spike firing faithfully represents climbing fiber activity. For short intervals (<300 ms) many Purkinje cells showed spontaneously rhythmic complex spike activity. However, our experiments designed to evoke conditional responses indicated that complex spikes are not predominantly predicated on stimulus history. Our realistic network model of the inferior olive explains the experimental observations via continuous phase modulations of the subthreshold oscillations under the influence of synaptic fluctuations. We conclude that complex spike activity emerges from a quasiperiodic rhythm that is stabilized by electrotonic coupling between its dendrites, yet dynamically influenced by the status of their synaptic inputs.
| A multitude of behavioral studies leave little doubt that the olivo-cerebellar system organizes appropriate timing in motor behavior [1–3], perceptual function [4–6] and motor learning [7–10]. Furthermore, the role of the inferior olive in motor function is evinced in (permanent and transient) clinical manifestations, such as tremors, resulting from olivary lesions and deficits [11–16]. Although the consequences of olivary dysfunctions are rather clear, the network dynamics producing functional behavior are controversial. At the core of the controversy is the question whether inferior olive cells are oscillating during the awake state and whether these oscillations affect the timing of the inferior olivary output [17–19]. The inferior olive is the sole source of the climbing fibers, the activity of which dictates complex spike firing by cerebellar Purkinje cells (for review, see [20]). Climbing fiber activity is essential for motor coordination, as it contributes to both initiation and learning of movements [8, 10, 21–26], and it may also be involved in sensory processing and regulating more cognitive tasks [27–30]. Understanding the systemic consequences of inferior olivary spiking is therefore of great importance.
The dendritic spines of inferior olivary neurons are grouped in glomeruli, in which they are coupled by numerous gap junctions [10, 31–33], which broadcast the activity state of olivary neurons. Due to their specific set of conductances [34–40], the neurons of the inferior olive can produce subthreshold oscillations (STOs) [41–43]. The occurrence of STOs does not require gap junctions per se [44], but the gap junctions appear to affect the amplitude of STOs and engage larger networks in synchronous oscillation [10, 16, 42]. Both experimental and theoretical studies have demonstrated that STOs may mediate phase-dependent gating where the phase of the STO helps to determine whether excitatory input can or cannot evoke a spike [45, 46]. Indeed, whole cell recordings of olivary neurons in the anesthetized preparation indicate that their STOs can contribute to the firing rhythm [42, 43] and extracellular recordings of Purkinje cells in the cerebellar cortex under anesthesia often show periods of complex spike firing around the typical olivary rhythm of 10 Hz [17, 47–49]. However, several attempts to capture clues to these putative oscillations in the absence of anesthesia have, so far, returned empty handed [19, 50].
It has been shown that in the anesthetized state both the amplitude and phase of the STOs can be altered by synaptic inputs [10]. Inhibitory inputs to the inferior olive originate in the cerebellar nuclei and have broadly distributed terminals onto compact sets of olivary cells [51–53]. Excitatory terminals predominantly originate in the spinal cord and lower brainstem, mainly carrying sensory information, and in the nuclei of the meso-diencephalic junction in the higher brainstem, carrying higher-order input from the cerebral cortex (Fig 1A) [15, 54, 55]. In addition, the inferior olive receives modulating, depolarizing, level-setting inputs from areas like the raphe nuclei [55]. Unlike most other brain regions, the inferior olive is virtually devoid of interneurons [56, 57]. Thus, the long-range projections to the inferior olive in conjunction with presumed STOs and gap junctions jointly determine the activity pattern of the complex spikes in Purkinje cells. How these factors contribute to functional dynamics of the inferior olive in awake mammals remains to be elucidated.
Here, we combine recordings in awake mice–in the presence and absence of gap junctions–with network simulations using a novel inferior olivary model to study the functional relevance of STOs in terms of resonant spikes. We are led to propose a view of inferior olivary function that is more consistent with the interplay between STOs, gap junctions and inputs to the inferior olive. Rather than acting as a strictly periodic metronome, the inferior olive appears more adequately described as a quasiperiodic ratchet, where cycles with variable short-lasting periods erase long-term phase dependencies.
To study the conditions for, and consequences of, rhythmic activity of the inferior olive, we made single-unit recordings of cerebellar Purkinje cells in lobules crus 1 and crus 2 (n = 52 cells in 16 awake mice) in the presence and absence of short-duration (30 ms) whisker air puff stimulation (Fig 1B and 1C). In the absence of sensory stimulation, the complex spikes of 35% of the Purkinje cells (18 out of 52) showed rhythmic activity (Fig 2A–2C; S1 Fig) with a median frequency of 8.5 Hz (inter-quartile range (IQR): 4.7–11.9 Hz). Upon sensory stimulation, 46 out of the 52 cells (88%) showed statistically significant complex spike responses. Of these, 31 (67%) had sensory-induced rhythmicity (Fig 2D–2F), which was a significantly larger proportion than during spontaneous behavior (p = 0.002; Fisher’s exact test). The median frequency of the oscillatory activity following stimulation was 9.1 Hz (IQR: 7.9–13.3 Hz). Hence, the preferred frequencies in the presence and absence of sensory stimulation were similar (p = 0.22; Wilcoxon rank sum test) (Figs 2C, 2F and 3). The duration of the enhanced rhythmicity following stimulation was relatively short in that it lasted no more than 250 ms. With our stringent Z-score criterion (>3), only a single neuron showed 3 consecutive significant peaks in the peri-stimulus time histogram (PSTH). The minimum inter-complex spike interval (ICSI) across cells was around 50 ms, putatively representing the refractory period. We conclude that complex spikes also display rhythmicity in awake behaving mice, and that sensory stimulation can amplify these resonances in periods of a few hundred milliseconds, even though stimulation is not required for the occurrence of rhythmicity per se.
The pattern of rhythmic complex spike responses that was apparent for a couple of hundred milliseconds after a particular air puff stimulus repeated itself in a stable manner across the 1,000 trials (applied at 0.25 Hz) during which we recorded (Fig 4A and 4B). For example, the level of rhythmicity of the first 100 trials was not significantly different from that during the last 100 trials (comparing spike counts in first PSTH peak, p = 0.824; χ2 test, or latency to first spike, p = 0.727, t test). This strongly indicates that there is–in a substantial fraction of the Purkinje cells–persistent oscillatory gating of the probability for complex spikes after a sensory stimulus resulting in time intervals (“windows of opportunity”) during which complex spikes preferentially occur (Fig 4B–4E). These windows of opportunity become even more apparent when sorting the trials on the basis of response latency: the first complex spikes with a long latency following the stimulus align with the second spikes of the short latency responses. Similarly, there are trials during which complex spikes appear only at the third cycle (Fig 4C, seen as a steeper rise around trial no. 650). The occurrence of spikes during later cycles, not predicated on prior spikes, argues against refractory periods or rebound spiking as the sole explanations for such rhythmic firing [58] and highlight the putative existence of network-wide coherent oscillations.
Since sensory stimulation of the whiskers can trigger a reflexive whisker protraction [59–61] and complex spike firing is known to correlate with the amplitude of this protraction [61], we examined the relation between periodic complex spike firing and whisker protraction. To this end, we further analyzed an existing dataset of simultaneously recorded Purkinje cells and whisker movements during 0.5 Hz air puff stimulation of the ipsilateral whisker pad. In line with our previous findings [61], trials during which a single complex spike occurred within 100 ms of whisker pad stimulation had on average a slightly, but significantly stronger protraction (from 6.1 ± 5.4° to 6.8 ± 5.3° (medians ± IQR), n = 35 Purkinje cells, p = 0.033, Wilcoxon-matched pairs test after Benjamini-Hochberg correction; S2A–S2C Fig). Our new analysis revealed that also the occurrence of a second complex spike was correlated with a stronger whisker protraction. This could be observed as a second period of increased protraction during trials with two complex spikes. When compared to the increase in trials with a single complex spike, this second protraction was highly significant (p < 0.001, Wilcoxon-matched pairs test after Benjamini-Hochberg correction; S2D Fig). The second complex spike was unlikely a mere reflection of stronger protraction following the first complex spike, as there was no difference in whisker protraction between the trials that had a complex spike during the first, but not the second 100 ms after stimulus onset, and the trials with the opposite pattern (a complex spike during the second, but not the first 100 ms; p = 0.980, Wilcoxon-matched pairs test after Benjamini-Hochberg correction). The rhythmic firing pattern of complex spikes was thus reflected in the behavioral output of mice.
The existence of windows of opportunity for complex spike activity is compatible with the assumption of an underlying STO, and cannot solely be explained by rebound activity without invoking circuit-wide extrinsic mechanisms. To test the implications of assuming olivary STOs, we proceeded to reproduce a detailed network with a tissue-scale computer model of the inferior olive neuropil. The model is constituted by 200 biophysically plausible model cells [40, 46, 62] embedded in a topographically arranged 3D-grid (Fig 5A–5C). It has the scale of a sheet of olivary neurons of about 10% of the murine principal olivary nucleus (cf. [63]). The model was designed to test hypotheses about the interaction between intrinsic parameters of olivary neurons, such as STOs and gap junctional coupling, and extrinsic parameters including synaptic inputs during the generation of complex spike patterns.
Each neuron in the model is composed of a somatic, an axonal and a dendritic compartment, each endowed with a particular set of conductances, including a somatic low threshold Ca2+ channel (Cav3.1; T-type), a dendritic high threshold Ca2+ channel (Cav2.1; P/Q-type) and a dendritic Ca2+-activated K+ channel, chiefly regulating STO amplitudes, while a somatic HCN channel partially determines the STO period (Fig 5B; see also Methods). The dendrites of each neuron are connected to the dendrites of, on average, eight nearby neighbors (within a radius of three nodes in the grid, representing a patch of about 400 μm x 400 μm of the murine inferior olive), simulating anisotropic and local gap junctional coupling (Fig 5C). As the inferior olive itself, our model has boundaries which have impact on local connectivity characteristics, such as the clustering coefficient, though these did not have significant impact on the average firing rate between edge and center cells (p = 0.812, comparing edge and center cells, Kolmogorov-Smirnov test; S3 Fig). The coupling coefficient between model cells varied between 0 and 10%, as reported for experimental data [45, 64, 65].
Sensory input was implemented as excitatory synaptic input, simulating the whisker signals originating from the sensory trigeminal nuclei that were synchronously delivered to a subset of model neurons. Additionally, a “contextual input” was implemented as a combination of inhibitory feedback from the cerebellar nuclei and a level setting modulating input (Figs 1A and 5A). This contextual input is modeled after an Ohrstein-Uhlenbeck process, essentially a random exploration with a decay parameter that imposes a well-defined mean yet with controllable temporal correlations (see Methods). The amplitude of the contextual input drives the firing rate of the model neurons, which we set around 1 Hz (S4 Fig), corresponding to what has been observed in vivo [28, 43, 66]. Thus, our model network recapitulates at least part of the neural behavior observed in vivo due to biophysically plausible settings of intrinsic conductances, gap junctional coupling and synaptic inputs.
Whether a model neuron at rest displays STOs or not is largely determined by its channel conductances. Activation of somatic T-type Ca2+ channels can trigger dendritic Ca2+-dependent K+ channels that can induce Ih, which in turn can again activate T-type Ca2+ channels, and so forth. This cyclic pattern can cause STOs that could occasionally produce spikes (Fig 5B). In our model, the conductance parameters were randomized (within limits, see Methods) so as to obtain an approximate 1:3 ratio of oscillating to non-oscillating cells (S5 Fig) guided by proportions observed in vivo [43]. Sensitivity analysis with smaller ratios (down to 1:5) did not qualitatively alter the results (data and analyses scripts are available online in https://osf.io/9hmpy/).
In the absence of contextual input, model neurons were relatively silent, but when triggered by sensory input, as occurred in our behavioral data (Figs 2 and 4), STOs synchronized by gap junctions would occur for two or three cycles (Fig 6A and 6B). Our network model confirms that gap junctional coupling can broaden the distribution of STO frequencies and that even non-oscillating cells may, when coupled, collectively act as oscillators (S6 Fig) [67]. Adding contextual input to the model network can lead to more spontaneous spiking in between two sensory stimuli. Compared to the situation in the absence of contextual input, the STOs are much less prominent and the post-spike reverberation is even shorter (Fig 6C). Accordingly, despite the significant levels of correlation in the contextual input (10%), the periods between oscillations are more variable due to the interaction of the noisy current and the phase response properties of the network. In addition, in the presence of contextual input our model could readily reproduce the appearance of preferred time windows for spiking upon sensory stimulation as observed in vivo (Fig 6D–6F, cf. Fig 4). This was particularly true for the model cells that directly received sensory inputs (Fig 6G–6H). Moreover, the observed rhythmicity in model cells as observed in their STO activity was in tune with that of the auto-correlogram (Fig 6D–6I) in that the timing of the STOs and that of the spiking were closely correlated (cf. Fig 5B). It should be noted though that model cells adjacent to cells directly receiving sensory input showed only a minor effect of stimulation. Thus, even though the gap junction currents in the model were chosen as the ceiling physiological value for the coupling coefficient (≤10%) [45, 67], these currents alone were not enough to trigger spikes in neighboring cells.
Both directly stimulated model cells and those receiving only contextual input exhibited phase preferences, seen in the spike-triggered membrane potential average as well as in the spike-triggered average of the input currents (Fig 6G–6I). Spike-triggered averages of membrane potentials for any cell showed depolarization followed by hyperpolarization. In contrast, trials in which no spike was generated showed a depolarization just before the occurrence of the input. Similarly, the average of the input showed a long-lived phase preference, not only for a hyperpolarization before the spike, but also a preference for a depolarization in the previous peak of the STO, more than 100 ms earlier. These results are in line in vitro experiments under dynamic clamp and noisy input [68, 69]. Likewise, the model indicates that for short durations STOs can induce clear phase dependencies for spiking, which fades under the variation of period durations dependent on the trial-specific contextual input (as seen in our data).
Depolarizing sensory input delivered onto a subset of the model cells can reset the STO phase in oscillatory cells and create resonant transients in others (Fig 7A and 7B; see also S7 Fig on the appearance of rebound firing). If a second stimulus is delivered during this short-lasting transient, the response probability is increased. As in most cells with resonant short-lasting dynamics, inputs delivered during different phases can cause phase advances or delays. Hyperpolarization advanced the phase between 0−π and delayed the phase between π−2π, whereas depolarization had roughly the opposite effect, in addition to phase advancements with spikes in later cycles between π−2π (Fig 7C–7E). Thus, there is a mutual influence of synaptic inputs and STOs on periodicity. While STOs can lead to phase-dependent gating, synaptic input can either modulate or reset the phase of the STOs, generating variable periods that range between 40–160 ms for the chosen amplitude of the contextual input (Fig 8; S6 Fig).
The only means of settling the question about the prevalence of STOs in awake and behaving mice would be intracellular recordings of inferior olivary neurons, which remains a daunting experimental challenge. We therefore looked for a less invasive method that could read out, from indirect and infrequent complex spikes, the presence or absence of STOs. We have developed one such paradigm inspired by auditory studies [70, 71] using a rhythmic gallop stimulus that we first applied to the network model (Fig 7F). In the gallop paradigm, stimuli are applied in quick succession with alternating intervals, comparable to the putative period of the underlying oscillation. Enough stimuli should be applied such that after multiple presentations the stimuli sample a uniform distribution of phases. In the context of auditory stimuli, the standard gallop experiment involves different tones and is used to test perceptual separation of auditory streams. Such rhythmic stimuli can help indicate resonances or physical limitations of the system, and distinguish across possible models for this separation (such as in neural resonance theory [71]). One possible mechanism of auditory stream separation is an underlying oscillatory process which resets in certain phases and is less responsive in others. According to the in vitro inferior olive literature [41–43] this behavior is to be expected, and hence, such a stimulus can help distinguish underlying processes.
If spikes are modulated by an oscillatory process, the presence of spikes on a short interval should be able to predict, in the next interval, the absence or presence of spikes. Indeed, if the underlying process producing spikes has oscillatory components and a relatively stable period, the probability of spikes in each interval is systematically different, which would appear as asymmetric ratios of response in the different intervals (Fig 7F). This can be inspected as the length of the empty and filled vertical bars representing ratio of probabilities of spiking for long or short stimulation intervals (Fig 7G). Thus, if the period of the STO rhythm would be regular and cause phase-dependent gating, complex spike responses following each stimulus interval are expected to show preferences for the short or long window of stimulation; these preferences were indeed observed (Fig 7G, left). However, this clear phase dependency only appears in the noiseless model scenario. After adding a moderate amount of contextual input, this dependency washes off, rendering the responses in the two windows more symmetric (Fig 7G, right), with only a few cells (5/200) displaying significant ratio differences (tested against bootstrap with shuffled spikes).
In line with the experimental in vivo data (e.g., Figs 2 and 4), the olivary spike rhythmicity in the network model was steadily present over longer periods, and for a wide range of contextual input parameters (S5 Fig). In addition, it also comprised, as in the experimental data, variations in frequency and amplitude during shorter epochs (Figs 8 and S6). Analysis of the network parameters indicates that these latter variations in oscillatory behavior can be readily understood by their sensitivity to both the amplitude (parameter 'sigma') and kinetics (parameter 'tau', temporal decay) of the contextual input. Indeed, because of the underlying Ornstein-Uhlenbeck process, the generation of contextual input converges to a specified mean and standard deviation, but in short intervals the statistics including the average network STO frequency can drift considerably (Figs 6C and 8). Since relatively small differences in oscillation parameters such as frequency can accumulate, they can swiftly overrule longer-term dependencies created by periodically resetting stimuli, as an analysis of phase distributions shows (Fig 9). Thus, based upon the similar outcomes of the network model and in vivo experiments, we are led to propose that (1) the STOs in the inferior olive may well contribute to the continuous generation of short-lasting patterns of complex spikes in awake behaving animals, and that (2) the synaptic input to the inferior olive may modify the main parameters of these STOs. Note that in the absence of input, periodic rhythmic behavior should be the default behavior of oscillating cells. Thus, in all likelihood, even if the inferior olive oscillates endemically, sustained but variable input should induce highly contextual spike responses to variable periods and render the olivary responses quasiperiodic, rather than regularly periodic as observed in reduced preparations.
In line with in vivo whole cell recordings made under anesthesia [10, 43] our awake data support the possibility that the moment of spiking may be related to the phase of olivary STOs, especially during the period of several hundred ms following stimulation (Figs 2 and 4). As discussed above, a gallop stimulus would expose such an oscillatory process underlying the response probabilities. Four idealized scenarios about the expected results can be constructed, as follows: first, one can start with a complete absence of STOs, which would result in a response probability unrelated to stimulus intervals; second, it could be that there were STOs, but no phase-dependent firing (to be expected if the STO amplitude is small), which would also lead to complex spike firing irrespective of stimulus intervals; third, there could be STOs, but each stimulus would evoke a phase-reset, which again, would not lead to interval dependencies; and fourth, there could be STOs in combination with phase-dependent gating, which would result in a clear dependency of complex spike firing on the previous interval length (Fig 7G, left panel). It should be noted that the large majority of studies on inferior olivary physiology, especially in reduced preparations, found evidence for the fourth situation (STOs + phase-dependent gating) [41, 43, 67].
To study whether phase-dependent gating in conjunction with an underlying oscillatory process could shape complex spike response timing in vivo we applied both a 250 vs. 400 ms and a 250 vs. 300 ms gallop stimulation using air puffs to the whiskers. Using only trials with a CS in the previous trial to calculate the ratio of responses ('conditional firing') a slight bias could be observed in the 250 vs. 400 ms paradigm (Fig 10A) and to a lesser extent in the 250 vs. 300 ms (Fig 10B). Analysis including all trials ('non-conditional') is included in S8 Fig and shows no significant bias for any of the cells tested. Hence, our in vivo data are in line with the results from the network model subjected to synaptic noise, and show that the timing of complex spike responses to sensory stimulation is biased but not strongly determined by STOs.
Our experimental data provided evidence for phase-dependent complex spike firing during brief intervals, but gallop stimulation did not expose a strong impact of STOs on complex spike response probabilities. Therefore, we sought an alternative approach to study the impact–if any–of STOs on complex spike firing in vivo. We reasoned that, if a sensory stimulus triggers a complex spike response with a certain probability, higher stimulation frequency should result in a proportional increase in complex spike firing. In particular, stimulus frequencies that would be in phase with the underlying STO would be expected to show signs of resonance and result in disproportionally increased complex spike firing. However, over periods of tens of seconds the complex spike frequency was resilient to varying the stimulus frequency between 1 and 4 Hz (linear regression = -0.02; R2 = 0.1) (Fig 10C) and did not show signs of resonance with any of the stimulus frequencies, as there were no frequencies at which the complex spike firing was substantially increased. Only a very high rate of sensory stimulation (10 Hz), commensurable with the average duration of windows of opportunity, could induce a mild increase in complex spike firing frequency, albeit at the cost of a highly reduced response probability (average increase: 71 ± 64% corresponding to an average increase from 1.12 Hz to 1.92 Hz; n = 5; p < 0.05; paired t test). This examination indicates that the average complex spike frequency is robust and stiff to modulation over longer time periods, imposing a hard limit on the frequency with which complex spikes can respond to sensory stimuli, confirming recent reports on complex spike homeostasis [72].
As stimulus triggered resonances were not observed at any of the stimulation frequencies, we turned to a more sensitive measure for the detection of oscillatory components in complex spike firing. We developed a statistical model that extrapolated from frequencies inferred through inter-complex spike intervals and stimulus triggered histograms (Fig 11A and 11B). We reasoned that phase-dependent gating would imply that the interval between the last complex spike before and the first one after sensory stimulation aligns to the preferred frequency. In contrast, if sensory stimulation would typically evoke a phase reset, as suggested by our network model (Fig 7), no such relation would be found. The method was applied only to Purkinje cells with highly rhythmic complex spike firing. For each of those, we calculated their preferred frequency in the absence (Fig 11C) or presence of sensory stimulation (Fig 11D). We used that frequency to construct statistical models representing idealized extremes of phase-dependent (oscillatory) and -independent (uniform) responses. For the oscillatory component we employed an oscillatory gating model, where the timing of the first complex spike after stimulus onset would be in-phase with the ongoing oscillation. This model was contrasted to a linear response model in which sensory stimulus could evoke a complex spike independent of the moment of the last complex spike before that stimulus, apart from a refractory period. For each Purkinje cell, we compared the distribution of the intervals between the last complex spike before and the first complex spike after stimulus onset with the predicted distributions based on the linear model, the oscillatory model and nine intermediate models, mixing linear and oscillatory components with different relative weights (Fig 11A–11E). For the two extreme models as well as for the nine intermediate models we calculated a goodness-of-fit per Purkinje cell. Overall, when using these relatively long periods (300 ms), the linear model was superior to the oscillatory model, although a contribution of the oscillatory model could often improve the goodness-of-fit (Fig 11F–11H). Despite the apparent failure of the oscillatory model to fit the data, the data did show an oscillatory profile for many of the cells (Fig 11E). This lends support to our observations that short-lived, but reliable, oscillations are apparent in complex spike timing, although they have little impact on the timing or probability of sensory triggered CS responses.
Apart from the STOs, extensive gap junctional coupling between dendrites is a second defining feature of the cyto-architecture of the inferior olive [31, 33, 73]. Absence of these gap junctions leads to relatively mild, but present deficits in reflex-like behavior and learning thereof [10, 74]. We analyzed the inter-complex spike interval times in Purkinje cells of mutant mice that lack the Gjd2 (Cx36) protein and are hence unable to form functional gap junctions in their inferior olive [69]. In line with the predictions made by our network model (Figs 5H and S6), the absence of gap junctions did not quench rhythmic complex spike firing during spontaneous activity (Fig 12A). In fact, the fraction of Purkinje cells showing significant rhythmicity in the Gjd2 KO mice was larger than that in the wild-type littermates (Gjd2 KO: 38 out of 65 Purkinje cells (58%) vs. WT: 15 out of 46 Purkinje cells (33%); p = 0.0118; Fisher’s exact test), with their average rhythmicity being significantly stronger (p = 0.003; Kolmogorov-Smirnov test), measured by Z-scores of side peaks (Fig 12B). Indeed, the variation in oscillatory frequencies across Purkinje cells of the mutants was significantly less than that in their wild-type littermates in that the latency to peak times per Purkinje cell were less variable (p = 0.0431; Mann-Whitney test; Fig 12C). This latter finding is at first sight contradictory to our findings in the network model, where we show that gap junctions promote more uniform firing rates through increased synchrony between neurons (Fig 5H). These simulations were run in the absence of synaptic input, though. Addition of contextual input also creates more variability in the wild type cells (S6B Fig). As the lack of gap junctions increases cell excitability [10, 69], it is likely that synaptic input has a larger impact in the absence of gap junctions, leaving less room for inter-cell heterogeneity. Overall, removal of gap junctions affected the temporal and spatial dynamics by increasing the stereotypical rhythmicity of complex spike firing.
We made paired recordings of Purkinje cells in awake mice to study the temporal relations of their complex spikes during spontaneous activity. The cell pairs were recorded with two electrodes randomly placed in a grid of 8 x 4, with 300 μm between electrode centers. For each pair of simultaneously recorded Purkinje cells, we made a cross-correlogram. The median number of complex spikes in the reference cell used for these cross-correlograms was 827 (range: 74–2174). Cell pairs showed coherent activity in that they could show a central peak and/or a side peak in their cross-correlogram (Fig 13A–13C). The side peaks could appear at different latencies, similar to the range observed in auto-correlograms of single Purkinje cells (cf. Fig 2B). Moreover, Purkinje cell pairs that did not produce signs of synchronous spiking in the center peak could still produce an “echo” in the side peak after 50–150 ms. Counter-intuitively, cross-correlograms of Purkinje cell pairs of the wild type mice showed less often a significant center peak than those of Gjd2 KOs (WT: 51 out of 96 pairs (53%; N = 4 mice); Gjd2 KO: 44 out of 61 pairs (72%; N = 7 mice); p = 0.0305; Fisher’s exact test). In line with the more stereotypic firing observed in single cells in the absence of gap junctions (Fig 12), the strength of the center peak was on average enhanced in the mutants (Z-scores of significant center peaks (median ± IQR): WT: 3.47 ± 1.82; Gjd2 KO: 5.75 ± 5.58; p = 0.0002; Mann-Whitney test) (Fig 13D–13F). Instead, the side peak of Gjd2 KO Purkinje cell pairs was not stronger than that of WTs (Z-scores of significant side peaks (median ± IQR): WT: 3.01 ± 0.89; Gjd2 KO: 3.04 ± 1.52; p = 0.194; Mann-Whitney test), leading to a lower ratio between center and side peak (mean ± SEM: WT: 90.70 ± 5.17%; Gjd2 KO: 72.86 ± 6.52%; p = 0.036; t = 2.143; df = 67; t test) (Fig 13E and 13F). Interestingly, the occurrence of side peaks in Purkinje cell pairs was unidirectional in approximately half the cell pairs (WT: 47 out of 82 pairs with at least one side peak (57%); Gjd2 KO: 25 out of 47 pairs (53%); p = 0.714; Fisher’s exact test), which means that one of the neurons of a pair was leading the other, but not vice versa. As this was consistent in the Gjd2 KO as well as the WT Purkinje cells, these data could reflect traveling waves across the inferior olive, which, however, must have extrinsic sources [44, 75]. Thus, the paired recordings are compatible with the findings highlighted above in that the presence of coupling can affect the coherence of STOs for short periods up to a few hundred milliseconds, while leaving the window for later correlated events open.
Given the major impact of complex spikes on Purkinje cell processing and motor behavior [10, 11, 25, 76, 77], resolving the mechanisms underlying their timing is critical to understand the role of the olivo-cerebellar system in motor coordination and learning [20, 22, 24, 78]. Synchronized subthreshold oscillations (STOs) of olivary neurons have been suggested to contribute to the formation of temporal patterning of complex spike firing, but most of the evidence for STOs has been obtained in decerebrate or anesthetized animals in vivo or, even more indirectly, in vitro [10, 41–43, 45, 47, 79, but see 80]. Our experiments and model were designed to inquire the existence of an in vivo STO, and on its ability to display phase-dependent responses in the awake brain. We evaluated rhythmic complex spike firing in behaving animals responding to peripheral stimuli and investigated their match with simulations of a tissue-scale model of the inferior olivary network. The model and data would be consistent with STOs whose period can be readily adjusted upon synaptic fluctuations from other brain regions, an effect that is consistent with the known response properties of the inferior olive [46, 67].
During spontaneous activity, Purkinje cells generally fire a complex spike roughly once a second, but this frequency can be increased to about 10 Hz by systemically applying drugs, like harmaline, which directly affect conductances mediating STOs in the inferior olive [41, 81]. Since these drugs also induce tremorgenic movements beating at similar frequencies, it has been proposed that the inferior olive may serve as a temporal framework for motor coordination [11, 82]. This oscillatory firing behavior of the olivary neurons may mirror limb resonant properties and act as an inverse controller, for example by dampening the dynamics of the muscles involved [83].
In line with previous recordings [22–24, 28, 61, 72, 76, 78, 84], the current data indicate that only a small fraction of Purkinje neurons respond to sensory stimulation with a complex spike response probability larger than 50%. This probability falls substantially with increasing frequency of stimulation, as the overall spike frequency only marginally increases to high frequency stimulation. Even after applying different temporal patterns of sensory stimulation for longer epochs, we observed no substantial deviation from the stereotypic 1 Hz firing rate. Moreover, it should be noted that even if the frequency of underlying oscillations has bearing on the pattern of responses of the gallop stimuli, conditional dependencies should be expected for most STO frequencies, unless the ratio of the interval of gallop and the STO period has no remainder. Given the seemingly consistent frequencies predicted by PSTH's and autocorrelograms of single cells (Figs 2 and 3, but also seen in cross correlograms, as in Fig 13), we chose gallop intervals with periods commensurate with a representative frequency of 8 Hz, each of which should sample different phases in the oscillation. If at all present, we should have observed conditional dependencies on at least a few cells.
In our study, complex spikes remain as unpredictable as ever. Thus, regulatory mechanisms keep the complex spike rate relatively stable over longer time periods [72]. No resonance is exhibited, irrespective of an enduring powerful sensory stimulus in a variety of frequencies. Save few exceptions, the presence of a complex spike in an interval is compensated by the absence in another. Thus, it looks as if the complex spikes rearrange themselves in time in order to keep close to its proverbial 1 Hz frequency. It remains to be shown to what extent the mechanisms involved are intrinsic (cell-dependent) and/or extrinsic (network-dependent). A possible candidate for setting the overall level of excitability through intrinsic mechanisms is given by Ca2+-activated Cl− channels, which are prominently expressed in olivary neurons along with Ca2+-dependent BK and SK K+ channels [85, 86]. In addition, the olivo-cerebellar module itself could partly impose this regulation [86–88]. Indeed, the long-term dynamics within the closed olivo-cortico-nuclear loop may well exert homeostatic control, given that increases in complex spikes lead to enhanced inhibition of the inferior olive via the cerebellar nuclei [20]. The impact of such a network mechanism may even be more prominent when changes in synchrony are taken into account [89]. We propose that a closed-loop experiment conducted while imaging from a wide field, producing stimulation as a function of the degree of complex spike synchrony, could tease out conditional complex spike probabilities. Increasing our capability of predicting complex spikes is instrumental to elucidate the control of inferior olivary firing.
The existence of temporal windows of opportunity for complex spike responses following sensory stimulation highlighted a potential impact of STOs on conditional complex spike gating [10, 41, 43, 90]. Indeed, autocorrelogram peaks correlated well with interspike intervals following stimulation, arguing for an underlying rhythm. Complex spikes could appear in a particular window even when they were not preceded by a complex spike in a previous window during a single trial, arguing against a prominent role of refractory periods in creating rhythmic complex spike responses. Comparing actual firing patterns with statistical models mixing linear or oscillatory interval distributions indicated a potential impact of oscillations. The mild impact of the oscillatory component on explaining the data may in part depend on the assumption that cells have a well-defined frequency. In other words, a variable rebound time could offset the phase response by a couple of milliseconds, reducing the contribution of the oscillatory model, though phase preferences due to prior spikes may still occur (i.e., Fig 11E). Our biophysical model suggests that fluctuating inputs, such as those mediating inhibition from the cerebellar nuclei or those relaying depolarizing modulation from the raphe nuclei [91, 92], may induce variations in the oscillation period on a cycle-by-cycle basis (Figs 8 and 9). As these contextual inputs are absent or suppressed in decerebrate or anesthetized preparations, as well as in vitro, they may also explain why many earlier studies systematically encountered cells with well-defined STO frequencies [10, 41, 43, 45, 47, 79, 93, 94]. In the network model, in which we mimicked the contextual input as an Ornstein-Uhlenbeck process with local variations but no long-term drifts of the mean [95], the results agree well with the experimental observations in terms of synchronous firing, phase shifts, cross-correlogram peaks and side peaks, as well as overall firing frequency. Indeed, the absence of resonant responses over longer time windows and the inconsistency of individual olivary cells to fire on every trial or cycle indicate that the STOs are not regularly periodic, but rather quasiperiodic, while still being synchronous.
Even though several lines of evidence suggest a role for STOs (see above), we did not observe an unequivocal, significant conditional dependence of complex spikes in the gallop paradigm, as expected by a noiseless model. How can a system with rhythmic responses at least partially fail to be phase modulated by such periodic stimuli? An attractive alternative explanation for rhythmicity might be the occurrence of high-threshold Cav2.1 P/Q-type Ca2+ channel-dependent rebound spikes (S7 Fig) [12, 62]. If impulse-like input to the olive can evoke a spike, and if this spike produces a rebound spike some tens of milliseconds later, this could explain the alignment between the PSTHs and cross-correlograms. However, this argument cannot explain stimulus triggered spikes at the second or third window of opportunity, without an earlier spike as observed in Fig 4. As the occurrence of the rebound spike is predicated on a prior spike, a spike in the second or third window without a prior spike cannot be explained by the rebound spiking phenomenon, at least not within the same cell. In other words, the spikes happening exclusively in the second (or third) window of opportunity cannot be the result of a previous spike in the same cell, unless there is a shared rhythm in the network. It is also conceivable that strong hyperpolarization that is synchronized with the complex spike rhythm could promote reverberating firing, but this is an extrinsic mechanism, discussed below. As they stand, our findings do not support the idea that the post-spike hyperpolarization is a prerequisite for the complex spike pattern observed. Multiple windows of opportunity could, according to our model, be enhanced by transient oscillations induced by resets relayed by gap junctions to the local olivary circuit.
Apart from the almost complete absence of interneurons, the presence of STOs and the exclusive projection to the cerebellum, the abundance of dendro-dendritic gap junctions is another defining feature of the inferior olive. The absence of these gap junctions does not lead to gross motor deficits, but prevents proper acquisition and execution of more challenging tasks [10, 16, 74], which is in line with the relatively minor impact found on complex spike activity in Gjd2 KO mice. At first sight, the effects of deleting gap junctions seem counterintuitive. Synchronous and rhythmic patterns are exacerbated, rather than diminished by the loss of gap junctions. However, the side peak of the auto-correlogram is significantly squashed, indicating that the gap junctions have a role in the increased coherence of the upcoming oscillation. Gap junctions do not only facilitate synchronization of coupled neurons, they also lower their excitability by increasing the membrane resistance [69]. Together, this results in less direct coupling, observed as reduced synchrony of direct neighbors [16, 96], and increased responsiveness to synaptic input. This leads to more long-range coherence and as a consequence gap junction networks may act as a “noise filter”, promoting short-range quorum-voting on phase (a term coined by Winfree [97]). This effect is visible in our model as spikes are most likely to occur when excitation follows inhibition (Fig 6H). This is in line with the finding that complex spikes of nearby Purkinje cells have a preference to fire together [72, 98, 99]. This concept also agrees with the possibility that coupled olivary neurons may control movements by dampening the dynamics of the muscles involved at an appropriate level [83, 100], as both the resonances and movement oscillations increase shortly after sensory stimulation in Gjd2 KO mice [16].
Network resonances are a pervasive feature of brain circuits and they can be induced by subthreshold oscillations of particular cell types [101, 102]. In addition to the autochthonous dynamics of the inferior olive, reverberating loops through the circuit could help explain some features of complex spike firing, including the occurrence of complex spike doublets and side peaks in cross-correlograms. Such phenomena could be explained by "network echoes", where complex spikes in one cycle would induce complex spikes in the next cycle [87, 103, 104]. The most obvious candidate loop to produce is that via the cerebellum and the nuclei of the meso-diencephalic junction [55, 105]. The output of the inferior olive is mainly directed via exceptionally strong synapses to the Purkinje cells [106]. These Purkinje cells in turn inhibit neurons of the cerebellar nuclei that can show rebound firing after a period of inhibition [88, 107]. This rebound activity can excite the inferior olive again via a disynaptic connection via the nuclei of the meso-diencephalic junction. While an isolated complex spike is unlikely to evoke such a rebound activity, a larger group of Purkinje cells could be successful in doing so [20, 107, 108]. The travel time for this loop (around 50–100 ms) has been indirectly assessed in the awake preparation [8, 10, 26], and corresponds to the latency of the rebound firing in the cerebellar nuclei under anesthesia [87, 88, 107, 109]. This implies that the travel times for the entire loop would be in the same order as found for the preferred frequencies of complex spike firing. Other, more elaborate loops involving for instance the forebrain may also exist [110] and could play an additional role in shaping complex spike patterns.
A putative impact of reverberating loops on rebound activity could be a network phenomenon, as the impact of an isolated complex spike may not be sufficient to trigger this loop. This is in line with the reduced “echo” in the cross-correlograms of the Gjd2 KO mice and enhanced doublets following lesions of the nucleo-olivary tract as occurs in olivary hypertrophy [111]. Taken together, rebound spiking, STOs and reverberating loops all seem to promote in a cooperative manner complex spike rhythmicity at a time scale of about 200 ms. Through modeling, we found that not only the state of the inferior olivary oscillations determines which inputs are transmitted, but that these inputs also determine the state of the network. Thus, inputs from both the cerebellum and the cerebrum determine the probability of complex spike responses on a cycle-by-cycle basis providing a quasiperiodic framework to align synchronous groups. This sharply contrasts with a view in which the inferior olive is a clock with regular periodicity. A circuit-wide understanding of cerebellar resonances on the basis of such a mechanism could open a novel pathway to explore the cerebellar gating by other brain regions.
The combination of delayed gap junctions and delayed inhibition, as found in the olivo-cerebellar loop [104, 112], can affect oscillatory behavior [113]. The interplay between STOs and delayed inhibition is therefore also relevant for other neural circuitries, for instance for creating filter settings for the perception of sounds with specific oscillatory properties [114–116] or orchestrating rhythmic movements as shown in the present study (see also [117]).
Well-coordinated movement sequences are not timed rigidly; they must be enacted flexibly and contextually. In order to catch a ball, or a prey, or to perform any other appropriately timed movement, it is essential to fine-tune the duration and onsets of multiple coordinated output systems. An inferior olive that responds contextually to time varying input by advancing and delaying cycles does not act as a rigid clock or metronome, but more contextually, as a ratchet-pole system, with the frequency of 'clicks' of the ratchet reflecting the recent history of applied torque. The properties we have encountered in this study are consistent with a 'ratchet-like' dynamics for the inferior olive, which integrates time-varying stimulus in a phase-dependent manner. According to this view, the inferior olive responds to all inputs (sensory and otherwise), by producing phase changes that are informative about the recent history of input, and dictate the appearance of coherent complex spike waves arriving at the cerebellar cortex.
All experimental procedures were approved a priori by an independent animal ethical committee (DEC-Consult, Soest, The Netherlands) as required under Dutch law.
Experiments were performed on 16 adult (9 males and 7 females of 25 ± 14 weeks old) homozygous Gjd2tm1Kwi (Gjd2 KO, formerly known as Cx36 KO mice [10]) mice which were compared to 15 wild-type littermates (8 males and 7 females of 26 ± 13 weeks old; means ± sd). The generation of these mice has been described previously [118]. The data described in S2 Fig originated from previously published recordings in 35 wild-type mice [61]. All mice had a C57BL6/J background. The mice received a magnetic pedestal that was attached to the skull above bregma using Optibond adhesive (Kerr Corporation, Orange, CA) and a craniotomy of the occipital bone above lobules crus 1 and crus 2. The surgery was performed under isoflurane anesthesia (2–4% V/V in O2). Post-surgical pain was treated with 5 mg/kg carprofen (“Rimadyl”, Pfizer, New York, NY) and 1 μg lidocaine (Braun, Meisingen, Germany). Mice were habituated during 2 daily sessions of 30–60 min.
Extracellular recordings of Purkinje cells were made in the cerebellar lobules crus 1 and 2 of awake mice as described previously [28]. Briefly, an 8 x 4 matrix of quartz-platinum electrodes (2–4 MΩ; Thomas Recording, Giessen, Germany) was used to make recordings that were amplified and digitized at 24 kHz using an RZ2 BioAmp processor (Tucker-Davis Technologies, Alachua, FL). The signals were analyzed offline with SpikeTrain (Neurasmus, Rotterdam, The Netherlands) using a digital band-pass filter (30–6,000 Hz). Complex spikes were recognized based on their waveform consisting of an initial spike followed by one or more spikelets. A recording was accepted as that of a single Purkinje cell when a discernible pause of at least 8 ms in simple spike firing followed the complex spikes and when the complex spikes were of similar shape and amplitude throughout the recording.
Sensory stimulation was applied as air puffs of 20 psi and 25 ms duration directed at the whisker pad ipsilateral to the side of recording. The stimuli were given in trains of 100 or 360 pulses either at regular or alternating intervals. During a recording, trains with different stimulus intervals were played in a random sequence.
Whisker videos were made from above using a bright LED panel as backlight (λ = 640 nm) at a frame rate of 1,000 Hz (480 x 500 pixels using an A504k camera from Basler Vision Technologies, Ahrensburg, Germany). The whiskers were not trimmed or cut. Whisker movements were tracked offline as described previously [119] using a method based on the BIOTACT Whisker Tracking Tool [120]. We used the average angle of all trackable large facial whiskers for further quantification of whisker behavior.
Of each Purkinje cell we computed the probability density function (PDF) of both its complex spike autocorrelogram and its distribution of intervals between consecutive complex spikes (inter-complex spike intervals (ICSIs)). PDFs were calculated with an Epanechnikov kernel (with finite support) with a width of 10 ms. In order to exclude stimulus-induced alterations in complex spike firing, complex spikes detected between 20 and 200 ms after a stimulus were omitted from this phase of the analysis. PDFs were calculated from 0 up till 500 ms. The peak in the ICSI PDF was considered as the “preferred ICSI interval” and its strength was expressed as the Z-score by dividing the peak value by the standard deviation of the PDF. To understand the impact of Purkinje cell with little or no preference for specific ICSI intervals on the analysis, we chose to look both at the Purkinje cells with high and low Z-scores. Thus, we grouped Purkinje cells into high and low level Z-scores, using a threshold of 3.
Air puff stimulations frequently triggered double complex spike response peaks, suggestive of an underlying inferior olivary oscillation. For further analysis of the conditional responses, an estimate of the putative inferior olivary frequency was derived from the interval between these two response peaks. First, it was established for each Purkinje cell whether two peaks were present in the PSTH. To this end, we set a threshold for each of these two peaks. For the first peak, this was calculated by reshuffling the ICSIs over the recording followed by calculating a stimulus-triggered pseudo-PSTH, repeating this procedure 10,000 times and selecting the 99% upper-bound. We considered the first response peak to be significant if it crossed the upper-bound uninterruptedly for at least 10 ms. Since the second response peak typically is much smaller than the first one, we calculated a new threshold for the second peak by excluding the time-window for the first response peak. This window was set from the time of the stimulus until where the response probability drops to the average response frequency, the response frequency as expected if stimuli do not trigger complex spikes, following the significant ‘first’ responsive peak. In 5 out of 98 Purkinje cells, the PDF of the response rate between clear peaks remained above the average response rate, in which case we used the time point where the amplitude drop in the PDF was more than twice the difference between upper bound and average response probability. The rest of the bootstrap method was identical to that for the first response peak. Only peaks up to 0.5 s after the stimulus were included in the population analysis.
In order to test whether the phase of the inferior olivary oscillations affected the complex spike response probability, we compared the complex spike intervals over an Air puff for each stimulus that triggered a complex spike. To this end, we analyzed the recordings of 25 Purkinje cells (10 WT and 15 Gjd2 KO) that were measured previously in crus 1 and crus 2 of awake, adult mice. We included only Purkinje cells that displayed clear oscillatory complex spike firing indicated by the display of a secondary complex spike response peak, as evaluated according to the bootstrap method described above, and/or significant peaks in the ICSI histogram. Only stable recordings covering at least 500 stimuli at frequencies below 1 Hz were considered for this analysis.
For each recording we compared two idealized statistical models of the observed ICSI distributions: an oscillatory model showing phase-dependent spiking and stable olivary oscillation frequencies and a uniform model lacking phase-dependencies. For the oscillatory model, we created complex spike probability functions for the pre-stimulus interval (-300 to 0 ms) based on the oscillatory period established either for the ICSI distribution or from the interval between the two complex spike response peaks. We fitted a sine wave with the observed frequency, having its peak at the moment of the first complex spike in the stimulus response window (20–200 ms after the air puff) and derived spike probability levels during the pre-stimulus interval from these fits, with the trough representing zero probability. Frequency and amplitude of every cycle were kept constant for the whole recording. In the uniform model, we calculated the pre-stimulus spiking probability with a uniform distribution based on the complex spike frequency of each Purkinje cell. We did include a refractory period, being the shortest ICSI observed for each recording, to reflect the inability of consecutive complex spikes to occur with a very short time interval. Refractory periods were comparable between mutants and wild types; 49 ± 15 ms for Gjd2 KO cells and 50 ± 20 ms for WT cells. Subsequently, we constructed compound fits consisting of linear summations of the two models. One extreme was the oscillatory model and the other the uniform model and we considered nine intermediate combinations (e.g., 0.3 x the oscillatory model + 0.7 x the uniform model). Every compound fit was run for 10,000 times. The goodness of fit was computed as the absolute differences of every single run of the model with the actual ICSI distribution.
The model networks used here are comprised of a topographical grid of 200 coupled cells, in a 10x10x2 lattice arrangement, which may resemble an area of about 400 μm x 400 μm of the inferior olive, for instance, the rostral portion of the dorsal lamella of the principle olive of the mouse. It is available online at https://github.com/MRIO/OliveTree, branch 'Warnaar'. For instructions on how to run the model and reproduce analysis, check README_Warnaar.txt'.
Each cell within these networks was modelled according to the single cell model described in [46], which is an elaboration of a previous model [62] with an added axon and modified fast sodium channel. Equations are provided in the appendix of that publication at (https://doi.org/10.1371/journal.pcbi.1002814.s002), and can be checked in the MATLAB functions IOcell and createDefaultNeurons in the codebase. The model includes three compartments (soma, dendrite, axon hillock) with 12 conductances. In addition to the ionic mechanisms, the dendrite of the model cell has a Ca2+ concentration state variable, which is related to the intrusion through the Cav2.1 channels. The main ionic conductances responsible for the oscillation are the somatic T-type Ca2+ and the Ca2+-activated K+ (SK) channels present in the dendritic compartment. The crucial parameters governing the emergence of subthreshold oscillations are randomized, reflecting the experimental facts that about one third of the cells oscillate endemically (in vivo) with intrinsic variations in oscillatory frequencies [43]. The behavior of the STO of the model cells in our network as a function of their parameters, for models with and without gap junctions, are included in S5 and S6 Figs. Cell parameters are found in S1 Table.
Connectivity is created with the function 'createW.m' in the MATLAB codebase. Briefly, cells within a specified radius of each other were connected according to a probability function such as to ensure the specified mean degree in the network (n = 8), chosen to resemble the observed connectivity distributions reported in the literature. The connectivity parameters (distance and average connection probability) were chosen to match experimental values (radius ≤ 120 μm) and average connection probability (~8 neighbors). The procedure to obtain connectivity is as follows. First, pairwise distances between all cells are calculated. Then, a binary adjacency matrix is created by thresholding those distances within a specified radius. Thereafter, we assign a random number between 0 and 1 to each link from a uniform distribution. Finally, this matrix is made binary by comparing each entry with a probability so that the average number of connections approximates a given mean connectivity. This binary adjacency matrix is then multiplied by the mean gap junction conductance parameter. Finally, gap junction conductance values are then randomized by a uniform jittering of the conductance by 10% of their original value.
The conductance of gap junctions was normalized with a saturating factor by difference of potential between the neighboring cells, according to [62] based on findings from [121] with the following function:
gc¯(∆V)=gc(0.8e(−∆V2/100)+0.2)
FORMULA 1
Where ΔV is the voltage difference between the connected cells, gc is the nominal coupling and gc¯ is the effective coupling.
Two inputs are given to the model, one emulating the sensory input from whisker pad stimulation and the other representing a stimulus-independent background reflecting diverse excitatory and inhibitory inputs to the inferior olive. The latter consisted of a continuous stochastic process with known mean and standard deviation with a relaxation parameter following the Ornstein-Uhlenbeck process [95], succinctly described underneath.
Only one subset of cells in the center of the network (40% of the cells in a mask spanning a radius of 3 cells from the center of the network) representing efferent arborization, receives the “sensory input”, with “sensory” currents being delivered to the soma of modelled cells (gAMPA = 0.15 mS/cm2). “Sensory input” was modeled according to O’Donnel et al. [122]. The mask is represented in Fig 5A.
The cells of the inferior olivary network most likely share input sources due to overlapping arborizations of efferent projections [123]. To represent both shared and independent input, we have modeled the current source in each cell as having an independent process and a shared process, with a mix parameter (alpha) of 10% input correlation shared by all the cells in the network. This level of correlation leads to a coherent background oscillation in the cells of the network, which is exacerbated in the presence of gap junctions (S5 Fig).
Ornstein-Uhlenbeck (OU) is a noise process that ensures that the mean current delivered is well behaved and that the integral of delivered current over time converges to a constant value [95]. The OU current is a good approximation for synaptic inputs originating in a large number of uncorrelated sources, where synaptic events are generated randomly and each event decays with a given rate (τ). We use a recursive implementation according to the following recursive formula:
ηi(t+1)=ηi(n,t)*exp(−δ/τ)+(1/τ)(μ−ηi(t))+σ*√δ*ξi
FORMULA 2
Where ηi(t) is the noise amplitude of neuron i at time t. The noise process is parameterized by τ,σ,μ where τ represents the synaptic decay time constant, δ is the integration step time for our forward Euler integrator, σ is the standard deviation of the noise process and μ is its mean. The random draw from a Gaussian distribution at every time step is represented by ξi.
Neurons in the inferior olive receive broad arborizations, leading to input correlations across nearby neurons. In our model this is represented via a mixture of an independent process for each neuron nindependent and a shared process, nall, common to all the neurons in the network, parameterized by a mixing parameter α, called 'noise correlation':
ni(t)=α*nindependent+(1−α)*nall(t)
FORMULA 3
Simulation results throughout the article come from simulations with noise use an α where neurons share 10% of their noise input. Reported results are qualitatively robust to changes in this value (S5 Fig).
To examine the dependence of network dynamics on the characteristics of the incoming input, we computed the 200 neuron network sweeping a grid of the main input parameters (τ,σ,μ,α) of the Ornstein-Uhlenbeck noise process. The network response in terms of STO frequency, population firing rates, proportion of firing neurons was analyzed with respect to a grid of input parameters. For comparability of statistics and reproducibility of results, all results displayed in this article were obtained from a single random seed. We have tested the network with multiple seeds and the results are qualitatively indistinguishable.
The parameters of the Ornstein-Uhlenbeck process were tuned such that the network emulating the wildtype network (with gap junctions, WT) produced an average frequency of 1 Hz and more than 95% of the model cells fire at least once every 5 seconds (the parameter space for the network responses including STO, population firing rate and proportion of cells that fire within 3s is found in S5B Fig). The parameters to achieve these criteria are dependent on the total leak through the gap junctions. There are multiple methods to compensate the absent leak in the gapless network. In the present case, the network without gap junctions has been tuned to produce the same firing frequency as the network with gaps by increasing the membrane leak currents from 0.010 to 0.013 mS/cm2. This results in a similar excitability but slightly lower STO frequency in the “mutant”. The average firing rate behavior of the network shows a linear relationship with the standard deviation of the OU process (S4 Fig). For the present network with balanced connectivity and a single gap conductance of 0.04 mS/cm2, the Ornstein-Uhlenbeck parameters are (τ,σ,μ,α),μ = −0.6 pA/cm2,σ = 0.6 pA/cm2 and τ = 20 ms. τ is a decay parameter that represents the synaptic decay times expected for olivary inputs, in this case chosen to emulate dendritic GABA according to Devor and Yarom [124].
Both synchrony and instantaneous frequency were estimated on the basis of a novel phase transformation of the membrane potential, which is more robust than the standard Hilbert transform, and can produce a linear phase response to the non-linear shape of the subthreshold oscillations [125]. This transformation improves the estimation of the momentary phase and compensates for the fact that ionic mechanisms induce different rates of membrane potential change at different phases of the oscillation. This phases analysis was conducted with the DAMOCO toolbox [126]. From the instantaneous phase, the instantaneous frequency is simply the inverse of the first order finite difference of phase. Synchrony across cells is estimated with the Kuramoto order parameter (K):
K(t)=|1N∑ei(Ψ(t)−ϕn(t))|
FORMULA 4
Where ϕn is the phase of each neuron, N is the number of neurons and Ψ is the phase average of all oscillators.
To estimate a phase response curve of the stimulated neurons, first a “sensory stimulus” is delivered at a phase known to produce an action potential (and resetting). The location of the first peak after stimulation is recorded. Subsequently, eight more simulations receive another stimulus, with same parameters as the resetting stimulus, but at different phases (at incremental intervals of 2π/8). The effect of that stimulation (delay or advance) on the next peak is recorded as a phase delta. Results are plotted in Fig 7A–7E.
|
10.1371/journal.pcbi.1002040 | Gene Expression in the Rodent Brain is Associated with Its Regional Connectivity | The putative link between gene expression of brain regions and their neural connectivity patterns is a fundamental question in neuroscience. Here this question is addressed in the first large scale study of a prototypical mammalian rodent brain, using a combination of rat brain regional connectivity data with gene expression of the mouse brain. Remarkably, even though this study uses data from two different rodent species (due to the data limitations), we still find that the connectivity of the majority of brain regions is highly predictable from their gene expression levels–the outgoing (incoming) connectivity is successfully predicted for 73% (56%) of brain regions, with an overall fairly marked accuracy level of 0.79 (0.83). Many genes are found to play a part in predicting both the incoming and outgoing connectivity (241 out of the 500 top selected genes, p-value<1e-5). Reassuringly, the genes previously known from the literature to be involved in axon guidance do carry significant information about regional brain connectivity. Surveying the genes known to be associated with the pathogenesis of several brain disorders, we find that those associated with schizophrenia, autism and attention deficit disorder are the most highly enriched in the connectivity-related genes identified here. Finally, we find that the profile of functional annotation groups that are associated with regional connectivity in the rodent is significantly correlated with the annotation profile of genes previously found to determine neural connectivity in C. elegans (Pearson correlation of 0.24, p<1e-6 for the outgoing connections and 0.27, p<1e-5 for the incoming). Overall, the association between connectivity and gene expression in a specific extant rodent species' brain is likely to be even stronger than found here, given the limitations of current data.
| Brain connectivity is believed to be associated with gene expression levels in the developing and the adult animal. Recently, this association has been explored in two model animals: the worm C. elegans at the level of single neurons; and the mouse, where specific subpopulations of neurons in the hippocampus were studied. Inspired by these studies, we set out to generalize their scope and examine the possibility of using gene expression signatures to predict regional connectivity in the whole rodent brain. Our results show a higher degree of association between connectivity and expression than shown before, and key genes are identified that are highly predictive of brain connectivity.
| Genes play a major role in the formation of the nervous system and in its continuous function. They specify neuronal cell types, help destine neurons into defined neural circuits, and provide important cues determining their connectivity [1]–[2]. Inspired by Roger Sperry's classical chemo-affinity hypothesis that states that neuronal wiring takes place by selective attachment guided by specific molecular identifiers, a large array of studies have described various gene families that are involved in axonal guidance and in determining their specific targets (see [3]–[7] for reviews). Another central paradigm has posited that a central driving force in determining synaptic connectivity are activity-dependent mechanisms, by which synapses are formed between neurons whose firing tends to be correlated in a self-organizing Hebbian manner (see [8]–[9] for reviews). A third paradigm has recently emphasized the potential role of random axonal outgrowth and location-dependent competition in establishing connectivity [10]. These paradigms are obviously not mutually exclusive and are likely to concur concomitantly, and quantifying the extent of association between gene expression and connectivity may provide global constraints on their relative contribution.
A few recent studies have examined the association between gene expression and connectivity on the neuronal level in the worm C. elegans, by studying the relation between a neuron's gene expression and its connectivity to and from other neurons. C. elegans offers a unique opportunity to perform such an investigation, as it is currently the only model organism for which both a large fraction of its synaptic connectivity and gene expression are known on an individual neuronal level. While [11]–[12] have set to predict the formation of synapses in the worm based on the expression pattern of the pertaining genes [13], aimed to do so while additionally considering their spatial proximity. Overall, these studies have shown that: (1) neuronal gene expression does contain significant information about its connectivity, but the predictive power it entails is rather moderate, at least with the current available data, and (2) it is still possible to use this information to identify genes that potentially play part in determining the neural architecture, on a genome scale. Here we aim to significantly go beyond these earlier studies and to investigate the fundamental relation between gene expression and connectivity in a mammalian brain, and to study it at the level of connectivity between different brain regions.
A recent study [14] has used the mouse brain data of the Allen mouse brain atlas (ABA) [15]–[16] and the accompanying spatial gene expression correlation map tool to study gene expression patterns within the CA1 field. Multiple observations have been made to suggest that gene expression associations between CA1 regions and other sub-cortical brain regions are indicative of direct or indirect projections to or from distinct spatial domains of the CA1 field. In another study [17], it was shown that a factorization of the hippocampus volume by the local gene expression levels leads to a spatial grouping that agrees with the known patterns of differential connectivity. Inspired by these studies, we set out here to generalize their scope and examine the possibility of using gene expression signatures to predict regional connectivity in a mammalian brain. Presently, as there is no adequate regional gene expression and connectivity data available for a single mammalian species, we therefore fuse data from two species: brain wiring data for the rat brain and regional gene expression data from the mouse brain, to study their relation in a prototypical rodent brain. The rat connectivity atlas [18] available online (http://brancusi.usc.edu/bkms/) provides connectivity information for the anatomical structures of the rat. The Allen mouse brain atlas (ABA) [15]–[16] provides gene expression images for the adult mouse brain. Although gene expression during embryogenesis and development would have ideally been more befitting, this data is still lacking on the large scale. Yet, major components of synapses (such as synaptic boutons and spines) are undergoing continuous turnover and are actively maintained during adult life (e.g., [19]–[20]), which raises the possibility that information on synaptic connectivity may also be manifested in adult gene expression. This, coupled with the success of the earlier studies in the worm in predicting connectivity from adult gene expression [11]–[13], has motivated us to explore this possibility in depth here. The Allen atlas also provides a mapping between image regions and brain structures. By matching the brain structures of the rat connectivity map and the brain structures of the mouse brain we are able to construct a combined gene expression/connectivity atlas of the rodent brain (Materials and Methods). Using the combined atlas we find that gene expression levels in different brain regions contain considerable predictive information on their connectivity (interestingly, more than the level found in previous studies in the worm) and identify the genes and functional annotations whose expression is most predictive. Obviously some errors may be introduced in this mapping due to inter-species variations in connectivity and expression levels that may hinder the statistical significance of our results. Hence, importantly, the results presented here are likely to be a lower bound on the actual magnitude of the relationship between gene expression and regional brain connectivity. In parallel to our study, another group demonstrated evidence for a correlation between gene expression and connectivity in the rodent brain by using similar sources for gene expression of the mouse brain and rat connectivity maps [21].
The combined expression/connectivity atlas of the rodent brain contains 176 brain regions. Each is associated here with three signatures. The first signature is a gene expression vector of size 20,936 obtained from processing the Allen Brain Atlas. The other two signatures specify brain region connectivity: one encodes the outgoing connections from each region (Efferent connectivity), and the other encodes the incoming connections to each region (Afferent connectivity). Connectivity is obtained from the BAMS atlas [18] using the nomenclature of [22], assuming that connections that are not reported do not exist [23].
Similarly to [11] we study the connectivity information contained in gene expression by considering both prediction accuracy and the expression/connectivity correlation. Prediction accuracy measures the extent to which connectivity is predicted given the gene expression data. It is estimated for each region separately via a standard cross validation procedure. The correlation between gene expression and connectivity is a global index that measures how similar are the distances between regions in connectivity terms to their distances in expression terms, for all regions at once. On top of predictability and correlation, we also bring further support to our results by examining the enrichment of connectivity-related predicted genes in various disorders that are believed to be related to alterations in brain connectivity.
Connectivity prediction ability was studied using a linear SVM classifier (see Materials and Methods). We first obtain results for outgoing connections: In order to examine each region only once, we consider those 146 regions that do not contain other regions, i.e., regions that are leaves of the regional hierarchy of ABA (Figure 1(a)). Additionally, all regions that have less than 5 outgoing connections are discarded, resulting in a set of 44 regions A1,…,A44. We then fix a region Ai and consider the expression signatures of all other leaf regions B1,…, B146. At each of the 5 cross-validation iterations, we train a classifier using 4/5 of the regions and obtain a mapping between gene expression of the target region Bj and the existence of an outgoing connection from Ai to Bj. The learned map is then applied to the remaining 1/5 regions in order to obtain predictions on the test data, unseen during training. These 5 iterations produce predictions to all regions B1,…,B146, and the overall prediction performance is quantified using the standard Area Under Curve (AUC) measure. A p-value is assigned to each region by performing a standard permutation test (see Materials and Methods). An analogous procedure was applied for predicting incoming connections.
The resulting prediction ability for outgoing connectivity is significant (p<0.05) for 32 out of the 44 regions (73%). The average AUC was 0.74 over all regions, and 0.79 for the significant regions. Significant prediction ability was observed also for the incoming connections. There are 57 regions that are not contained in other regions and which have at least 5 incoming connections. Out of these regions 32 (56%) have statistically significant (p<0.05) prediction accuracy. The average AUC is 0.73 for all the 57 regions and 0.83 for the 32 significant ones. The results for the prediction experiments (combining incoming and outgoing) are provided in Table S1, and the significant regions are portrayed in Figure 1(b,c). The outgoing and the incoming experiments share 35 brain regions that have at least 5 outgoing and 5 incoming connections, out of which 15 are successfully predicted in both incoming and outgoing sets.
In several regions of the hierarchy, the BAMS atlas is more detailed than the Allen Brain Atlas, therefore there are known BAMS connections that exist between substructures of the given leafs of the Allen Brain Atlas. In our study, such connections are eliminated since they arise from localized substructures that might have specific gene expression profiles, not necessarily matching that of the larger structures. This conservative approach is in line with the incompleteness of BAMS [24], i.e., the conservative connectivity map is geared to allow for more missing links rather than erroneously including spurious ones. However, for completeness, we also report the results obtained when taking a more liberal approach, which propagates links between BAMS substructures up to regions that have ABA analogs, are also presented in Table S1. This ‘liberal’ connectivity matrix contains well studied links that do not appear in the conservative connectivity map, such as the projection from the dentate gyrus to Ammon's horn. In this experiment too, there are many regions for which the connectivity prediction is significantly above chance −49% of the efferent regions and 58% of the afferent regions show significant predictability. While this is somewhat lower than the results obtained using the conservative connectivity matrix (73% and 56%), this drop in performance is expected due to the addition of regions with only few known connections, and the specificity of the connections to and from sub-regions that go beyond the resolution of the maps.
Several other alternative choices were also made in order to demonstrate the robustness of the experimental design and results, and are also depicted in Table S1. When choosing a threshold of 10 connections instead of 5, the average AUC obtained is similar; When replacing the SVM algorithm with the ensemble algorithm gentleBoost [25], results remain similar or slightly improve. Interestingly, when using the Nearest Neighbor algorithm as the classifier, the results somewhat deteriorate, suggesting that the connectivity predicting patterns are not metrically related in a trivial manner. To provide further support to the validity of the prediction method in the face of missing connectivity data (as BAMS is probably not comprehensive [24]), we also run simulations on synthetic connectivity graphs where one can carefully control the level of missing information (Materials and Methods). The results show that it is possible to have significantly correct predictions even if a large majority of the connections are missing.
Supplementary Table S2 shows predictions for individual connections that were obtained by aggregating the results over individual brain regions. Shown are both connections which are known to exist (230 outgoing and 207 incoming) and newly predicted connections that currently have not been reported in the literature (416 outgoing and 390 incoming), obtained with the natural SVM detection threshold at zero.
Using the connectivity prediction paradigm described above we employ a zero-norm SVM feature selection procedure (see Materials and Methods) to select the genes whose expression levels are most predictive of connectivity. For each region, the top 500 genes (out of 20,936) are selected, and a list of the 500 most frequently selected genes over all regions is formed, one for predicting the outgoing and one for predicting the incoming connections (Materials and Methods). As can be seen in Figure 2, many genes are selected repeatedly over the different regions in each of the outgoing and the incoming experiments.
Remarkably, 241 genes (out of the 500 most selected) are shared by both the outgoing and the incoming lists (the expected number of shared genes according to the hypergeometric distribution is approximately 12, p<1e-5). The lists of genes selected are reported in Supplementary Table S3. Thus, in parallel to our finding that the connectivity of many brain regions is predictable on both the outgoing and incoming side, we also find that many genes are informative of both the incoming and outgoing connectivity. Since the outgoing predictions are based on the gene expression vectors of the target regions, and the incoming predictions are based on those of the source regions, the two sets of experiments use two halves of the data and the intersection of the two gene lists is not a statistical necessity. As a control test, we check whether those genes that show the highest region-to-regions variability are those that get selected as predictive. If this were the case, one could attribute their selection to the increased variability and not to their ability to predict connectivity. To this end, all genes were ranked according to their region-to-regions variability, measured as the mean distance from the average expression value, and put in equally sized bins. Then, the intersection of each bin with the two lists of the most informative genes was computed. As is evident from Figure 3 the selected connectivity-predicting genes are not necessarily those genes with the highest region-to-region variability and the two sets are inherently different. Apparently, a large amount of variability points to the influence of other factors that are not related to connectivity.
Having such lists gives as an opportunity to estimate the level of involvement of neural connectivity alterations in different brain disorders. To this end, we assembled from the literature lists of the top 100 genes that have been associated with each disorder examined, and quantified the number of (both efferent and afferent) connectivity related genes in each such list – the higher this number is, the more likely it is that connectivity alterations may play a role in the pathogenesis of the said disorder (Materials and Methods). Ranked by this measure (supp Table S4), the disorders we examined are (from the most associated to the least associated) Autism, attention deficit disorder, Schizophrenia, anxiety disorder, major depression, Parkinson's disease, bipolar disorder, Alzheimer's disease, obesity, glioma, and cardiovascular diseases. This ranking order fits fairly well with the prominent role ascribed to neuronal connectivity alterations in schizophrenia and autism. To obtain a rough estimate of the role of neuronal connectivity in these disorders as perceived in the literature, we recorded the number of web documents reported by the Google search engine that contained both the name of the disorder and the term “neuronal connectivity” and compared the latter to the connectivity-involvement measure we computed above. The web frequency count, as collected between March 28 and March 30, 2010 (supp Table S4), shows that the disorders examined can be divided to three main groups - high (schizophrenia and autism), low (obesity, glioma and cardiovascular) and medium level (the remaining ones). Quite remarkably, the high-frequency group has the highest mean of predicted connectivity related genes (15), followed by the medium level group (11.8) and then the low level one (3). These differences are statistically significant. Notably, one disorder originally belonging to the medium-level group (attention deficit disorder) has a similar number of connectivity-related genes as those in the high level group, possibly suggesting a potential role of connectivity alterations in its pathogenesis. A recent comprehensive meta-analysis of genes associated with Schizophrenia [21], listing 75 Schizophrenia related genes, has provided us an opportunity to examine our pertaining predictions in light of this gene association data. A random intersection of 500 genes would include less than 1.8 genes on average. The list of incoming connectivity genes intersects this list by 7 genes (p<0.002), and the outgoing lists intersects it by 4 genes (p = 0.1).
To estimate the global correlation (i.e., across all regions) between gene expression and connectivity we represent each of these two information sources as a square matrix that depicts the correlation in either gene expression or the connectivity profiles between every two regions (see Materials and Methods). Three 146×146 matrices are hence obtained: one based on similarity in gene expression and two for the similarity in incoming and outgoing connectivity profiles. Following previous work [11], [26], we compute the Pearson correlation between the lower triangular part of the matrices to evaluate correlation between data sources. The correlation between gene expression and outgoing connectivity is 0.26 (p<1e-7, empirical p<1e-4) and the one to outgoing connectivity is 0.23 (p<1e-6, empirical p<1e-4), showing again that there is a robust and significant relation between gene expression and regional brain connectivity.
We then employ such a correlation test to evaluate the connectivity information content of four different sets of genes of interest (Materials and Methods): an axon guidance list based on [27], a compilation of presynaptic genes [28], the list of predictive genes identified in C. elegans [11], and the list of genes that were found to bear an embryologic imprint [29]. The first two lists represent known gene sets that given their axonal/synaptic function are potentially, likely to be involved in determining and maintaining brain connectivity. The Third set has been previously found to be predictive in the worm. The last set might be correlated with connectivity since developmental relationships are sometimes mirrored in connectivity [30]. For each of these four sets we compute the 146×146 expression similarity matrix and examine its correlation to the original connectivity matrix obtained between the 146 different leaf regions. The results are presented in Table 1. Quite remarkably, only the genes known to be associated with axon guidance from the literature are significantly correlated with the brain regional connectivity and a significant correlation is absent for the three other groups. It is intriguing to find such an association between axon guidance and connectivity-related genes, even when looking at adult expression data. In addition to the four sets of genes, Table 1 presents the p-values of the connectivity correlation test applied to the lists of genes that were collected for each of the medical conditions mentioned above. These results are similar to the expected ranking, with various brain disorder genes showing an inter-region distribution that is significantly correlated with brain connectivity.
To further study which gene annotation groups are informative with respect to connectivity, we also applied the correlation test to individual functional annotation groups. For each of 1,616 annotation groups in DAVID [31] that were at least partly expressed in the 20,936 genes at hand, we compute its 146×146 expression regional expression similarity matrix and examine its correlation to the original connectivity matrix. The results are summarized in Table 2 for the outgoing connectivity and Table 3 for the incoming connectivity, and are given in full in Table S5. Reassuringly, the top listed functional annotation groups are generally mostly related to neurogenesis, cell-cell signaling, synaptic activity and axonogenesis (both tables), and to neurotransmitter binding and receptor activity on the incoming side. There were 276 outgoing groups with p-value smaller than 0.05, and 200 incoming groups and the two lists share 156 annotation groups (18 expected by random).
Finally, it is interesting to compare the association we found between expression and connectivity of brain regions in rodents to the linkage previously found for single neurons in nematodes. To this end, we reanalyzed the data used in [11] using the global correlation test and created a list of functional annotation groups that are most correlated with connectivity in C. elegans (Table S6). A Pearson correlation test reveals that the list of p-values obtained for each functional annotation group in the worm is significantly correlated with the similar list obtained for rodents. For outgoing (incoming) connectivity, the correlation value is of 0.24, p-value 1e-5 (0.27, p-value 1e-6). Hence, there is a certain similarity in the functional gene groups that are associated with neural/brain connectivity across fairly distant phyla and across neuroanatomical scales.
Our work follows a direction set forth by previous work done for single neurons in C. elegans [11]–[13]. Despite obvious differences in the brain complexity, connectivity type, and the amount and quality of the data, it is interesting to compare the prediction performance obtained here to that of its preceding C. elegans investigation. In the previous study of [11], the mean Area Under the ROC curve (AUC) for the prediction experiments is only about 0.6 for both incoming and outgoing connectivity. In our results, the average AUC is markedly higher (0.73 and 0.74). For all 289 genes used in [11], the correlation between connectivity and expression in the worm was 0.176 for outgoing connectivity, and 0.075 for incoming connectivity. Looking at all of the 20 thousands plus genes used in this work at once, the equivalent correlations are 0.26 and 0.23. Moreover, there is considerable variance in the predictability in different regions and some regions achieve quite high predictive values (0.83 and 0.79 mean AUC values over the significant regions, with maximal AUC values reaching 0.99).
Our results are further supported by the recent parallel contribution of French and Pavlidis [21], in which a similar correlation test yields a score of 0.22 and 0.26 for incoming and outgoing connectivity respectively. The work of [21] is focused on the correlation assay and the authors state that they were unable to perform convincing predictive experiments. Here, in difference, we show that there is a considerable predictive signal. In fact, the prediction capability is considerably stronger than that found in the worm, and many of the brain regions present a marked and highly significant level of predictability. This prediction ability is further used here to select the lists of connectivity-related genes. A predictive test is, in our minds, a more solid foundation for gene selection than a correlation test. This is because a combination of even uninformative features can produce a correlation map that is similar to a given input map, while the separation between train and test data in the prediction experiments is much less prone to this pitfall. The lists of selected connectivity-related genes we obtain are verified here by comparing them to various lists obtained from the literature, again, going beyond the results presented in [21].
Regions of high predictability do not seem to be clustered in specific parts of the hierarchy. While smaller nuclei with many connections and therefore more available data seem somewhat easier to predict, a comparison between a structure's volume and the predictability of its connectivity map shows that regions of all sizes depict good predictability (Supplementary Figure S2). This might suggest that all regions are potentially of high predictability; however, the quality of the data currently available limits our ability to uncover their true predictability. The correlation between spatial proximity and connectivity is 0.11 and 0.10 for outgoing and incoming connectivity (compared to 0.26 and 0.23). Thus, while in the brain nearby regions are more likely to be connected, this association is significantly lower than the association between gene expression and connectivity.
To build the combined rodent brain atlas that contains both expression and connectivity, we rely on available resources that are not fully compatible or complete. Some of the connectivity that is currently absent in the rat atlas may actually exist in the rodent brain. The assumption of conservation of connectivity and expression between mouse and rat, underlying the construction of a combined atlas of a common rodent ancestor, probably holds only partly. Furthermore, the gene expression data was not measured during brain development, as would ideally have been more befitting. Yet, as both connectivity and expression are associated with common factors such as functionality, it is perhaps not surprising that considerable pertaining information can be delineated in adult expression patterns of neurons. As evident, the latter permit a considerable level of connectivity prediction, exhibit significant correlations with the connectivity data, and show a marked overlap between genes that are discriminative for incoming and outgoing connectivity. Finally, strictly speaking, we identify an association and not a causal relation from genes to connectivity. Although this causal direction is expected based on current consensus, it is certainly possible that connectivity in turn affects gene expression – one possible route for such effects may be indeed via activity-dependent mechanisms that shape synaptic formation and maintenance, mentioned earlier [8]–[9].
Despite the above limitations to the quality of the data, we were able to uncover a fairly marked association between gene expression and connectivity. Thus, we are able to make a significant advancement toward the long term goal of inferring the connectome from the genome [32]. Naturally, had our data been richer, for example, alleviating the need to rely on conservation across species, even better results could be expected. However, especially given these limitations, the magnitude of the association found here is truly remarkable, and the large-scale analysis approach presented here will undoubtedly show its continuing value in future studies as more refined data accumulates. This type of analysis is valid for both single neuron connectivity and connectivity between brain regions, and it is likely to be valid for intermediate, mesoscopic scales [24], [33]. In the nearby future, such efforts can be applied to link between newly established connectivity maps in humans (e.g. [34]) with accumulating regional gene expression data in the human brain. Moreover, once the genetic atlas of the developing brain [16] is processed to register gene maps, a distinction can be drawn between genes that are associated in maintaining connectivity and genes that are dominant during the initial formation of brain connectivity.
With the future advent of better and more accurate data we might be able to perform the analysis presented here focusing solely on the gene expression of neuronal cells while disregarding other cell types. To gain preliminary experimental insight into the role played by cell type in determining the link between expression and connectivity, we have examined the human data available from two recent papers. The first paper [35] has microarray data collected from the brains of AD patients and controls. In the second paper [36], care was taken such that the gene expression data was collected from neurons only. Therefore, for a first approximation, we have samples that are glia + neurons and samples that are only neurons. By comparing the two sets of samples we can identify genes that are over-expressed in glia and not over-expressed in neuron samples (Note that the situation is not symmetric and the opposite list cannot be extracted without further assumptions). Working with the mouse homologs of the identified human genes, we find that those genes that tend to be over expressed in glia are less informative than a typical group of the same size. The p-value of this finding is borderline though – 0.02 for efferent correlation test and 0.17 for the afferent correlation test. Future studies analyzing neuronal vs glial expression data comparatively are hence needed to shed further light on this intriguing question.
Our study has been made possible thanks to the innovative open approach of the Allen Brain project [16]. Gene expression data was obtained from the Allen Mouse Brain Atlas (ABA) dataset [15] for gene expression in the adult mouse brain composed of 20,936 genes (http://mouse.brain-map.org/). For each gene a 200 micron 3d volume of gene expression in the mouse brain is available (a vector of length ∼150 k). Some genes have several scans. Scans are available in one of two planes: Coronal and Sagittal. We compiled a dataset of voxel gene expressions based on sagittal scans. When numerous scans exist for a single gene a mean is taken (maximum was also tried – resulting in only minute, negligible differences in results reported).
For linking voxels to brain structures we use the structural annotation available at ABA (http://mouse.brain-map.org/pdf/Allen_Reference_Atlases.pdf). It defines a nomenclature of 209 brain structures organized in a hierarchy. The gene expression for each brain structure is computed as the average of all voxels contained within that region. Once more, experiments were also performed by taking the maximal value instead of the mean with little, negligible influence on the connectivity prediction ability and on the results reported.
One should note that during the preparation of this work partial results on the developing mouse brain have been uploaded to the ABA website. These results are not complete enough to enable us to run our experiments on a developing brain. For example, there is no mapping currently available between voxels and brain structures.
Rat connectivity information is obtained from [18]. To match rat connectivity to mouse gene expression we link the rat nomenclature of [22] and the ABA mouse nomenclature, by creating a mapping between identical terms. The mapping is given in Table S7.
It sometimes occurs that a region is identified in the mouse nomenclatures and at least one of the children of this region is not identified. Even in such cases, we do not perform the analysis on the non-leaf regions. This policy simplifies the framework and minimizes borderline cases, for example, when some of the leaves are identified and some are not.
We use a Linear Support Vector Machine (SVM) [37] classification with a fixed parameter of C = 1 for prediction. The learned binary labels correspond to the existence or non-existence of a connection between regions. Regions with less than 5 positive examples (i.e. connections) are discarded. For each region separately, a balanced 5-fold cross-validation is performed on this data with 80% training and 20% testing. Since each connection (existing or not) is tested exactly once, the cross validation procedure produces a connectivity prediction value for each possible connection. We consider the real value which is the signed distance from the learned classifier's separating hyperplane, and use it to compute the Area Under Curve (AUC) statistics. To eliminate dependence on the random split used, each such cross-validation experiment is repeated 20 times, and the mean AUC is recorded. In order to evaluate statistical significance, the entire experiment is repeated 1,000 times while permuting the labels.
To demonstrate the validity of the prediction assay in the face of missing connectivity data we perform the following synthetic data experiment: A random network was created of a similar cardinality as the BAMS network used in our experiments, such that the degrees of the nodes are five times higher than those of the BAMS network (varying between nodes, similarly to BAMS). Synthetic random vectors of “gene expression” were created in such a way that nodes that are connected to a specific node have for a subset of the genes a somewhat similar pattern, randomly varied around a certain central pattern, i.e., tend to have some genes overexpressed and some genes underexpressed in a similar manner. Then, we run the same protocol as in our prediction assay and measure success by computing the mean AUC obtained from all regions (the equivalent success in the real data experiments is 0.73). This experiment is then repeated when some of the initially given positive connections are held out and marked as ‘non existing’ (i.e., incorporating missing data in a controlled manner).
The results of the simulations for specific missing data values, averaged over many runs are presented below in Supplementary Figure S1 . As can be seen, even for such challenging simulations where the prediction for the full dataset is at 80%, the results degrade nicely with the number of missing connections. In these noisy conditions the results vs the simulated atlas remain well above chance even when only 15% of the connections are retained (i.e., ‘known’, blue-line). Moreover, the classifiers learned with the missing data are useful for predicting the complete (no missing data) simulated connections (red-line).
To examine the correlation between a genetic pattern and a connectivity pattern across all brain structures under investigation, we used an assay similar to the one used by Toledo–Rodriguez et al [26]. This assay was also used in [11]. Given a set of N = 146 structures, we constructed two N×N similarity matrices, S1 and S2, where S1 (S2) represents the pairwise similarity between the expression data (connectivity) of every two brain structures. Pearson correlation is used as a measure of those pairwise similarities for both gene expression and connectivity, both between the vectors of gene expression, and the connectivity vectors. The (N * N/2 – N) entries forming the lower triangle of S1 (S2) are concatenated to form a covariation vector v1 (v2). The Pearson correlation between the two covariation vectors v1 and v2 describes the extent to which similarities in gene expression imply similarities in connectivity and vice-versa. The statistical significance of the resulting correlation is computed using an empiric null hypothesis constructed from repeating the procedure with shuffling. On each repetition the gene expression signatures were shuffled amongst all regions, thus disassociating a region and its gene expression. The p-values are calculated by repeating the shuffling 1,000 times and computing the probability to achieve a score equal or higher than the score of the non-shuffled data.
Similarly to the prediction assay, for each brain region we take connected regions gene expression as positive examples and non-connected regions as negative examples. This is done once for outgoing connections, and once for incoming connections, where the two experiments are performed independently. At each time, feature (gene) selection was performed using zero norm SVM algorithm [38]. Zero norm SVM works by iteratively training an SVM while reweighing the feature vectors until convergence. In order to select a fixed number of features, we have selected the 500 features with the highest weights provided by the zero-norm SVM procedure. This is repeated for each brain structure which has at least 5 connections, i.e., to 44 regions in the outgoing experiment and to 57 regions in the incoming experiment.
To obtain two global lists of selected genes that are informative to either outgoing connectivity or incoming connectivity, the individual lists obtained for each region are combined. This is done by counting for each gene the number of times it was selected across the brain structures in each of the two experiments. The 500 genes that appeared most frequently in the individual outgoing experiments form the list of selected outgoing genes, and similarly for the incoming list.
To gain more insight into the nature of the selected genes, we have employed the DAVID functional annotation tools [31] to determine the most prominent annotations in the two lists formed above. The details of this experiment are provided in Supplementary Table S8.
To alleviate potential concerns about the influence of artifacts in the gene expression data on the prediction and gene selection process, we have compared the prevalence of artifacts in the data of selected genes to that of a disjoint sample of genes. 50 genes were sampled randomly from the groups 241 genes that are found to be predictive for both outgoing and incoming connectivity. Another group of 50 genes was sampled from the 1000 most brain active genes that do not appear in either list of predictive genes. For further control, genes that were not highly expressed in the brain were removed from the study since their images are expected to contain less data and therefore fewer artifacts.
The results show that for the sample of connectivity predictive genes, 58% of the slices contained local artifacts such as localized stains. The equivalent number for the background group is 57%. The ratio of global artifacts such as folds and scratches are also quite similar between the two groups: 11% and 17% respectively. Overall, we do not observe a tendency for more artifacts in the selected genes in comparison to the general population of brain-expressed genes. Supplementary table S9 contains the raw data of this analysis.
The top 100 genes associated with each disorder were extracted from the HuGe database [39], and the size of the intersection of these lists and the two lists of connectivity genes extracted by the feature selection method above were computed. The expected size of a random intersection is 2.5 genes.
There were 4 such lists. (1) Axon guidance genes were obtained from the gene families discussed in [27]: Netrin, Slit, Semaphorin, Ephrin, DCC, UNC5, Robo, Robo3, Neuropilin, Plexin and Eph. A total of 86 homologous members of these families were matched in the ABA gene set. (2) A group of 103 pre-synaptic gene homologs was obtained from a list of 107 genes appearing in [28]. (3) C.elegans genes were obtained from mouse homologies on the most highly ranked genes shown to be involved in neural connectivity in [11]. ABA homologies of 19 outgoing (31 incoming) were obtained from 30 outgoing (53 incoming) C.elegans genes. (4) The list of genes which are indicative of embryonic history taken from [29]. 83 such genes were identified within the ABA gene list out of 93 in the original list.
In order to compute the significance of the correlation assay results obtained by a group of genes, such as the three literature based gene-lists or the 1,616 DAVID groups, we have compared the p-value obtained using the correlation assay with the p-values obtained for 1000 random groups of the same size. This procedure eliminates bias caused by the group size.
|
10.1371/journal.pgen.1002786 | FANCJ/BACH1 Acetylation at Lysine 1249 Regulates the DNA Damage Response | BRCA1 promotes DNA repair through interactions with multiple proteins, including CtIP and FANCJ (also known as BRIP1/BACH1). While CtIP facilitates DNA end resection when de-acetylated, the function of FANCJ in repair processing is less well defined. Here, we report that FANCJ is also acetylated. Preventing FANCJ acetylation at lysine 1249 does not interfere with the ability of cells to survive DNA interstrand crosslinks (ICLs). However, resistance is achieved with reduced reliance on recombination. Mechanistically, FANCJ acetylation facilitates DNA end processing required for repair and checkpoint signaling. This conclusion was based on the finding that FANCJ and its acetylation were required for robust RPA foci formation, RPA phosphorylation, and Rad51 foci formation in response to camptothecin (CPT). Furthermore, both preventing and mimicking FANCJ acetylation at lysine 1249 disrupts FANCJ function in checkpoint maintenance. Thus, we propose that the dynamic regulation of FANCJ acetylation is critical for robust DNA damage response, recombination-based processing, and ultimately checkpoint maintenance.
| The BRCA1–Fanconi anemia (FA) pathway is required for both tumor suppression and cell survival, particularly following treatment with DNA damaging agents that induce DNA interstrand crosslinks (ICLs). ICL processing by the BRCA–FA pathway includes promotion of homologous recombination (HR) and DNA damage tolerance through translesion synthesis. However, little is known about how the BRCA–FA pathway or these ICL processing mechanisms are regulated. Here, we identify acetylation as a DNA damage–dependent regulator of the BRCA–FA protein, FANCJ. FANCJ acetylation at lysine 1249 is enhanced by expression of the histone acetyltransferase CBP and reduced by expression of histone deacetylases HDAC3 or SIRT1. Furthermore, acetylation on endogenous FANCJ is induced upon treatment of cells with agents that generate DNA lesions. Consistent with this post-translation event regulating FANCJ function during cellular DNA repair, preventing FANCJ acetylation skews ICL processing. Cells have reduced reliance on HR factor Rad54 and greater reliance on translesion synthesis polymerase polη. Our data indicate that FANCJ acetylation contributes to DNA end processing that is required for HR. Furthermore, resection-dependent checkpoint maintenance relies on the dynamic regulation of FANCJ acetylation. The implication of these findings is that FANCJ acetylation contributes to DNA repair choice within the BRCA–FA pathway.
| The hereditary breast cancer associated gene product, BRCA1 is an essential tumor suppressor. To promote genomic stability, BRCA1 interacts with multiple protein partners. In particular, through its C-terminal BRCT repeats, BRCA1 directly interacts with Abraxas, CtIP and FANCJ (also known as BRIP1 or BACH1 (BRCA1-associated C-terminal helicase 1)). These BRCT-interacting proteins contribute to the function of BRCA1 in the DNA damage response (DDR). Abraxas serves to localize BRCA1 to sites of DNA damage and CtIP promotes the initiation of DNA end resection, which is critical for HR [1]–[3]. FANCJ also participates in localizing BRCA1 to sites of DNA damage, in DNA repair, and in checkpoint signaling; however, its distinct function is less clear.
Elucidating how FANCJ functions in the DDR is important, as mutations in the FANCJ gene are associated with hereditary breast cancer as well as with the rare cancer prone syndrome Fanconi anemia (FA) within the FANCJ patient complementation group (FA-J) [4]. As a DEAH-family helicase, it is expected that FANCJ metabolizes DNA substrates to facilitate DNA repair. Consistent with this idea, recombinant-FANCJ is a 5′-3′ helicase and translocase that can unwind D-loops and displace RAD51 [5]. In cells, FANCJ also localizes to sites of DNA damage. Furthermore, when FANCJ is absent, catalytically inactive, or lacks BRCA1 binding, cells display defects in double strand break repair (DSBR) and HR [6]–[9]. Recently, FANCJ was identified as a factor essential for maintaining the DNA damage induced checkpoint in response to ionizing radiation [10]. Despite these findings, FANCJ-deficient cells are only mildly sensitive to agents that induce DSBs [11].
To explain these findings, it has been proposed that FANCJ functions in DSBR, but has a more significant role in processing replication forks stalled at lesions, such as DNA interstrand crosslinks (ICLs). In support of this idea, FANCJ-null cells, similar to other FA patient cells, are extremely sensitive to agents that induce ICLs, such as cisplatin, melphalan, or mitomycin C (MMC) [7], [12], [13]. This sensitivity is reversed by complementation of FA-J cells with wild-type FANCJ (FANCJWT), but not with catalytically inactive FANCJ mutants [6], [8], [14]. Interestingly, the mechanism by which FANCJ mediates ICL processing is regulated by BRCA1 binding. HR is favored when BRCA1 binds FANCJ. When BRCA1 binding is prevented, lesion bypass is favored by a mechanism requiring the translesion synthesis polymerase polη [9]. Thus, complementation of FA-J cells with a BRCA1-interaction defective mutant FANCJS990A reverses ICL sensitivity but does not fully restore FANCJ function.
Here, we present evidence that FANCJ contributes to lesion processing by promoting a robust DDR. Essential for this function is FANCJ acetylation on a specific lysine residue. As such, preventing FANCJ acetylation suppresses DNA end resection that normally serves to engage recombination-based processing. Thus, both BRCT-interacting proteins, CtIP and FANCJ undergo DNA damage induced changes in acetylation that further regulates their function in the DDR to promote genomic stability.
As observed for CtIP, FANCJ binds directly to the BRCT domains of BRCA1 [6], [9], [15]. Given that CtIP function is inactivated by acetylation [16], we addressed whether FANCJ was similarly modified. For this analysis, myc-tagged FANCJ was co-transfected with various Flag- or HA-tagged histone acetyltransferases. In an immunoblot probed with a pan-acetyl lysine antibody, we found that the precipitated FANCJ was acetylated only when CBP was over-expressed (Figure 1A). Moreover, FANCJ acetylation was induced by CBP in a dose dependent manner (Figure 1B).
FANCJ acetylation was preserved most effectively by the inclusion of two types of deacetylase inhibitors, trichostatin-A (TSA) and nicotinamide (NAM) (Figure 1C). Thus, we considered that more than one class of histone deacetylase (HDAC) could deacetylate FANCJ. TSA inactivates class 1 and class II HDACs, whereas NAM inactivates the nicotinamide adenoine dinucleotide (NAD+)-dependent sirtuin (class III) family of HDACs (including SIRT1 to SIRT7) [17]. FANCJ acetylation was reduced more when either Flag-tagged-HDAC3 or SIRT1 were overexpressed in 293T cells than observed upon overexpression of HDAC1, HDAC2, or SIRT6 (Figure 1D). Titration of the SIRT1 expression vector transfected into 293T cells revealed that 0.01 µg of the SIRT1 construct matched the expression level of 4 µg of the HDAC3 construct. At this similar level of expression, HDAC3 more efficiently deactylated FANCJ than did SIRT1 (Figure 1E). Together, these data implicate that FANCJ can be acetylated by CBP and deacetyated by HDAC3 as well a SIRT1 when over-expressed.
To identify the FANCJ acetylation site(s), myc-tagged C-terminal FANCJ truncation mutants were co-transfected with CBP into 293T cells. By Immunoblot analysis using the pan-acetyl antibody, we found that acetylation of FANCJ required amino acids 1239 to 1249 (Figure 2A, 2C). Consistent with this region being modified, a C-terminal domain of FANCJ similar to a C-terminal p53 control was acetylated in vitro by a HAT-domain protein (Figure 2B, 2C). To determine, which of three lysine (K) residues in this C-terminal region were required for acetylation, we generated three independent FANCJ mutant constructs that converted lysines 1240, 1242, or 1249 to arginine (R). Further transfection experiments revealed that the K1249 was the dominant site for FANCJ acetylation, a lysine that is not conserved in chicken or C. elegans FANCJ species (Figure 2D, 2E).
Next, we sought to provide more conclusive evidence that CBP-induced acetylation on FANCJ was at the K1249 site. We purified FANCJ from 293T cells transfected with a C-terminal myc-tagged FANCJWT or the FANCJK1249R mutant species by immunoprecipitation using a myc antibody. Isolated proteins were then digested with trypsin and subjected to tandem mass spectrometry analysis (LC-MS/MS). FANCJ-derived peptides covering the entire sequence were analyzed, and acetylation sites were identified using MASCOT search algorithm. Most of the acetylated lysine residues were detected in overlapping peptides derived from at least two independent protein preparations. In the FANCJWT, one of these sites was K1249 (Figure 2F). Interestingly, even though by antibody detection, the FANCJK1249R mutant scores unmodified as in Figure 2D; FANCJWT and FANCJK1249R mutant had three additional acetylation marks detected by mass spectrometry (Figure S1). Furthermore, the K1249R mutant had five additional acetylated lysines not found in wild-type FANCJ, suggesting that these sites are not available when K1249 is acetylated (Figure S1). Thus, immunoblot and mass spectrometry analysis confirm that the very last amino acid of FANCJ, lysine 1249 is acetylated.
Given that DNA damage reduces CtIP acetylation [16], we addressed whether DNA damage could alter FANCJ acetylation. Endogenous FANCJ acetylation was enhanced in MCF7 cells treated with zeocin, camptothecin (CPT), or hydroxyurea (HU) as compared to ultraviolet radiation (UV), MMC, or methyl methanesulfonate (MMS) at the dose and time-post treatment analyzed (Figure 3A). Notably, zeocin had a more robust induction of FANCJ acetylation despite the dose of zeocin, CPT, or UV having similar affect on cell survival (Figure S2; data not shown). As found previously, DNA damage did not measurably alter FANCJ co-precipitation with BRCA1 with the exception of UV damage, which could reflect the UV-induced BRCA1 degradation [11], [18] (Figure 3A). DNA damage also induced FANCJ acetylation in HeLa cells, in response to not only CPT, but also MMC (Figure 3A). In response to DNA damage, we also noted that FANCJ protein levels were sometimes enhanced (Figure 3A). To clarify whether acetylation or our ability to detect acetylation was induced by DNA damage, we sought to induce DNA damage in cells in which our ability to detect FANCJ acetylation was not limiting. Indeed, the amount of acetylation on similar levels of exogenous FANCJWT achieved with low dose CBP expression was considerably enhanced following treatment with zeocin or CPT (Figure 3B, 3C). Interactions with BRCA1 and MLH1 were not required for the CBP-induced acetylation of FANCJ, because BRCA1- and MLH1-interaction-defective mutants, FANCJS990A and FANCJK141/142A were readily modified (data not shown). In contrast, following treatment with CPT, acetylation was not detected on the FANCJK1249R mutant (Figure 3C), indicating that DNA damage-induced FANCJ acetylation requires the C-terminal K1249 residue. It remains to be determined, however if FANCJ acetylation is induced by a distinct type of DNA damage.
The enhanced FANCJ acetylation following DNA damage led us to hypothesize that this modification facilitated FANCJ function in DNA repair. To address this possibility, we made use of this lysine to arginine FANCJK1249R mutant that prevents acetylation and also generated a lysine to glutamine FANCJK1249Q mutant to structurally mimic acetylation. Consistent with these mutants being functional, the purified recombinant proteins displayed similar catalytic activities as FANCJWT (Figure S3). In addition, they were expressed at similar levels as FANCJWT in FANCJ-null FA-J cells (Figure 4A). Similar to FANCJWT, FANCJK1249R and FANCJK1249Q precipitated with known FANCJ interacting partners, BRCA1 and MLH1 [6], [8] (Figure 4B). In addition, the mutants co-localized with BRCA1 in response to DNA damage and the FA-J cells expressing FANCJWT or mutants had similar asynchronous cell cycle profiles (Figure 4C, 4D). The acetylation mutants also restored MMC resistance and the ability of FA-J cells to exit from an abnormal G2/M accumulation, albeit in a manner slightly more robust than FANCJWT (Figure 4E, 4F). Together, these findings suggested that the mutants were enzyme active and functional in vivo; however the mechanism by which the FANCJ mutants restore ICL resistance could be distinct from FANCJWT.
Previously, complementation of FA-J cells with a BRCA1-binding defective mutant, FANCJS990A gave the semblance of FANCJWT function. In particular, MMC resistance was restored [8]. However, in contrast to FANCJWT, FANCJS990A provides resistance to MMC by a mechanism dependent on the DNA damage tolerance pathway. Within this tolerance pathway, translesion synthesis polymerases can bypass DNA lesions such as unhooked ICLs and intra-strand crosslinks generated by UV, but not DSBs generated by zeocin. Evidence that FANCJS990A skewed lesion processing towards DNA damage tolerance was based on several findings. First, the sensitivity to MMC in these cells was restored upon depletion of the essential tolerance factor, Rad18 or the translesion polymerase polη, but not upon depletion of the HR protein, Rad54. Second, in comparison to FANCJWT, cells expressing FANCJS990A were hyper-resistant to UV, a phenotype that was reversed upon polη-depletion. Third, in comparison to FANCJWT, FANCJS990A-expressing cells were sensitive to zeocin, indicating reduced DSBR [9]. Thus, we sought to determine whether similar to the BRCA1-binding mutant, the acetylation mutants also functioned differently from FANCJWT. To test this idea, the FA-J cell lines were left untreated or treated with increasing doses of MMC, zeocin, or UV. In comparison to the other FA-J cell lines, the FA-J cell line expressing the acetylation mutant FANCJK1249R was hyper-resistant to UV, but unable to restore normal levels of zeocin resistance. In contrast, the FA-J cell line expressing the acetylation mimic FANCJK1249Q displayed greater resistance to zeocin (Figure 5A; Figure S2). Thus, in response to UV and zeocin, cells expressing the acetylation mutants are distinct from each other as well as from cells expressing FANCJWT.
To further validate these results, we targeted recombination or DNA damage tolerance pathways by using siRNA reagents to Rad54 or polη. Significantly, depletion of Rad54 suppressed the zeocin resistance of the FA-J cell line expressing FANCJK1249Q (Figure 5A, 5C). Likewise, depletion of polη suppressed the UV hyper-resistance of the FA-J cell line expressing FANCJK1249R (Figure 5A, 5C). Furthermore, depletion of polη, but not Rad54 reversed the MMC resistance of the FA-J cell line expressing FANCJK1249R (Figure 5B). In contrast, depletion of Rad54, but not polη reduced the MMC resistance of the FA-J cell line expressing FANCJK1249Q (Figure 5B). Together, these results indicate that the acetylation of FANCJ at lysine 1249 contributes to the mechanism of lesion processing; preventing acetylation favors DNA damage tolerance and constitutive acetylation favors recombination.
How could FANCJ acetylation affect lesion processing? Because both CtIP and FANCJ are acetylated and directly bind to the BRCA1-BRCT domain, we speculated that FANCJ might similarly have a role in DNA end resection. In particular, the affect of CtIP acetylation on DNA end resection was analyzed in response to CPT [16]. We found RPA foci formation at 1 h post-CPT was more robust (64% and 65%) in the FANCJWT and FANCJK1249Q FA-J cell lines as compared to vector and FANCJK1249R FA-J cell lines that had 47% and 29%, respectively (Figure 6A). Thus, as measured by RPA foci formation, FANCJWT and the acetylation mimic FANCJK1249Q were more active in DNA end resection.
RPA loading onto ssDNA also leads to its subsequent phosphorylation on Ser4 and Ser8 [3]. We found that the FA-J cell lines had a similar phosphorylation of Chk2 and γ-H2AX following exposure to two different dose of CPT, indicating that FANCJ or its ability to be acetylated is not required for DSB formation in response to CPT (Figure 6B). Likewise, at 1 h post-CPT treatment, Chk1 phosphorylation was detected (Figure 6B). In contrast, RPA phosphorylation was most robust in the CPT-treated FANCJWT and FANCJK1249Q FA-J cell lines (Figure 6B). In support of these findings, reduced RPA phosphorylation was also detected in CPT-treated FANCJ-deficient U2OS cells generated by siRNA reagents (Figure S4). Furthermore, at 4–24 h post CPT treatment, we noted diminished RPA phosphorylation in FANCJK1249R as compared with FANCJWT and FANCJK1249Q FA-J cell lines (Figure 6C). At this time, RPA phosphorylation in the FANCJK1249R FA-J cells was also reduced compared to vector FA-J cells that had gained considerable RPA phosphorylation as compared to 1 h post-CPT (Figure 6B, 6C). In the response to zeocin, which induces DSBs independent of replication, RPA phosphorylation was similar in FA-J cell lines with or without FANCJWT (Figure S5). Together, these results suggest a role for FANCJ and its acetylation in DNA end resection at stalled replication forks as induced by CPT.
To address whether the contribution of FANCJ acetylation to DNA end resection was sufficient to enhance HR, we next analyzed Rad51 foci formation. In response to CPT, we found that Rad51 foci were the most robust in FA-J cells complemented with FANCJWT or the FANCJK1249Q mutant. Instead, Rad51 foci in the FA-J cells with vector or FANCJK1249R were more anemic (Figure 7A). Furthermore, a greater number of FANCJK1249Q expressing FA-J cells were positive for Rad51 foci as quantitated between 2–16 h after CPT treatment (Figure 7A). In contrast, the γ-H2AX foci did not have a significant difference between the FA-J cells lines. Thus, a greater proportion of γ-H2AX co-staining Rad51 foci were detected in FANCJK1249Q or FANCJWT, as compared to vector or FANCJK1249R expressing FA-J cells (Figure 7A merge). Together, these findings demonstrate that in response to CPT, FANCJ and its acetylation at 1249 promote DNA end processing events that enhance RPA phosphorylation, and both RPA and Rad51 focal accumulation.
Given these findings and the recent identification that FANCJ promotes checkpoint maintenance [10], we considered that FANCJ acetylation could be essential for maintaining the checkpoint. Defects in checkpoint maintenance were evaluated by determining if CPT treated FA-J cells traversed prematurely to mitosis. FA-J cells lacking FANCJWT entered mitosis by 24 h post-CPT as indicated by a positive histone H3 phosphorylation (Figure 7B). These results are consistent with FANCJ acetylation supporting checkpoint maintenance. However, FA-J cells expressing FANCJK1249R or FANCJK1249Q also failed to maintain the checkpoint, showing H3 phosphorylation by 24 h (Figure 7B). Substantiating this finding, at time points greater than 4 h post-CPT treatment, both mutants had reduced Chk1 phosphorylation as compared to FA-J cells expressing FANCJWT (Figure 7B). Collectively, these findings suggest that FANCJ acetylation enhances the initial DDR to facilitate recombination-based repair and limit translesion synthesis. Checkpoint maintenance however, requires FANCJ and its dynamic regulation by acetylation (Figure 7C).
Here we identify acetylation as a DNA damage-dependent regulator of the BRCA-FA protein, FANCJ. We show that acetylation at lysine 1249 is a critical regulator of FANCJ function during cellular DNA repair. We analyzed the expression of two FANCJ mutants that mimicked either the constitutive deacetylated FANCJK1249R or acetylated FANCJK1249Q protein isoforms. While the mutants functioned similar to FANCJWT in several assays and restored MMC resistance and exit from an abnormal G2/M checkpoint response to FA-J cells, the mutants were distinct from FANCJWT with respect to lesion processing. Notably, FA-J cells expressing the acetylation mutants differentially relied on repair and tolerance factors for resistance to DNA damaging agents. Our findings further demonstrate that FANCJ has the ability to potentiate HR and DNA damage induced acetylation is important for this function.
Another BRCA1-BRCT interacting protein, CtIP is acetylated and functions in DNA end resection. Thus, we considered that recombination-based lesion processing by the FANCJ acetylation mimic, FANCJK1249Q resulted from a function for FANCJ acetylation in DNA end resection. To test this idea, the FA-J cells were treated with CPT, which generates breaks in S-phase. Indeed, FA-J cells expressing the acetylation mutants were distinct in the initial response to CPT. Specifically, FA-J cells expressing the FANCJK1249Q, but not FANCJK1249R, promoted DNA end resection post-CPT exposure as measured by presentation of RPA foci or its phosphorylation at serine residues 4 and 8. Furthermore, FA-J cells expressing the FANCJK1249Q, as compared to FANCJK1249R had 2.5-fold more cells with CPT-induced Rad51 foci. This more robust DDR could reflect a role for FANCJ acetylation in loading RPA, as shown for FANCJ in response to HU [19]. Our data do not implicate a global role for FANCJ in DNA end resection given that FANCJ-deficiency did not affect the amount of RPA phosphorylation following zeocin, an agent that induces DSBs independent of replication. However, FANCJ is acetylated when cells are exposed to CPT or zeocin. Thus, DSB-induced FANCJ acetylation that is not associated with stalled or broken replication forks may contribute to some other aspect of the DDR, such as checkpoint maintenance.
In fact, we find that FANCJ as well as its acetylation are essential for checkpoint maintenance. Specifically, in the absence of FANCJ or its DNA damage induced acetylation, Chk1 phosphorylation was induced, but not maintained and correspondingly cells underwent a more rapid transit into mitosis post-CPT. Interestingly, we found that similar to FA-J cells expressing the acetylation mutant FANCJK1249R, FA-J cells expressing the acetylation mimic FANCJK1249Q failed to maintain the checkpoint despite an initial DDR to CPT. Thus, some other aspect of checkpoint signaling is perturbed in FA-J cells that express the acetylation mimic. Perhaps this mutant fails to mediate a protein interaction or act upon a DNA substrate important for checkpoint maintenance. Instead FANCJ acetylation could serve as a switch, in which acetylation and de-acetylation is essential to maintain the checkpoint (Figure 7C). Consistently, a role for FANCJ in checkpoint maintenance was reported in a recent study [10].
It follows that defects in initiating the DDR, engaging HR, and maintaining the checkpoint impact cellular DNA damage resistance. Reduced DNA repair and/or checkpoint maintenance defects could explain why FA-J cells expressing the acetylation mutant FANCJK1249R were sensitive to zeocin. Defects in repair and in maintaining the checkpoint may not increase cellular sensitivity if backup lesion processing mechanisms serve to process or bypass the lesion. Compensatory pathways could explain the lack of CPT-sensitivity in the FA-J cells with or without acetylation mutants (Figure S2). In support of this idea, our data reveal that FA-J cells expressing the acetylation mutant were resistant to DNA damage by relying on tolerance factors. As such, depletion of polη in FA-J cells expressing the non-acetylatable FANCJK1249R mutant reversed the UV and MMC resistance. Instead, FA-J cells expressing the acetylation mimic FANCJK1249Q maintained zeocin and MMC resistance in a Rad54-dependent manner. These findings suggest that the toxicity to ICLs lesions as found in cells deficient for FANCJ is avoided because FANCJ enzyme active acetylation mutants facilitate recombination in S phase or translesion synthesis bypass of unhooked ICL lesions perhaps in mitosis. In the absence of a maintained checkpoint, however recombination similar to translesion synthesis bypass is likely to be error-prone.
Previously, we found that BRCA1 binding to FANCJ altered FANCJ function in HR and translesion synthesis pathways. Indeed, we find that similar to FA-J cells expressing the acetylation FANCJK1249R mutant, FA-J cells expressing the BRCA-interaction defective mutant, FANCJS990A were hyper-resistant to UV induced damage, sensitive to zeocin induced damage, and relied on polη for MMC resistance [7]. Data also indicate that similar to FANCJK1249R, the FANCJS990A mutant fails to maintain the checkpoint. In response to melphalan treatment, FA-J cells expressing the FANCJS990A mutant, as compared to FANCJWT, underwent a reduced and more rapid G2/M checkpoint exit [7]. These similar outcomes do not reflect common defects in BRCA1 binding or acetylation. Indeed, the FANCJS990A mutant was acetylated upon co-transfection of CBP to levels similar to those observed for FANCJWT (data not shown). Moreover, co-precipitation experiments demonstrated that the FANCJK1249R mutant bound BRCA1 as well as FANCJWT. Thus, BRCA1 binding and acetylation of FANCJ may be distinct events. Nevertheless, defects in BRCA1 binding at serine 990 or acetylation at lysine 1249 could have similar outcomes for FANCJ function because both mutants fail to maintain a robust checkpoint and Rad51-based repair is reduced [9].
A stalled replication fork with exposed single stranded and double stranded regions could provide an ideal DNA substrate for FANCJ. Indeed, FANCJ requires several nucleotides for binding and metabolizing DNA [20]. FANCJ function in replication fork processing could also be similar to other 5′-3′ DNA helicase/translocases such as Ecoli RecD and yeast Rad3. Rad3 facilitates exonucleolytic degradation of DNA ends, which restricts recombination between short homologous sequences [21]. Interestingly, RecD regulates resection and recombination by changes in helicase speed, which can also facilitate a polymerase swap, in which bypass polymerases diminish fork break down [22]. Conceivably, enhanced FANCJ enzyme activity or altered substrate preference due to acetylation could generate more single-stranded DNA to elicit checkpoint responses such as RPA loading as proposed [19]. Alternatively, checkpoint maintenance could require reduced FANCJ enzyme activity so that FANCJ does not displace proteins from lesions, such as RAD51 or interacting partners BRCA1, RPA and BLM helicase [6], [23]–[25]. In this context, it is worth noting that changes in motor speed have been associated with FANCJ clinical mutants. The breast cancer associated mutant, M299I is enzyme activating and both unwinds and translocates DNA more efficiently than FANCJWT, whereas the P47A mutant is enzyme inactivating [26], [27]. Whether changes in FANCJ function derive from acetylation and/or partners that bind via this modification remains to be determined. Furthermore, based on our current data, it is unclear if distinct DNA lesions selectively induce FANCJ acetylation.
In summary, our findings indicate that FANCJ has the ability to potentiate HR through dual roles in DNA end processing and checkpoint maintenance. These two functions require FANCJ lysine 1249, a site not conserved in FANCJ orthologues such as chicken FANCJ and C. elegans Dog-1. Interestingly, unlike in human cells, FANCJ does not function in HR in chicken and C-elegans systems [28], [29]. It is not surprising that regulators of FANCJ acetylation state, HDACIII, SIRT1, and CBP have roles in DNA repair and genomic stability [30]–[32]. It remains to be determined, however, whether associated repair defects are related to failure to regulate FANCJ acetylation. Complicating this analysis, HDACIII, SIRT1, and CBP have many other histone and non-histone protein substrates that also have role in DNA repair and genomic stability. For example, SIRT1 deacetylation plays an important role in regulating the function of DNA double strand break repair proteins, such as Ku70 [33], WRN [34], and NBS1 [35]. Moreover, p300/CBP functions to regulate the activities of multiple proteins at the replication fork including PCNA [36]. CBP also regulates the activity of other helicases, including WRN [37]. Whether HDAC or HAT associated defects derive from a failure to regulate FANCJ acetylation will be an important question for future studies.
MCF7, HeLa, and 293T cells were grown in DMEM supplemented with 10% fetal bovine serum and penicillin/streptomycin (100 U/mL each). FA-J (EUFA30-F) cells were cultured with 15% fetal bovine serum and penicillin/streptomycin (100 U/mL each). FA-J cells were infected with the POZ retroviral vector [38] containing no insert, WT, K1249R, or K1249Q FANCJ inserts. Stable FA-J POZ cell lines were selected as before [8].
Cells were harvested, lysed, and processed for Western blot analysis as described previously using an NETN lysis buffer (20 mM Tris, 150 mM NaCl, 1 mM EDTA, and 0.5% NP-40) containing 10 mM NaF and 1 mM NaVO3 [7]. For acetylation detection, unless otherwise noted cells were lysed with 150 mM NETN buffer supplemented with 10 µM TSA and 5 mM nicotinamide. For γ-H2AX detection, cell pellets were collected and dissolved and boiled in 2× lysis buffer (50 mM Tris pH 6.8, 2% SDS, 1% B-ME). Antibodies used for immunoprecipitation (IP) and Western blot assays include FANCJ polyclonal Abs E67 [26], β-Actin (Sigma), pRPA S3/4 (Bethyl), RPA (Bethyl), pChk1 S317 (Bethyl), Chk1 (Bethyl), pChk2 (Cell signaling), Chk2 (Cell Signaling), γ-H2AX S139 (Millipore), H2AX (Bethyl), Flag (Sigma), HA (12C4), pan-acetylated lysine (Cell signaling), MLH1 (BD Bioscience), BRCA1 monoclonal (ms110), pH 3 (Millipore), H3 (Abcam), polη (Abcam), Rad54 (Abcam), Rad51 (Abcam), and Myc monoclonal (9E10).
FA-J stable cell lines were either mock treated or treated with 0.25 µg/ml of melphalan (Sigma) and incubated for various times. Cells were fixed with 90% methanol in PBS overnight and then incubated 10 min with PBS containing 30 µg/ml DNase-free RNase A and 50 µg/ml propidium iodide. 1×104 cells were analyzed using a FACs Calibur instrument (Becton-Dickinson, San Jose, CA). Aggregates were gated out and the percentage of cells in G2/M was calculated using Flow Jo software.
The pCDNA3-myc.his vector (Invitrogen) was digested by Not1/Apa1 and different FANCJ fragments generated by PCR and digested by Not1/Apa1 were inserted. Primers are available upon request. Reverse primers used for K1249R-pCDNA3 and K1249RQ-pCDNA3 are 5′TTTTGGGCCCCCTAAAACCAGGAAACATGCC3′ and 5′TTTTGGGCCCCTGAAAACCAGGAAACATGCC3′, respectively. The K1249R and K1249Q pOZ vectors were generated with the QuickChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA) by using the FANCJ- pCDNA3-myc.his or FANCJ-pOZ as a template and the following primers: (K1249R-pOZ-Forward) 5′GGCATGTTTCCTGGTTTTAGGGCGGCCGCTGGAGGAGCA3′ and (K1249R-pOZ-Reverse) 5′GTCTCCTCCAGCGGCCGCCCTAAAACCAGGAAACATGCC3′; (K1239Q-pOZ-Forward) 5′GGCATGTTTCCTGGTTTTCAGGCGGCCGCTGGAGGAGAC3′ and (K1239Q-pOZ-Reverse) 5′GTCTCCTCCAGCGGCCGCCTGAAAACCAGGAAACATGCC3′; Recombinant FANCJ protein production was made in insect cells using the PVL13.2 vector as before [26]. Full-length WT FANCJ was used as a template to generate the acetylation mutants using the following primers: (K1249R-PVL132 Forward) 5′GGCATGTTTCCTGGTTTTAGGGACTACAAGGAGACG3′ and (K1249-PV132 Reverse) 5′CGTCGTCCTTGTAGTCCCTAAAACCAGGAAACATGCC3′. (K1249Q-PVL132 Forward) 5′ GGCATGTTTCCTGGTTTTCAGGACTACAAGGACGACG3′ and (K1249Q-PVL132 Reverse) 5′ CGTCGTCCTTGTAGTCCTGAAAACCAGGAAACATGCC3′. The pGEX-5X vector (GE Healthcare Life Sciences) was digested by Sal1/Not1 and the FANCJ C-terminal fragment was generated by PCR and digested by Sal1/Not1 and inserted. Primers are available upon request. All DNA constructs were confirmed by DNA sequencing.
Stable FA-J cell lines were untransfected or transfected with siRNA previously described against Luc, Rad54, or polη [9]. Cells were seeded onto 6 well plates and incubated overnight. Seeded cells were either untreated or treated with increasing dose of MMC (1 h, serum free), UV, CPT, (1 h, serum free), or zeocin (1 h, serum free). To assay for percent survival, cells were counted 5–8 days post infection and percent survival was calculated as before [9].
Helicase assay reaction mixtures (20 µl) contained 40 mM Tris-HCl (pH 7.4), 25 mM KCl, 5 mM MgCl2, 2 mM dithiothreitol, 2% glycerol, 100 ng of bovine serum albumin/µl, 2 mM ATP, 10 fmol of 19-bp duplex DNA substrate (0.5 nM), and the concentrations of FANCJ (acetylated or non acetylated) indicated in the figures. Helicase reactions were initiated by the addition of FANCJ, and the reaction mixtures were incubated at 30°C for 15 min unless otherwise indicated. Reactions were quenched with the addition of 20 µl of 2× Stop buffer (17.5 mm EDTA, 0.3% SDS, 12.5% glycerol, 0.02% bromophenol blue, 0.02% xylene cyanol). For standard duplex DNA substrates, a 10-fold excess of unlabeled oligonucleotide with the same sequence as the labeled strand was included in the quench to prevent reannealing. Reaction products were resolved on nondenaturing 12% (19∶1 acrylamide-bisacrylamide) polyacrylamide gels, and quantitated as described previously [27].
Stable FA-J cell lines were seeded onto 6 well plates and incubated overnight. Cells were either untreated or treated with 1 mM HU (24 h) or 0.25 µM CPT (1 h). Cells were fixed with 3% paraformaldehyde/2% sucrose for 10 min at RT, and permeabilized with 0.5% Triton X-100 in 20 mM HEPES for 5 min on ice. Incubation with antibodies and washes were described previously [6]. For Rad51 staining, cells were fixed with 3% paraformaldehyde/2% sucrose for 10 min at RT, permeabilized with ice-cold methanol for 30 min, and blocked with 4% BSA for 1 h. Staining was as described previously [6].
The acetyltransferase assays were performed in 30 µl of reaction, which includes reaction buffer (50 mM HEPES (ph 8.0), 10% glycerol, 1 mM DTT, 1 mM PMSF, 10 mM Na-butyrate), 1 µL [3H]-acetyl-CoA, 1 µl recombinant HAT domain of p300 (gift of Dr. Luo), and recombination FANCJ-CT or p53-CT [39]. Reaction were carried out at 30°C for 1 h and separated by SDS-PAGE, analyzed by autoradiography. Concentrations of recombinant proteins were determined by comassie staining from Invitrogen.
Gel bands containing FANCJ1 were de-stained twice with 25 mM ammonium bicarbonate in 50% acetonitrile for 30 min in 37°C, reduced with 7.6 mg/ml dithiothreitol at 60°C for 10 min, and alkylated with 18.6 mg/mL iodoacetamide at room temperature for 1 hour. The bands were then washed twice with 25 mM ammonium bicarbonate in 50% acetonitrile for 15 min at 37°C prior to shrinking with 50 µL acetonitrile for 10 min at room temperature. 100 ng trypsin (Promega) was added to each sample and 25 mM ammonium bicarbonate was added until the gels were fully swollen (∼10–50 µL) and the digestion proceeded overnight at 30°C. Following digestion, peptide extracts were transferred into new tubes and the gels were further extracted with 50 µL of 50% acetonitrile containing 5% formic acid (v/v) and following 15 min were added to the initial extracts. The latter process was repeated for a total of three extractions. Extracts were then dried on a SpeedVac and reconstituted in 20 µL of 0.1% formic acid for LC-MS/MS analysis.
Tryptic peptides (2 µL) were directly loaded at 4 µL/min for 7 min onto a custom-made trap column (100 µm I.D. fused silica with Kasil frit) containing 2 cm of 200 Å, 5 µm Magic C18AQ particles (Michrom Bioresources). Peptides were then eluted using a custom-made analytical column (75 µm I.D. fused silica) with gravity-pulled tip and packed with 25 cm 100 Å, 5 µm Magic C18AQ particles (Michrom). Peptides were eluted with a linear gradient from 100% solvent A (0.1% formic acid∶acetonitrile (95∶05)) to 35% solvent B (acetonitrile containing 0.1% formic acid) in 35 min at 300 nL/min using a Proxeon Easy nanoLC system directly coupled to a LTQ Orbitrap Velos mass spectrometer (Thermo Scientific) [40]. Data were acquired using a data-dependent acquisition routine of acquiring one mass spectrum from m/z 350–2000 in the Orbitrap (resolution 60,000) followed by tandem mass spectrometry scans in the LTQ linear ion trap of the 10 most abundant precursor ions found in the mass spectrum. Charge state rejection of singly-charged ions and dynamic exclusion was utilized to minimize data redundancy and maximize peptide identification [41].
The raw data files were processed and searched against the human index of the SwissProt database (version 09/21/11) containing both the mutant and wild-type forms of FANCJ1 with Mascot (version 2.3.02; Matrix Science) using parent mass tolerances of 15 ppm and fragment mass tolerances of 0.5 Da. Full tryptic specificity with 2 missed cleavages was used and variable modifications of acetylation (protein N-term and lysine), pyro-glutamination (N-term glutamine), and oxidation (methionine), and fixed modification of carbamidomethylation (cysteine) were considered. Mascot search results were also loaded into Scaffold (Version 3.3.1; Proteome Software) for comparative analyses using spectral counting of tandem mass spectra and full annotation of the data [42].
|
10.1371/journal.pgen.1005423 | SLIRP Regulates the Rate of Mitochondrial Protein Synthesis and Protects LRPPRC from Degradation | We have studied the in vivo role of SLIRP in regulation of mitochondrial DNA (mtDNA) gene expression and show here that it stabilizes its interacting partner protein LRPPRC by protecting it from degradation. Although SLIRP is completely dependent on LRPPRC for its stability, reduced levels of LRPPRC persist in the absence of SLIRP in vivo. Surprisingly, Slirp knockout mice are apparently healthy and only display a minor weight loss, despite a 50–70% reduction in the steady-state levels of mtDNA-encoded mRNAs. In contrast to LRPPRC, SLIRP is dispensable for polyadenylation of mtDNA-encoded mRNAs. Instead, deep RNA sequencing (RNAseq) of mitochondrial ribosomal fractions and additional molecular analyses show that SLIRP is required for proper association of mRNAs to the mitochondrial ribosome and efficient translation. Our findings thus establish distinct functions for SLIRP and LRPPRC within the LRPPRC-SLIRP complex, with a novel role for SLIRP in mitochondrial translation. Very surprisingly, our results also demonstrate that mammalian mitochondria have a great excess of transcripts under basal physiological conditions in vivo.
| Mitochondria provide most of the energy required for key metabolic and cellular processes that are essential for life. The biogenesis of the mitochondrial oxidative phosphorylation system, the site of energy conversion, is dependent on the coordinated expression of the mitochondrial and nuclear genomes. Mitochondrial gene expression is largely regulated at the post-transcriptional level by RNA-binding proteins, including the LRPPRC-SLIRP complex. It is still unclear how the proteins within this complex regulate mitochondrial RNA metabolism. Here, we have knocked out the Slirp gene in mice to dissect the individual roles of LRPPRC and SLIRP and provide further insights into the mechanisms governing post-transcriptional regulation of mitochondrial gene expression. LRPPRC is required for the maintenance of mitochondrial mRNA polyadenylation whereas SLIRP, by facilitating the presentation (or association) of mRNAs to the mitochondrial ribosome, regulates the rate of translation. In addition, we demonstrate that mitochondrial mRNAs in mammals are present in quantities that far exceed those needed to maintain normal physiology under basal conditions.
| Mitochondria are double-membrane bound organelles that have fundamental roles in energy metabolism, cell health and death, making them essential for life. The oxidative phosphorylation (OXPHOS) system is the major site of ATP production in mitochondria and is composed of proteins encoded by two genomes, the nuclear genome and mitochondrial DNA (mtDNA). Consequently coordinated regulation of nuclear and mitochondrial gene expression is required to meet the changing energy demands of the cell. The compact size and organization of mtDNA in animals has necessitated the evolution of unique mechanisms to regulate the expression of the 13 subunits of the OXPHOS system that are mitochondrially encoded. Mitochondrial gene expression is complex and predominantly regulated at the post-transcriptional level [1,2] by nuclear-encoded mitochondrial RNA-binding proteins that control the processing, maturation, translation, stabilization and degradation of mitochondrial RNAs [3]. The mitochondrial RNA polymerase (POLRMT) stimulated by mitochondrial transcription factor A (TFAM) and B2 (TFB2M) produces near-genome length polycistronic transcripts [3]. Because animal mtDNA lacks introns, the 22 mitochondrial tRNA genes that are arranged between the 2 rRNA and 11 mRNA coding genes act as punctuation marks to signal the processing of the polycistronic transcripts [4] by mitochondrial tRNA (mt-tRNA) processing enzymes [5–7]. The processed transcripts undergo extensive maturation, including polyadenylation at the 3′ end of the mitochondrial mRNAs (mt-mRNAs) [8,9], and mitochondrial rRNAs (mt-rRNAs) and mt-tRNAs are modified enzymatically at specific nucleosides to enable proper folding and biogenesis of the translation machinery [2,3]. The matured mt-mRNAs are translated on mitochondrial ribosomes (mitoribosomes) [10], although it is not clear how they are recognized as they lack conventional 5′ and 3′ untranslated regions (UTRs), Shine-Dalgarno sequences and 5′ 7- methylguanosine caps [11].
The mammalian family of RNA-binding pentatricopeptide repeat domain (PPR) proteins consists of seven nuclear-encoded mitochondrial proteins, each of which has a specific role in regulating mitochondrial gene expression from transcription and processing to maturation and translation [12]. The PPR protein LRPPRC first came to attention when a mutation of the LRPPRC gene was shown to cause a rare French-Canadian variant of Leigh syndrome characterized by cytochrome c oxidase deficiency [13]. In cultured cells, LRPPRC knockdown (KD) causes a reduction of mt-mRNA levels [14–16] and impaired mitochondrial translation [16]. LRPPRC physically interacts with SLIRP, which has an RNA recognition motif (RRM), consistent with a role in mitochondrial RNA metabolism [14]. LRPPRC and SLIRP form a complex that mediates mt-mRNA stability [15–17] and both proteins are co-stabilized within this complex because reduction of LRPPRC levels leads to concomitant reduction of SLIRP [14–18]. In mice, the LRPPRC-SLIRP complex regulates mt-mRNA stability, polyadenylation and coordinated mitochondrial translation [17]. We have also demonstrated that the bicoid stability factor (BSF, renamed DmLRPPRC1), one of the two Drosophila melanogaster orthologues of mammalian LRPPRC [19,20], has a very similar function as the mammalian one [21]. Furthermore, DmLRPPRC1 associates with one of the two fly orthologues of SLIRP [19,21], suggesting that the interaction between PPR-motif- and RRM-containing proteins is important for mitochondrial RNA metabolism and has been conserved through evolution.
To address the unclear in vivo role of SLIRP and its function within the LRPPRC-SLIRP complex, we generated Slirp knockout mice. Molecular analyses revealed that SLIRP is required to stabilize LRPPRC. In addition, our findings show that LRPPRC and SLIRP have distinct roles within the mt-mRNA-stabilizing complex they form, i.e. LRPPRC is required for maintenance of polyadenylation whereas SLIRP regulates the rate of translation. Very surprisingly, we also report that mice lacking SLIRP are apparently healthy despite a very drastic (50–70%) depletion of mt-mRNAs. These findings show that mt-mRNAs in mammalian mitochondria are present in quantities that far exceed those needed to maintain normal physiology.
In mammals SLIRP forms a complex with the mitochondrial protein LRPPRC [16–18] and the complex is required for the stability of mt-mRNAs, polyadenylation and coordinated mitochondrial translation [15–17]. SLIRP is predicted to localize to mitochondria with a probability of 94.4% using the MitoProtII software [22] and we confirmed this prediction by using immunocytochemistry to show that endogenous SLIRP co-localizes with the mitochondrial ATPase complex (S1A Fig) in 143B cells.
To investigate the specific role of SLIRP within mitochondria in vivo, we generated a germline Slirp knockout (KO) mouse model (Slirp-/-) via excision of the floxed exon 2 of Slirp by expressing the Cre-recombinase under the control of the β-actin promoter (S1B Fig). The resulting Slirp+/- mice (S1C Fig) were inter-crossed to generate Slirp-/- mice and all expected genotypes were obtained at Mendelian ratios, thus showing that SLIRP, in contrast to LRPPRC [17], is not required for embryonic development. Mice lacking SLIRP were apparently healthy with no obvious phenotype, except a slight reduction in body weight (S1D Fig). In contrast to a previous report [23], we also found that lack of SLIRP does not impair fertility as crosses between Slirp-/- males or females and wild-type mice produced normal litter sizes. These findings show that the in vivo roles of SLIRP and LRPPRC are at least partly divergent.
Steady-state SLIRP levels have been shown to correlate with those of LRPPRC [14–16] and conditional KO of Lrpprc causes complete loss of SLIRP [17]. Therefore, we investigated LRPPRC levels by immunoblotting of mitochondria isolated from Slirp-/- mice and found that ~25% of the LRPPRC protein remained in heart, liver and kidney of these mice (Fig 1A). The reduction of LRPPRC protein levels upon Slirp deletion was also confirmed by immunocytochemistry on mouse embryonic fibroblasts (MEFs) derived from Slirp-/- mice (Fig 1B). Our findings thus show that low levels of LRPPRC can be maintained without forming a complex with SLIRP, whereas in wild-type mice LRPPRC and SLIRP form a stable complex in both heart and liver, as confirmed by co-immunoprecipitation (S1E Fig). We found that LRPPRC is reduced by ~50% in heart, liver and kidney mitochondria in Lrpprc+/- mice, consistent with our previous results [24], and these levels were further reduced to ~25% in the Slirp-/- mice (Fig 1A). This finding shows that low levels of LRPPRC are sufficient for healthy survival in mice lacking SLIRP, and that SLIRP, which requires LRPPRC for its stability, may have acquired a role in fine-tuning mitochondrial gene expression.
We proceeded to investigate how loss of SLIRP will impact mitochondrial gene expression and found normal levels of mtDNA in Slirp-/- mitochondria (S2A Fig). Also the transcription rates, measured by de novo transcription assays, were normal in liver mitochondria and slightly decreased in heart mitochondria, in the absence of SLIRP (Fig 1C). The occurrence of normal de novo transcription was further confirmed by the finding of unaltered steady-state levels of mature mt-tRNAs in Slirp-/- hearts (Fig 1D). In contrast, the steady-state levels of mt-mRNAs were strikingly reduced in heart (Figs 1E and S2B) and liver (S2C Fig) mitochondria isolated from Slirp-/- mice, consistent with in vitro results from SLIRP KD in cultured cells [14–16]. Our finding of reduced mt-mRNA levels despite normal de novo transcription shows that the mt-mRNA stability is reduced in the absence of SLIRP.
Next, we compared the steady-state levels of mt-mRNAs in hearts of Slirp-/- and Lrpprc conditional KO mice at 12 weeks (Figs 1E and S2B), which is the age at which the Lrpprc conditional KO mice start to die [17]. We found less pronounced decrease of mt-mRNA steady-state levels, with the exception of Nd1, Nd2 and Nd5, in the Slirp-/- hearts in comparison with Lrpprc conditional KO hearts (Figs 1E and S2B). This differential decrease in mt-mRNA stability could be accounted for by the fact that there is a remaining fraction of LRPPRC protein in the Slirp-/- mice, whereas LRPPRC is completely absent in the hearts of Lrpprc conditional KO mice. It is surprising that the Slirp-/- mice are apparently healthy, with the exception of a slight weight loss (S1D Fig), despite such profound reduction (50–70%) of the mt-mRNA steady-state levels in all investigated tissues. This finding shows that mt-mRNAs are in significant excess in vivo and suggests that respiratory chain (RC) dysfunction only occurs if the transcript levels drop below a certain minimal threshold.
In mammalian mitochondria, mRNAs, with the exception of Nd6, contain short poly(A) tails [8,9], which are necessary to complete the termination codon for seven of the total 11 mt-mRNAs. The LRPPRC-SLIRP complex, which is involved in mt-mRNA stability, has been found to maintain polyadenylation [15,17], but the role for poly(A) tails in the regulation of mt-mRNA stability is unclear. However, polyadenylation appears to have roles in mitochondrial translation that are distinct from termination codon formation [25,26], which is consistent with the requirement of LRPPRC for coordinated translation in mammalian mitochondria [16,17]. Interestingly, we found that poly(A) tail length was intact in the absence of SLIRP (Figs 1F and S2D), which shows that SLIRP is not required for the in vivo maintenance of mitochondrial polyadenylation. This finding also demonstrates that the presence of poly(A) tails is not sufficient to ensure mt-mRNA stability in vivo when SLIRP is lost (Fig 1E and 1F). LRPPRC has been shown to promote polyadenylation by the mitochondrial poly(A) polymerase [15,27] and our findings show that normal poly(A) tail length can be maintained even if the levels of LRPPRC are low, as it is the case in the Lrpprc+/- and Slirp-/- mice (Fig 1A and 1F). A corollary of this is that SLIRP may have an additional function besides maintaining mt-mRNA stability as part of the LRPPRC-SLIRP complex.
Next we investigated how decreased levels of mt-mRNAs affected the protein synthesis machinery in mitochondria from Slirp-/- mice. We measured the steady-state levels of the mitochondrial 12S and 16S rRNAs and found that they were increased in Slirp-/- relative to control mice (Figs 1E and S2B and S2C). The increase in 16S rRNA correlated with an increased amount of MRPL37, a mitochondrial ribosomal protein (MRP) of the large subunit, in Slirp-/- heart and liver mitochondria (S3A Fig). This apparent increase in mitoribosome biogenesis is presumably a compensatory response to the reduced mt-mRNA stability observed upon SLIRP loss.
To assess the state of association of mt-mRNAs with the mitoribosome we performed sucrose sedimentation gradient analyses of mitochondrial extracts. We used qRT-PCR to determine the sedimentation profile of the small (28S) and large (39S) ribosomal subunits and the fully assembled mitoribosome (55S) (Figs 2A and S3C). The 12S rRNA co-migrated with the MRPS35 protein of the 28S subunit and the 16S rRNA co-migrated with MRPL37 (Figs 2A and S3C, left panels), which shows that the 28S subunit was mainly present in fractions 6–7, the 39S subunit in fraction 9 and the 55S mitoribosome in fractions 11–12 in control mice. Strikingly, in Slirp-/- mitochondria of liver (Fig 2A, right panel) and heart (S3C Fig, right panel), the ribosome profiles were altered as shown by the continuous distribution of MRPL37 and 16S rRNA between fractions 9 and 12. This continuous distribution may occur as a consequence of the increased steady-state levels of mt-rRNAs and MRPs (S3A Fig). We also measured the abundance of mt-mRNAs in the different fractions of the gradient by qRT-PCR and could identify two distinct pools, one translationally inactive in fractions 4–5 and a second one, translationally active, that co-migrates with the assembled mitoribosome (Figs 2A and S3C). To investigate the proportion of mt-mRNAs engaged with the mitoribosome without being misled by the global decrease of mt-mRNA levels in Slirp-/- mice, we normalized Slirp-/- mt-mRNA levels to those of the control samples (Fig 2B). Interestingly, after normalization, we found that mt-mRNAs were less engaged with the assembled mitoribosome in the Slirp-/- liver mitochondria in comparison with controls (Fig 2B). Strikingly, the profile was the opposite in heart where we found increased engagement of mt-mRNAs with the 55S mitoribosome in the Slirp-/- heart mitochondria in comparison with controls (S3D Fig).
Next we investigated the association of mt-mRNAs with the mitoribosome by performing RNA sequencing (RNAseq) of fractions from liver mitochondria that corresponded to the 28S and 39S subunits and to the 55S mitoribosome. In addition we carried out RNAseq of the fractions between the 39S subunit and the 55S mitoribosome, as we observed a continuous distribution of large subunit proteins and rRNA in this region of the gradient in Slirp-/- mitochondria. Differential expression analyses of the mt-mRNAs indicate a global and dramatic decrease of their abundance across the ribosomal fractions in liver mitochondria where SLIRP is lost (Fig 2C), which is in line with the reduced mt-mRNA steady-state levels previously assessed (S2C Fig). The levels of the Nd6 mt-mRNA associated with the mitoribosome were not affected by the loss of SLIRP (Fig 2C), suggesting that the association of Nd6 with the mitoribosome is possibly independent from the LRPPRC-SLIRP complex. Furthermore, both Slirp+/+ and Slirp-/- datasets revealed that mt-mRNAs preferentially co-migrate with the 55S mitoribosome and with the 28S subunit as most mt-mRNAs were found in fractions 6 and 11–12 (Fig 2C). We observed a greater enrichment of mt-mRNAs, albeit to varying extents for different mt-mRNAs, with the 28S subunit compared to the 39S subunit, indicating that mt-mRNAs engage the small subunit initially, as is the case for bacterial and cytoplasmic ribosomes [28]. This trend was also observed in the Slirp-/- datasets, despite the significant reduction in mt-mRNAs (Fig 2C). The levels of mitochondrial transcripts in Slirp-/- ribosomal fractions relative to the levels in the corresponding Slirp+/+ fractions (Fig 2D), confirmed the global decrease in the abundance of all mt-mRNAs across those fractions, with the exception of Nd6. In addition we confirmed the increase in the mt-rRNA levels in Slirp-/- liver mitochondria, as previously shown by qRT-PCR (S2C Fig). Furthermore we observed a reduced presence of mt-mRNAs in fractions 11 and 12 (Fig 2D), confirming our finding that mt-mRNAs were less engaged with the assembled mitoribosome in the Slirp-/- liver mitochondria in comparison with controls (Fig 2B). Interestingly, the greatest decrease in mt-mRNA levels was found in fractions 6 and 9 (Fig 2D), suggesting that loss of SLIRP may affect the ordered association or disassembly of the ribosomal components with mt-mRNAs and that SLIRP may have a role in regulating the presentation of mature mt-mRNA to the mitochondrial ribosome.
We proceeded to measure mitochondrial protein synthesis to assess the biological significance of the altered engagement of mt-mRNAs with the mitoribosome in the absence of SLIRP. We determined the rate of translation by following 35S-methionine incorporation into newly synthetized mitochondrial polypeptides over time in MEFs (Fig 3A) and in isolated heart, liver (Fig 3B) and kidney (S4A Fig) mitochondria. Interestingly, we found that the translation rate was impaired in the Slirp-/- MEFs as well as liver and kidney mitochondria (Figs 3A and S4A), which is in line with the observed reduced engagement of the mt-mRNAs with the mitoribosome (55S) in Slirp-/- liver mitochondria (Fig 2B and 2C). In contrast, in Slirp-/- heart mitochondria the incorporation of 35S-methionine was comparable to that of control heart mitochondria (Fig 3B), with the exception of Nd2 and Cox1/Nd4 whose translation seemed to be affected by the loss of SLIRP. The maintenance of a comparable translation rate despite Slirp knockout is consistent with the observed increased engagement of mt-mRNAs with the 55S mitoribosome in Slirp-/- hearts (S3D Fig). These findings suggest that SLIRP is involved in presenting mature mRNAs to the mitoribosome in order to promote mitochondrial translation, but that its loss can be compensated for in certain tissues such as the heart. We found that the steady-state levels of the mitochondrial translation initiation factor 3 (mtIF3) were increased in Slirp-/- mitochondria in comparison with controls, especially in the liver (S3B Fig), which likely constitutes a compensatory response to the impaired rate of translation. The tissue-specific mitochondrial translation defect, which is very minor in the heart and more apparent in the liver and kidney, does not seem to impact the assembly of the RC subunits, as their steady-state levels are similar in liver and heart mitochondria from control and Slirp-/- mice (Fig 3C). Furthermore the respiration was normal under phosphorylating, non-phosphorylating, and uncoupled conditions in Slirp-/- mitochondria from liver (Fig 3D) and heart (S4C Fig). Consistently, the RC enzyme activities of complexes I, II and IV were comparable in mitochondria from liver (Fig 3E) and heart (S4D Fig) in Slirp-/- and control mice.
Together, these data argue that SLIRP can act as a general activator of mitochondrial translation, whose loss (i) can be overcome by an unknown mechanism in tissues such as the heart and (ii) is not sufficient to induce OXPHOS dysfunction in tissues where translation rate is affected such as the liver and kidney. We hypothesize that despite their reduced stability and impaired loading onto the mitoribosome, mt-mRNAs in Slirp-/- mice are still translated at rates sufficient to preserve normal OXPHOS activity, which likely explains the absence of a pathophysiology in the Slirp-/- mice under basal conditions. In addition, the stability of the mitochondria-encoded RC subunits as observed by 35S-methionine pulse-chase assay in the Slirp-/- MEFs (S4B Fig) is likely contributing to the maintenance of normal OXHPOS function despite the decrease in translation.
In plants, PPR proteins have been found to associate via protein-protein interactions with additional RNA-binding proteins, including RRM proteins to regulate gene expression [29]. The co-stabilization of SLIRP and LRPPRC as a complex [16–18] has provided a challenge to specifically decipher their individual roles in mitochondria. Our data suggest that the decreased levels of mt-mRNA in Slirp-/- mitochondria could possibly be a consequence of the decreased LRPPRC protein levels, where SLIRP could act as a stabilizing partner for LRPPRC without directly affecting mt-mRNA stability. To further investigate this hypothesis, we generated mice ubiquitously overexpressing Lrpprc on a Slirp KO background (Fig 4A), in an attempt to overcome the co-stability dependence of the two proteins. Interestingly, we found that LRPPRC protein levels could not be restored in the absence of SLIRP thus confirming that SLIRP is essential for LRPPRC protein stabilization. Mitochondrial protein turnover is regulated by several proteases [30], among which LONP1 has been shown to target components of the mitochondrial gene expression machinery [31,32]. By knocking down the expression of Lonp1 in Slirp-/- MEFs we could partially restore LRPPRC protein levels, demonstrating that LRPPRC is targeted for degradation by this matrix protease in the absence of SLIRP (Fig 4B). We used this rescue model to determine if the increased steady-state levels of LRPPRC would restore mt-mRNA levels in the absence of SLIRP (Fig 4C). However, the significant rescue of LRPPRC steady-state levels induced by the Lonp1 KD (Fig 4B) did not significantly increase mt-mRNA levels (Fig 4C), yet in the absence of any adverse effect on the mt-RNA degradation machinery (Fig 4B) [33]. In previous work we have shown that Lrpprc+/- mice, that have a ~50% reduction of LRPPRC protein levels, have normal mitochondrial transcript stability [24]. In contrast, we show here that a similar LRPPRC level reduction combined with loss of SLIRP, as seen in Slirp-/- MEFs upon Lonp1 KD (Fig 4B), induced a significant reduction in the steady-state levels of mitochondrial transcripts (Fig 4C). Taken together, these results show that SLIRP has a role in mt-mRNA stability that can be disconnected from its function in stabilizing LRPPRC. We thus conclude that both LRPPRC and SLIRP are required for maintaining mt-mRNA steady-state levels independent of their roles in stabilizing the other partner of the LRPPRC-SLIRP complex.
It has previously been shown that the stability of SLIRP is absolutely dependent on the presence of LRPPRC [14–18]. Here, we show that a small fraction of LRPPRC can be maintained even if SLIRP is absent. It should be noted that SLIRP is necessary for maintaining normal levels of LRPPRC, which can otherwise be degraded by mitochondrial matrix proteases such as LONP1. Beyond their roles in co-stabilization, SLIRP and LRPPRC share a common direct role on mitochondrial transcript stability as we have shown that both proteins are required to maintain mt-mRNA steady-state levels. Indeed, mt-mRNA stability could not be restored by the sole rescue of LRPPRC levels in the absence of SLIRP. An alternative hypothesis is that the rescued LRPPRC is not fully functional in the absence of SLIRP and can therefore not fulfill its mt-mRNA stabilizing function. This is however very unlikely as we have shown that low levels of LRPPRC, independent of the LRPPRC-SLIRP complex, are sufficient, and therefore functional, for mt-mRNA poly(A) tail maintenance.
Interestingly we show that in contrast to its partner LRPPRC, SLIRP is not involved in the maintenance of the poly(A) tails of mt-mRNAs in vivo. This result was surprising given the reduced poly(A) tail abundance and subsequent accumulation of mt-mRNA oligo(A) tails reported upon SLIRP KD in cells [15], but was in line with the observation by the same authors that LRPPRC alone could stimulate mt-mRNA polyadenylation in vitro and that SLIRP only had a supportive role in this assay through the stabilization of LRPPRC [15]. However this last observation contrasts with an in vitro study showing that the extension of the poly(A) tail was enhanced when LRPPRC was complexed with SLIRP, compared to LRPPRC alone [27]. This is likely because recombinant PPR proteins can be unstable and prone to precipitation [34], and LRPPRC would require SLIRP for its in vitro stability and thereby would enhance its intrinsic activity required for poly(A) tail maintenance.
We find that SLIRP is not involved in poly(A) tail maintenance in vivo, but instead has a role in fine-tuning the rate of mitochondrial protein synthesis. Indeed, we have shown using RNAseq that SLIRP can globally facilitate the ordered association of mature mt-mRNAs with mitoribosome components, thereby affecting the rate of translation. The only exception was Nd6 mRNA, whose stability required the presence of LRPPRC and SLIRP but whose engagement with the mitoribosome seemed in contrast to be unaffected by the loss of SLIRP. Whether this independence is conferred by the absence of a poly(A) tail is not clear. Furthermore, as mentioned above, the residual levels of LRPPRC are sufficient to stabilize the poly(A) tails of mt-mRNAs, enabling normal protein synthesis in the heart but not in the liver and kidney. This effect is independent of the interaction between LRPPRC and SLIRP as we have confirmed that this complex is present both in mouse heart and liver mitochondria. The apparent tissue-specific effect of SLIRP loss on mitochondrial translation could be explained by an unknown mechanism compensating for the absence of SLIRP in the heart. We found that the translation rate was also impaired in the Slirp-/- MEFs and we therefore hypothesize that the consequence of SLIRP loss could be linked to the proliferative status of the tissue. Indeed, faster rates of mitochondrial translation may be required in proliferating cells such as hepatocytes and MEFs. The role of SLIRP in maintaining the translation rate would therefore be better illustrated in proliferative tissues, where its absence would confer a more obvious disadvantage. However, irrespective of the effects on protein synthesis, polypeptides are made in sufficient amounts for proper assembly of the OXPHOS complexes and SLIRP loss does not compromise mitochondrial respiration. The observation that a moderate decrease in mitochondrial translation does not lead to a reduced abundance of the steady state levels of mitochondria-encoded RC subunits is likely due to the stability of the RC subunits.
Notably, mice lacking SLIRP are apparently healthy, with the exception of a slight weight loss, despite having a profound (50–70%) depletion of mt-mRNAs. The levels of the mtDNA-encoded mt-mRNAs are thus present in great excess under normal physiological conditions. It is interesting to speculate that the excess of transcripts would enable robust and rapid activation of mitochondrial protein synthesis in response to sudden changes in metabolic demand. Our findings also indicate that excess mt-mRNAs could provide a buffer that can cope with dramatic reduction of transcription of mtDNA-encoded genes, as might occur when mtDNA undergoes replication in rapidly dividing cells.
The targeting vector for disruption of Slirp in ES cells (derived from C57BL/6N mice) was generated by using BAC clones from the C57BL/6J RPCI-23 BAC library by Taconic Artemis. To generate a conditional Slirp knockout (KO) allele, exon 2 of the Slirp locus was flanked by loxP sites. The puromycin resistance marker (PuroR) was flanked by F3 sites and inserted into intron 2. The puromycin resistance cassette was removed by mating Slirp+/loxP-Puro mice with transgenic mice ubiquitously expressing Flp recombinase. The resulting Slirp+/loxP mice were mated with mice ubiquitously expressing the Cre recombinase under the dependence of the β- actin promoter to generate Slirp heterozygous KO (Slirp+/-) animals. The Slirp+/- mice were then backcrossed to C57BL/6N mice for seven generations and intercrosses were used to generate wild-type (WT, Slirp+/+) and Slirp homozygous KO (Slirp-/-) animals.
This study was performed in strict accordance with the recommendations and guidelines of the Federation of European Laboratory Animal Science Associations (FELASA). The protocol was approved by the “Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen”.
The mice were housed in specific pathogen-free conditions with a 12 hr light-dark cycle and had free access to water and food. Phenotypical characterization was performed at the German Mouse Clinic on 15 Slirp+/+ and 15 Slirp-/- mice of each gender aged between 9 and 21 weeks. At the German Mouse Clinic, the mice were maintained according to the GMC housing conditions and German laws and the tests were performed as outlined in the standard operating procedures (SOP) linked to the EMPReSS website http://empress.har.mrc.ac.uk
The MPAT assay was adapted from previous protocols [15,35]. RNA was extracted with TRIzol Reagent (Invitrogen) from heart mitochondria. An adaptor DNA oligonucleotide (sequence in S1 Table) was phosphorylated by the T4 PNK (New England Biolabs) and 2.5 pmol of this phosphorylated adaptor DNA oligonucleotide was then ligated to the 3’ end of 0.3 μg of total mitochondrial RNA, for each RNA species to be tested. The ligation reaction was performed using the T4 RNA ligase (New England Biolabs) for 2 hrs at 37°C. The ligated RNA was extracted with the TRIzol Reagent and reverse transcribed using the High capacity cDNA reverse transcription kit (Applied Biosystems) and a primer specific of the adaptor DNA oligonucleotide sequence (anti-adaptor, sequence in S1 Table). A first round of PCR was carried out for 29 cycles using a gene-specific upper primer and the anti-adaptor primer. The PCR products were purified on G-50 micro columns (GE Healthcare) in order to remove the primers. A nested PCR was then carried out for 12 cycles using a gene-specific lower primer and an inner anti-adaptor primer (sequence in S1 Table), in order to improve specificity. The PCR products were then cloned into the pCR4-TOPO vector and transformed into chemically-competent bacteria. Finally, DNA from selected colonies was extracted and sequenced in order to assess the length of the poly(A) stretch on the 3’ end of each mt-mRNA species.
In organello transcription and translation assays were performed on mitochondria isolated from mouse tissues by differential centrifugation as detailed in the S1 Text. Mitochondria, 800 μg, were collected for each in organello transcription assay and washed in 1 ml of transcription buffer (10 mM Tris pH 7.4, 25 mM sucrose, 75 mM sorbitol, 100 mM KCl, 10 mM K2HPO4, 50 μM EDTA, 5 mM MgCl2, 10 mM glutamate, 2.5 mM malate, 1 mg/ml BSA and 1 mM ADP). An aliquot of mitochondria was collected for immunoblotting with the VDAC antibody (Millipore) to ensure equal loading. The remaining mitochondria were pelleted by centrifugation at 10,000 g, for 3 min at 4°C, suspended in 750 μl of transcription buffer supplemented with 30 μCi of 32P-UTP (PerkinElmer) and incubated for 20 min at 37°C. After the incubation, mitochondria were washed once and suspended in 750 μl of fresh transcription buffer in the presence of 0.2 mM of cold UTP. A short chase was performed for 5 min at 37°C in order to decrease the background and mitochondria were washed three times in 10 mM Tris pH 6.8, 0.15 mM MgCl2 and 10% glycerol. The mitochondrial pellet was suspended in 1 ml TRIzol (Invitrogen) for RNA extraction according to the manufacturer’s instructions. The isolated RNAs were analyzed by northern blotting and the radiolabeled transcripts were visualized by autoradiography.
Mitochondria, 500 μg, were collected for in organello translation assays and incubated in 750 μl translation buffer (100 mM mannitol, 10 mM sodium succinate, 80 mM KCl, 5 mM MgCl2, 1 mM KPi, 25 mM HEPES pH 7.4, 5 mM ATP, 20 μM GTP, 6 mM creatine phosphate, 60 μg/ml creatine kinase and 60 μg/ml of all amino acids except methionine). An aliquot of the mitochondrial preparation was set aside for immunoblotting to ensure equal loading as described above. Mitochondria were supplemented with 150 μCi of 35S methionine (PerkinElmer) for 10, 30 or 60 min at 37°C. After labeling, mitochondria were washed in translation buffer and suspended in a SDS-PAGE loading buffer. Translation products were resolved by SDS-PAGE and analyzed by autoradiography.
The mitochondrial translation rate was also assessed in cultured primary MEFs following a previously described method [36]. The translation products were labeled for 10, 30 and 60 min with 80 μCi/ml of a mixture of 35S-methionine/cysteine (Perkin Elmer) in DMEM lacking methionine and cysteine and in the presence of 100 μl/ml of the cytoplasmic translation inhibitor emetine (Sigma). After pulse labeling, a short chase was performed for 5 min at 37°C to decrease the background. The cells were then washed and lysed with RIPA lysis buffer. Protein concentration was measured and 50 μg of total cell extracts were resolved by SDS-PAGE and analyzed by autoradiography.
Knockdown of the mitochondrial LONP1 protease was performed on Slirp+/+ and Slirp-/- primary MEFs plated on 10 cm diameter dishes. MEFs at 80% confluence were transfected with 1.4 μg of either scrambled or Lonp1 siRNA (Stealth siRNA negative control, Med. GC and Stealth siRNA Lonp1 respectively, Life technologies) in 12 μl of Lipofectamine RNAi Max (Invitrogen) per dish. Cells were harvested after 72 hrs either in TRIzol Reagent (Invitrogen) for RNA extraction or in RIPA lysis buffer [37] for total cell protein extraction.
For detection of mitochondrial RNAs, northern blotting and qRT-PCR were performed as described in the S1 Text.
Protein steady-state levels were assessed by immunoblotting as described in the S1 Text.
Heart and liver mitochondria, 1.2 mg, were lysed in the presence of 1% n-Dodecyl β-D-maltoside (Sigma). Lysates were loaded on 10–30% sucrose gradients and separated by centrifugation overnight as previously described [17,38]. Gradient fractions were collected as 750 μl aliquots. RNA was extracted from one third of each fraction by using the TRIzol LS Reagent (Invitrogen) according to the manufacturer’s recommendations. The samples were subsequently treated with DNase I and used for cDNA synthesis. The transcript abundance in each fraction was assessed by qRT-PCR analysis using the Taqman probes listed in S1 Table. The remaining two thirds of each fraction (500 μl) were precipitated with trichloracetic acid, resolved by SDS-PAGE and ribosome-containing fractions were detected by immunoblotting using antibodies specific for individual proteins from the 28S (MRPS35, Proteintech) and 39S (MRPL37, Sigma) ribosomal subunits.
RNA from mitochondrial sucrose gradient fractions 6 and 9–12 was isolated and the concentration, purity, and integrity were confirmed using a BioAnalyser. The libraries were constructed using the Illumina TruSeq Sample Prep Kit and deep sequencing of the mitochondrial RNAs was performed by the Cologne Center for Genomics at the University of Cologne on an Illumina MiSeq according to the manufacturer’s instructions. Raw sequencing reads were aligned to the mouse mitochondrial genome (chrM; mm10) with Bowtie2 v2.2.4 (-p 20—very-sensitive) [39]. Gene-specific read counts were summarised with featureCounts [40] from the Subread package v1.3.5-p4 (-b-T 20-s 2) using the Ensembl 78 (GENCODE VM4) annotation, modified to merge Nd4L/Nd4 and Atp8/Atp6 annotations into bicistronic transcripts. Raw fragment counts were normalised as fragments per kilobase per million mapped reads (FPKM) and expression changes calculated as log2 fold changes of FPKM values. Heat maps were made with gplots v2.15.0 and transcripts were hierarchically clustered by complete linkage of their euclidean distances with the hclust and dist functions of the base stats package in R v3.1.2.
The mitochondrial oxygen consumption flux and the respiratory chain complex activities were measured as described in the S1 Text and in previous works [41].
|
10.1371/journal.pcbi.1006080 | Bamgineer: Introduction of simulated allele-specific copy number variants into exome and targeted sequence data sets | Somatic copy number variations (CNVs) play a crucial role in development of many human cancers. The broad availability of next-generation sequencing data has enabled the development of algorithms to computationally infer CNV profiles from a variety of data types including exome and targeted sequence data; currently the most prevalent types of cancer genomics data. However, systemic evaluation and comparison of these tools remains challenging due to a lack of ground truth reference sets. To address this need, we have developed Bamgineer, a tool written in Python to introduce user-defined haplotype-phased allele-specific copy number events into an existing Binary Alignment Mapping (BAM) file, with a focus on targeted and exome sequencing experiments. As input, this tool requires a read alignment file (BAM format), lists of non-overlapping genome coordinates for introduction of gains and losses (bed file), and an optional file defining known haplotypes (vcf format). To improve runtime performance, Bamgineer introduces the desired CNVs in parallel using queuing and parallel processing on a local machine or on a high-performance computing cluster. As proof-of-principle, we applied Bamgineer to a single high-coverage (mean: 220X) exome sequence file from a blood sample to simulate copy number profiles of 3 exemplar tumors from each of 10 tumor types at 5 tumor cellularity levels (20–100%, 150 BAM files in total). To demonstrate feasibility beyond exome data, we introduced read alignments to a targeted 5-gene cell-free DNA sequencing library to simulate EGFR amplifications at frequencies consistent with circulating tumor DNA (10, 1, 0.1 and 0.01%) while retaining the multimodal insert size distribution of the original data. We expect Bamgineer to be of use for development and systematic benchmarking of CNV calling algorithms by users using locally-generated data for a variety of applications. The source code is freely available at http://github.com/pughlab/bamgineer.
| We present Bamgineer, a software program to introduce user-defined, haplotype-specific copy number variants (CNVs) at any frequency into standard Binary Alignment Mapping (BAM) files. Copy number gains are simulated by introducing new DNA sequencing read pairs sampled from existing reads and modified to contain SNPs of the haplotype of interest. This approach retains biases of the original data such as local coverage, strand bias, and insert size. Deletions are simulated by removing reads corresponding to one or both haplotypes. In our proof-of-principle study, we simulated copy number profiles from 10 cancer types at varying cellularity levels typically encountered in clinical samples. We also demonstrated introduction of low frequency CNVs into cell-free DNA sequencing data that retained the bimodal fragment size distribution characteristic of these data. Bamgineer is flexible and enables users to simulate CNVs that reflect characteristics of locally-generated sequence files and can be used for many applications including development and benchmarking of CNV inference tools for a variety of data types.
| The emergence and maturation of next-generation sequencing technologies, including whole genome sequencing, whole exome sequencing, and targeted sequencing approaches, has enabled researchers to perform increasingly more complex analysis of copy number variants (CNVs)[1]. While genome sequencing-based methods have long been used for CNV detection, these methods can be confounded when applied to exome and targeted sequencing data due to non-contiguous and highly-variable nature of coverage and other biases introduced during enrichment of target regions[1–5]. In cancer, this analysis is further challenged by bulk tumor samples that often yield nucleic acids of variable quality and are composed of a mixture of cell-types, including normal stromal cells, infiltrating immune cells, and subclonal cancer cell populations. Circulating tumor DNA presents further challenges due to a multimodal DNA fragment size distribution and low amounts of tumor-derived DNA in blood plasma. Therefore, development of CNV calling methods on arbitrary sets of tumor-derived data from public repositories may not reflect the type of tumor specimens encountered at an individual centre, particularly formalin-fixed-paraffin embedded tissues routinely profiled for diagnostic testing.
Due to lack of a ground truth for validating CNV callers, many studies have used simulation to model tumor data[6]. Most often, simulation studies are used in an ad-hoc manner using customized formats to validate specific tools and settings with limited adaptability to other tools. More generalizable approaches aim at the de novo generation of sequencing reads according to a reference genome (e.g. wessim[3], Art-illumina[7], and dwgsim[8]. However, de novo simulated reads do not necessarily capture subtle features of empirical data, such as read coverage distribution, DNA fragment insert size, quality scores, error rates, strand bias and GC content[6]; factors that can be more variable for exome and targeted sequencing data particularly when derived from clinical specimens. Recently, Ewing et al. developed a tool, BAMSurgeon, to introduce synthetic mutations into existing reads in a Binary alignment Mapping (BAM) file[9]. BAMSurgeon provides support for adjusting variant allele fractions (VAF) of engineered mutations based on prior knowledge of overlapping CNVs but does not currently support direct simulation of CNVs themselves.
Here we introduce Bamgineer, a tool to modify existing BAM files to precisely model allele-specific and haplotype-phased CNVs (Fig 1). This is done by introducing new read pairs sampled from existing reads, thereby retaining biases of the original data such as local coverage, strand bias, and insert size. As input, Bamgineer requires a BAM file and a list of non-overlapping genomic coordinates to introduce allele-specific gains and losses. The user may explicitly provide known haplotypes or chose to use the BEAGLE[10] phasing module that we have incorporated within Bamgineer. We implemented parallelization of the Bamgineer algorithm for both standalone and high performance computing cluster environments, significantly improving the scalability of the algorithm. Overall, Bamgineer gives investigators complete control to introduce CNVs of arbitrary size, magnitude, and haplotype into an existing reference BAM file. We have uploaded all software code to a public repository (http://github.com/pughlab/bamgineer)).
For all proof-of-principle experiments, we used exome sequencing data from a single normal (peripheral blood lymphocyte) DNA sample. DNA was captured using the Agilent SureSelect Exome v5+UTR kit and sequenced to 220X median coverage as part of a study of neuroendocrine tumors. Reads were aligned to the hg19 build of the human genome reference sequence and processed using the Genome Analysis Toolkit (GATK) Best Practices pipeline.
Following the validation of our tool for readily-detected chromosome- and arm-level events, we next used Bamgineer to simulate CNV profiles mimicking 3 exemplar tumors from each of 10 different cancer types profiled by The Cancer Genome Atlas using the Affymetrix SNP6 microarray platform: lung adenocarcinoma (LUAD); lung squamous cell (LUSC); head and neck squamous cell carcinoma (HNSC); glioblastoma multiformae (GBM); kidney renal cell carcinoma (KIRC); bladder (BLCA); colorectal (CRC); uterine cervix (UCEC); ovarian (OV), and breast (BRCA) cancers (Table 1).
To select 3 exemplar tumors for each cancer type, we chose profiles that best represented the copy number landscape for each cancer type. First, we addressed over-segmentation of the CNV calls from the microarray data by merging segments of <500 kb in size with the closest adjacent segment and removing the smaller event from the overlapping gain and loss regions. We then assigned a score to each tumor that reflects its similarity to other tumor of the same cancer type (S7 Fig). This score integrates total number of CNV gain and losses (Methods, Eq 6), median size of each gain and loss, and the overlap of CNV regions with GISTIC peaks for each cancer type as reported by The Cancer Genome Atlas (Table 1). We selected three high ranking tumors for each cancer type such that, together, all significant GISTIC[15] peaks for that tumor type were represented. A representative profile from a single tumor is shown in Fig 2C.
Subsequently, for each of the 30 selected tumor profiles (3 for each of 10 cancer types), we introduced the corresponding CNVs at 5 levels of tumor cellularity (20, 40, 60, 80, and 100%) resulting in 150 BAM files in total. For each BAM file, we used Sequenza to generate allele-specific copy number calls as done previously. Tumor/normal log2 ratios are shown in Fig 3 for one representative from each cancer type. From this large set of tumors, we next set out to compare Picard metrics and CNV calls as we did for the arm- and chromosome-level pilot.
We evaluated Bamgineer using several metrics: tumor allelic ratio, SNP phasing consistency, and tumor to normal log2 ratios (Fig 4). As expected, across all regions of a single copy gain, tumor allelic ratio was at ~0.66 (interquartile range: 0.62–0.7) for the targeted haplotype and 0.33 (interquartile range: 0.3–0.36) for the other haplotype. As purity was decreased, we observed a corresponding decrease in allelic ratios, from 0.66 down to 0.54 (interquartile range: 0.5–0.57) for targeted and an increase (from 0.33) to 0.47 (interquartile range: 0.43–0.5) for the other haplotype for 20% purity (Fig 4A and 4B). These changes correlated directly with decreasing purity (R2 > 0.99) for both haplotypes. Similarly, for single copy loss regions, as purity was decreased from 100% to 20% the allelic ratio linearly decreased (R2 > 0.99) from ~0.99 (interquartile range: 0.98–1.0) for targeted haplotype to ~0.55 (interquartile range: 0.51–0.58) for targeted haplotype and increases from 0 to ~0.43 (interquartile range: 0.4–0.46) for the other haplotype (Fig 4B). The results for log2 tumor to normal depth ratios of segments normalized for average ploidy were also consistent with the expected values (Methods, Eq 2). For CNV gain regions, log2 ratio decreased from ~0.58 (log2 of 3/2) to ~0.13 as purity was decreased from 100% to 20%. For CNV loss regions, as purity was decreased from 100% to 20%, the log2 ratio increased from -1 (log2 of 1/2) to -0.15, consistent with Eq 2 (Fig 4C; S1-S4 for individual cancers).
Ultimately, we wanted to assess whether Bamgineer was introducing callable CNVs consistent with segments corresponding to the exemplar tumor set. To assess this, we calculated an accuracy metric (Fig 4D) as:
accuracy=TP+TFTP+TF+FP+FN
where TP, TF, FP and FN represent number of calls from Sequenza corresponding to true positives (perfect matches to desired CNVs), true negatives (regions without CNVs introduced), false positives (CNV calls outside of target regions) and false negatives (target regions without CNVs called). TP, TF, TN, FN were calculated by comparing Sequenza absolute copy number (predicted) to the target regions for introduction of 1 Mb CNV bins across the genome. As tumor content decreased, accuracy for both gains and losses decreased as false negatives became increasingly prevalent due to small shifts in log2 ratios. We note that (as expected), decreasing cancer purity from 100% to 20% generally decreases the segmentation accuracy. Additionally, we observe that segmentation accuracy is on average, significantly higher for gain regions compared to the loss regions for tumor purity levels below 40% (Fig 4D). This is consistent with previous studies that show the sensitivity of CNV detection from sequencing data is slightly higher for CNV gains compared to CNV losses[16]. We also note that with decreasing cancer purity, the decline in segmentation accuracy follows a linear pattern of decline for gain regions and an abrupt stepwise decline for loss regions (Fig 4D; segmentation accuracies are approximately similar for 40% and 20% tumor purities).
Finally, we observed a degree of variation in terms of segmentation accuracy across individual cancer types (S1–S4 Figs). Segmentation accuracy was lower for LUAD, OV and UCEC compared to other simulated cancer types for this study. The relative decline in performance is seen in cancer types where CNV gains and losses cover a sizeable portion of the genome; and hence, the original loss and gain events sampled from TCGA had significant overlaps. As a result, after resolving overlapping gain and loss regions (S7 Fig), on average, the final target regions constitute a larger number of small (< 200 kb) loss regions immediately followed by gain regions and vice versa; making the accurate segmentation challenging for the CBS (circular binary segmentation) algorithm implemented by Sequenza relying on presence of heterozygous SNPs. This can cause uncertainties in assignments of segment boundaries.
In summary, application of an allele-specific caller to BAMs generated by Bamgineer recapitulated CNV segments consistent with >95% (medians: 95.1 for losses and 97.2 for gains) of those input to the algorithm. However, we note some discrepancies between the expected and called events, primarily due to small CVNs as well as large segments of unprobed genome between exonic sequences.
To evaluate the use of Bamgineer for circulating tumor DNA analysis, we simulated the presence of an EGFR gene amplification in read alignments from a targeted 5-gene panel (18 kb) applied to a cell-free DNA from a healthy donor and sequenced to >50,000X coverage. To mirror concentrations of tumor-derived fragments commonly encountered in cell-free DNA[17,18], we introduced gain of an EGFR haplotype at frequencies of 100, 10, 1, 0.1, and 0.01%. This haplotype included 3 SNPs covered by our panel, which were phased and subject to allele-specific gain accordingly. As with the exome data, we observed shifts in coverage of specific allelic variants, and haplotype representation consistent with the targeted allele frequencies (Fig 5A, Supplemental S1 Table). Furthermore, read pairs introduced to simulate gene amplification retain the bimodal insert size distribution characteristic of cell-free DNA fragments (Fig 5B and 5C).
While this experiment showcases the ability of Bamgineer to faithfully represent features of original sequencing data while controlling allelic amplification at the level of the individual reads, these subtle shifts are currently beyond the sensitivity of conventional CNV callers when applied to small, deeply covered gene panels. Therefore, it is our hope that Bamgineer may be of value to aid develop of new methods capable of detecting copy number variants supported by a small minority of DNA fragments in a specimen.
Bamgineer is computationally intensive and the runtime of the program is dictated by the number of reads that must be processed, a function of the coverage of the genomic footprint of target regions. To ameliorate the computational intensiveness of the algorithm, we employed a parallelized computing framework to maximize use of a high-performance compute cluster environment when available. We took advantage of two features in designing the parallelization module. First, we required that added CNVs are independent for each chromosome (although nested events can likely be engineered through serial application of Bamgineer). Second, since we did not model interchromosomal CNV events, each chromosome can be processed independently. As such, CNV regions for each chromosome can be processed in parallel and aggregated as a final step. S8 Fig shows the runtimes for The Cancer Genome Atlas simulation experiments. Using a single node with 12 cores and 128 GB of RAM, each synthetic BAM took less than 3.5 hours to generate. We also developed a version of Bamgineer that can be launched from sun grid engine cluster environments. It uses python pipeline management package ruffus to parallelize tasks automatically and log runtime events. It is highly modular and easily updatable. If disrupted during a run, the pipeline can continue to completion without re-running previously completed intermediate steps.
Here, we introduced Bamgineer, to introduce user-defined haplotype-phased allele-specific copy number events into an existing Binary Alignment Mapping (BAM) file, obtained from exome and targeted sequencing experiments. As proof of principle, we generated, from a single high coverage (mean: 220X) BAM file derived from a human blood sample, a series of 30 new BAM files containing a total of 1,693 simulated copy number variants (on average, 56 CNVs comprising 1800Mb i.e. ~55% of the genome per tumor) corresponding to profiles from exemplar tumors for each of 10 cancer types. To demonstrate quantitative introduction of CNVs, we further simulated 4 levels of tumor cellularity (20, 40, 60, 80% purity) resulting in an additional 120 new tumor BAM files. We validated our approach by comparing CNV calls and inferred purity values generated by an allele-specific CNV-caller (Sequenza[14]) as well as a focused comparison of allelic variant ratios, haplotype-phasing consistency, and tumor/normal log2 ratios for inferred CNV segments (S1–S4 Figs). In every case, inferred purity values were within ±5% of the targeted purity; and majority of engineered CNV regions were correctly called by Sequenza (accuracy > 94%; S1–S4 Figs). Allele variant ratios were also consistent with the expected values both for targeted and the other haplotypes (Median within ±3% of expected value). Median tumor/normal log2 ratios were within ±5% of the expected values.
To demonstrate feasibility beyond exome data, we next evaluated these same metrics in a targeted 5-gene panel applied to a cell-free DNA sequencing library generated from a healthy blood donor and sequenced to >10,000X coverage[17] To simulate concentrations of tumor-derived fragments typically encountered in cancer patients, we introduced EGFR amplifications at frequencies of 100, 10, 1, 0.1, and 0.01%. As with the exome data, we observed highly specific shifts in allele variant ratios, log2 coverage ratios, and haplotype representation consistent with the targeted allele frequencies. Our method also retained the bimodal DNA insert size distribution observed in the original read alignment. However, it is worthwhile noting that, these minute shifts are currently beyond the sensitivity of existing CNV callers when applied to small, deeply covered gene panels. Consequently, we anticipate that Bamgineer may be of value to aid develop of new methods capable of detecting copy number variants supported by a small minority of DNA fragments.
In the experiments conducted in this study, we limited ourselves to autosomes and to maximum total copy number to 4. Naturally, Bamgineer can readily simulate higher-level copy number states and alter sex chromosomes as well (S10 Fig). While chromosome X in diploid state (e.g. XX in normal female) is treated identically to autosomes, for both X and Y chromosomes beginning in haplotype state (e.g. XY in normal male), the haplotype phasing step is skipped and Bamgineer samples all reads on these chromosomes independently. For high-level amplifications, the ability of Bamgineer to faithfully retain the features of the input Bam file (e.g. DNA fragment insert size, quality scores and so on), depends on the intrinsic factors such as the length of the desired CNV, mean depth of coverage and fragment length distribution of the original input BAM file (see Materials and Methods).
The significance of this work in the context of CNV inference in cancer is twofold: 1) users can simulate CNVs using their own locally-generated alignments so as to reflect lab-, biospecimen-, or pipeline-specific features; 2) bioinformatic methods development can be better supported by ground-truth sequencing data reflecting CNVs without reliance on generated test data from suboptimal tissue or plasma specimens. Bamgineer addresses both problems by creating standardized sequencing alignment format (BAM files) harbouring user-defined CNVs that can readily be used for algorithm optimization, benchmarking and other purposes.
We expect our approach to be applicable to tune algorithms for detection of subtle CNV signals such as somatic mosaicism or circulating tumor DNA. As these subtle shifts are beyond the sensitivity of many CNV callers, we expect our tool to be of value for the development of new methods for detecting such events trained on conventional DNA sequencing data. By providing the ability to create customized user-generated reference data, Bamgineer will prove valuable inn development and benchmarking of CNV calling and other sequence data analysis tools and pipelines.
The work presented herein can be extended in several directions. First, Bamgineer is not able to reliably perform interchromosomal operations such as chromosomal translocation, as our focus has been on discrete regions probed by exome and targeted panels. Second, while Bamgineer is readily applicable to whole genome sequence data, sufficient numbers of reads are required for re-pairing when introducing high-level amplifications. As such, shallow (0.1-1X) or conventional (~30X) whole genome sequence data may only be amenable to introduction of arm-level alterations as smaller, focal targets may not contain sufficient numbers of reads to draw from to simulate high-level amplifications. Additionally, in our current implementation, we limited the simulated copy numbers to non-overlapping regions. Certainly, such overlapping CNV regions occur in cancer and iterative application of Bamgineer may enable introduction of complex, nested events. Finally, introduction of compound, serially acquired CNVs may be of interest to model subclonal phylogeny developed over time in bulk tumor tissue samples.
Bamgineer uses several python packages for parsing input files (pyVCF[19], VCFtools[20], and pybedtools[21], manipulating BAM files (pysam[22], Samtools[23], Sambamba[24] and BamUtils[25]), phasing haplotypes (BEAGLE[10], and distributing compute jobs in cluster environments (ruffus[26]). HaplotypeCaller from the Genome Analysis Toolkit (GATK)[27] is used to call germline heterozygous SNPs (het.vcf) if known haplotype SNP data is not provided. The analysis workflow is outlined in S9 Fig.
The user provides 2 mandatory inputs to Bamgineer as command-line arguments: 1) a BAM file containing aligned paired-end sequencing reads (“Normal.bam”), 2) a BED file containing the genome coordinates and type of CNVs (e.g. allele-specific gain) to introduce (“CNV regions.bed”). Bamgineer can be used to add four broad categories of CNVs: Balanced Copy Number Gain (BCNG), Allele-specific Copy Number Gain (ASCNG), Allele-specific Copy Number Loss (ACNL), and Homozygous Deletion (HD). For example, consider a genotype AB at a genomic locus where A represents the major and B represents the minor allele.
Bamgineer can be applied to convert that genomic locus to any of the following copy number states:
{A,B,ABB,AAB,ABB,AABB,AAAB,ABBB,…}
An optional VCF file containing phased germline calls can be provided (phased_het.vcf). If this file is not provided, Bamgineer will call germline heterozygous single nucleotide polymorphisms (SNPs) using the GATK HaplotypeCaller and then categorize alleles likely to be co-located on the same haplotypes using BEAGLE and population reference data from the HapMap project.
To obtain paired-reads in CNV regions of interest, we first intersect Normal.bam with the targeted regions overlapping user-defined CNV regions (roi.bed). This operation generates a new BAM file (roi.bam). Subsequently, depending on whether the CNV event is a gain or loss, the algorithms performs two separate steps as follows.
To introduce copy number gains, Bamgineer creates new read-pairs constructed from existing reads within each region of interest. This approach thereby avoids introducing pairs that many tools would flag as molecular duplicates due to read 1 and read 2 having start and end positions identical to an existing pair. If desired, these read pairs can be restricted to reads meeting a specific SAM flag. For our exome experiments, we used read pairs with a SAM flag equal to 99, 147, 83, or 163, i.e. read paired, read mapped in proper pair, mate reverse (forward) strand, and first (second) in pair. To enable support for the bimodal distribution of DNA fragment sizes in ctDNA, we removed the requirement for “read mapped in proper pair” and used read pairs with a SAM flag equal to 97, 145, 81, or 161. Users considering engineering of reads supporting large inserts or intrachromosomal read pairs may also want to remove the requirement for “read mapped in proper pair”. Additionally, we required that the selection of the newly paired read is within ±50%(±20% for ctDNA) of the original read size.
The newly created read- pairs are provided unique read names to avoid confusion with the original input BAM file. To enable inspection of these reads, these newly created read pairs are stored in a new BAM file, gain_re_paired_renamed.bam, prior to merging into the final engineered BAM. Since we only consider high quality reads (i.e. properly paired reads, primary alignments and mapping quality > 30), the newly created BAM file contains fewer reads compared to the input file (~90–95% in our proof-of-principle experiments). As such, at every transition we log the ratio between number of reads between the input and output files.
High-level copy number amplification (ASCN > = 4). To achieve higher than 4 copy number amplifications, during the read/mate pairing step, we pair each read with more than one mate read (Fig 1) to generate more new reads (to accommodate the desired copy number state). Though, since as stated a small portion of newly created paired reads do not meet the inclusion criteria, we aim to create more reads than necessary in the initial phase and use the sampling to adjust them in a later phase. For instance, to simulate copy number of 6, in theory we need create two new read pairs for every input read. Hence, in the initial “re-pairing” step we aim to create four paired reads per read (instead of 3), so that the newly created Bam file includes enough number of reads (as a rule of thumb, we use read-paring window size of ~20% higher than theoretical value). It should be noted that the maximum copy number amplification that can faithfully retain the features of the input BAM file (e.g. DNA fragment insert size, quality scores and so on), depends on the intrinsic factors such as the length of the desired CNV, mean depth of coverage and fragment length distribution of the original input BAM file.
Introduction of mutations according to haplotype state. To ensure newly constructed read-pairs match the desired haplotype, we alter the base at heterozygous SNP locations (phased_het.vcf) within each read according to haplotype provided by the user or inferred using the BEAGLE algorithm. To achieve this, we iterate through the set of re-paired reads used to increase coverage (gain_re_paired_renamed.bam) and modify bases overlapping SNPs corresponding to the target haplotype (phased_het.vcf). We then write these reads to a new BAM file (gain_re_paired_renamed_mutated.bam) prior to merging into the final engineered BAM (S9 Fig).
As an illustrative example consider two heterozygous SNPs, AB and CD both with allele frequencies of ~0.5 in the original BAM file (i.e. approximately half of the reads supporting reference bases and the other half supporting alternate bases. To introduce a 2-copy gain of a single haplotype, reads to be introduced must match the desired haplotype rather than the two haplotypes found in the original data. If heterozygous AB and CD are both located on a haplotype comprised of alternative alleles, at the end of this step, 100% of the newly re-paired reads will support alternate base-pairs (e.g. BB and DD). Based on the haplotype structure provided, other haplotype combinations are possible including AA/DD, BB/CC, etc.
Sampling of reads to reflect desired allele fraction. Depending on the absolute copy number desired for the for CNV gain regions, we sample the BAM files according to the desired copy number state. We define conversion coefficient as the ratio of total reads in the created BAM from previous step (gain_repaired_mutated.bam) to the total reads extracted from original input file (roi.bam):
ρ=no.ofreadsingain_re_paired_mutated.bamno.ofreadsinroi.bam
(1)
According to the maximum number of absolute copy number (ACN) for simulated CNV gain regions (defined by the user), two scenarios are conceivable as follows.
Copy number gain example. For instance, to achieve the single copy gain (ACN = 3, e.g. ABB copy state), the file in the previous step (gain_re_paired_renamed_mutated.bam), should be sub-sampled such that on average depth of coverage is half that of extracted reads from the target regions from the original input normal file(roi.bam). Thus, the final sampling rate is calculated by dividing ½(0.5) by ρ (subsample gain_re_paired_renamed_mutated.bam such that we have half of the roi.bam depth of coverage for the region; in practice adjusted sampling rate is in the range of 0.51–0.59 i.e. 0.85 < ρ < 1 for CN = 3) and the new reads are written to a new BAM file (gain_re_paired_renamed_mutated_sampled.bam) that we then merge with the original reads (roi.bam) to obtain gain_final.bam. Similarly to obtain three copy number gain (ACN = 5) and the desired genotype ABBBB, the gain_re_paired_renamed_mutated.bam is subsampled such that depth of coverage is 3/2(1.5) that of extracted reads from the target regions from the original input normal file(note that as explained during the new paired-read generation step, we have already created more reads than needed).
To introduce CNV losses, Bamgineer removes reads from the original BAM corresponding to a specific haplotype and does not create new read pairs from existing ones. To diminish coverage in regions of simulated copy number loss, we sub-sample the BAM files according to the desired copy number state and write these to a new file. The conversion coefficient is defined similarly as the number of reads in loss_mutated.bam divided by number of reads in roi_loss.bam (> ~0.98). Similar to CNV gains, the sampling rate is adjusted such that after the sampling, the average depth of coverage is half that of extracted reads from the target regions (calculated by dividing 0.5 by conversion ratio, as the absolute copy number is 1 for loss regions). Finally, we subtract the reads in CNV loss BAMs from the input.bam (or input_sampled.bam) and merge the results with CNV gain BAM (gain_final.bam) to obtain, the final output BAM file harbouring the desired copy number events.
To validate that the new paired-reads generated from the original BAM files show similar probability distribution, we used two-sided Kolmogorov–Smirnov (KS) test. The critical D-values where calculated for α = 0.01 as follows:
Dα=c(α)n1+n2n1n2
(2)
where coefficient c(α) is obtained from Table of critical values for KS test (https://www.webdepot.umontreal.ca/Usagers/angers/MonDepotPublic/STT3500H10/Critical_KS.pdf; 1.63 for α = 0.01) and n1 and n2 are the number of samples in each dataset. To assess tumor allelic ratio consistency, for each SNP the theoretical allele frequency parameter was used as a reference point (Eq 3). Median, interquartile range and mean were drawn from the observed values for each haplotype-event pair for all the SNPs. The boxplot distribution of the allele frequencies were plotted and compared against the theoretical reference point. To assess the segmentation accuracy, we used log2 tumor to normal depth ratios of segments normalized for mean ploidy as the metric; where the mean ploidy is (Eqs 4 and 5). To benchmark the performance of segmentation accuracy, we used accuracy as the metrics. Statistical analysis was performed with the functions in the R statistical computing package using RStudio.
Theoretical expected values. The expected value for tumor allelic frequencies at heterozygous SNP loci for tumor purity level of p (1-p: normal contamination) is calculated as follows:
AF(snp)=pAFtcnt+(1−p)AFncnnpcnt+(1−p)cnn
(3)
where AFt and AFn represent the expected allele frequencies for tumor and normal and cnt and cnn the expected copy number for tumor and normal at specific SNP loci. For CNV events used in this experiment AFt are (1/3 or 2/3) for gain and (1 or 0) for loss CNVs according to the haplotype information (whether or not they are located on the haplotype that is affected by each CNV). The expected value for the average ploidy (∅^) is calculated as follows
∅^=1W(∑i=1ncngwgi+∑j=1mcnlwlj+∑i=1ncnn(W−G−L)
(4)
, where cng, cnl, cnn, wg and wl represent the expected ploidy for gain, loss and normal regions, and the length of individual gain and loss events respectively. W, G, and L represent total length (in base pairs) for gain regions, loss regions, and the entire genome (~ 3e9).
The expected log2ratio for each segment is calculated as follows
log2ratio(seg)=log2(p×cnseg+(1−p)×cnn∅^)
(5)
,where cnseg is the segment mean from Sequenza output, p tumor purity and ∅^ is the average ploidy calculated above. cnn is the copy number of copy neutral region (i.e. 2)
Similarity score to rank TCGA tumors. The similarity score for specific cancer type (c) and sampled tumor (t) is calculated as follows:
S(c,t)=1/(|2gt−Gc−Go|+|2lt−Lc−Lo|+ϵ)
(6)
, where gt, Gc, Go represent the total number of gains for specific tumor sampled from Cancer Genome Atlas (after merging adjacent regions and removing overlapping regions), median number of gains for specific tumor type, and number of gain events overlapping with GISTIC peaks respectively; lt, Lc, Lo represent the above quantities for CNV loss regions(ϵ is an arbitrary small positive value to avoid zero denominator). The higher score the closer is the sampled tumor to an exemplar tumor from specific cancer type.
|
10.1371/journal.pbio.2006738 | Evolutionary emergence of infectious diseases in heterogeneous host populations | The emergence and re-emergence of pathogens remains a major public health concern. Unfortunately, when and where pathogens will (re-)emerge is notoriously difficult to predict, as the erratic nature of those events is reinforced by the stochastic nature of pathogen evolution during the early phase of an epidemic. For instance, mutations allowing pathogens to escape host resistance may boost pathogen spread and promote emergence. Yet, the ecological factors that govern such evolutionary emergence remain elusive because of the lack of ecological realism of current theoretical frameworks and the difficulty of experimentally testing their predictions. Here, we develop a theoretical model to explore the effects of the heterogeneity of the host population on the probability of pathogen emergence, with or without pathogen evolution. We show that evolutionary emergence and the spread of escape mutations in the pathogen population is more likely to occur when the host population contains an intermediate proportion of resistant hosts. We also show that the probability of pathogen emergence rapidly declines with the diversity of resistance in the host population. Experimental tests using lytic bacteriophages infecting their bacterial hosts containing Clustered Regularly Interspaced Short Palindromic Repeat and CRISPR-associated (CRISPR-Cas) immune defenses confirm these theoretical predictions. These results suggest effective strategies for cross-species spillover and for the management of emerging infectious diseases.
| The probability that an epidemic will break out is highly dependent on the ability of the pathogen to acquire new adaptive mutations and to induce evolutionary emergence. Forecasting pathogen emergence thus requires a good understanding of the interplay between the epidemiology and evolution taking place at the onset of an outbreak. Here, we provide a comprehensive theoretical framework to analyze the impact of host population heterogeneity on the probability of pathogen evolutionary emergence. We use this model to predict the impact of the fraction of susceptible hosts, the inoculum size of the pathogen, and the diversity of host resistance on pathogen emergence. Our experiments using lytic bacteriophages and CRISPR-resistant bacteria support our theoretical predictions and demonstrate that manipulating the diversity of resistance alleles in a host population may be an effective way to limit the emergence of new pathogens.
| Understanding the factors that govern the ability of pathogens to invade a new host population is of paramount importance to design better surveillance systems and control policies. Mathematical epidemiology can provide key insights into these dynamics [1–4]. For instance, simple deterministic models identified critical vaccination thresholds, above which pathogens are driven extinct, which informed policy guidelines for vaccination campaigns [1]. However, chance events and rapid pathogen evolution can also play a critical role in determining the outcome of disease dynamics [2,4–6]. For example, recent experimental studies indicated that the dramatic size of the 2013–2016 Ebola epidemic can at least be partially explained by the acquisition of genetic mutations that increased transmissibility to humans [7,8].
Stochastic models of epidemiology can help to understand the emergence of evolving pathogen populations [5,9–13]. These models, however, often make the unrealistic assumption that the pathogen is spreading in a well-mixed and homogeneous host population, in which all hosts are equally susceptible. Although a handful of theoretical studies have shown that host heterogeneity could have an important impact on pathogen emergence, these models either relied on phenomenological or numerical approaches [12,13], or assumed that the hosts only differ in their number of contacts but not their susceptibility to pathogens [11]. Here, we extend this line of inquiry by (i) building a mechanistic model of pathogen emergence in a diverse host population, in which only some hosts are resistant to the pathogen, (ii) deriving analytical expressions for the probability of evolutionary emergence of the pathogen, and (iii) providing the first experimental test of theoretical predictions on pathogen evolutionary emergence using a bacteria–phage interaction. We demonstrate that realistic increases in the diversity of host resistance alleles strongly reduce the probability of evolutionary emergence of novel pathogens, hence suggesting new strategies to manage the emergence of diseases. Crucially, using bacteria with distinct Clustered Regularly Interspaced Short Palindromic Repeat (CRISPR) immunity and their lytic viruses (bacteriophages) [14–17], we experimentally explore the effect of host population heterogeneity on the emergence and evolution of pathogens. The experimental validation of our theoretical predictions with this microbial system confirms the ability of our mathematical model to capture the complexity of the interplay between the epidemiology and evolution of emerging pathogens in this model system.
In order to predict how the composition of host populations impacts the probability of pathogen emergence, we developed a branching process model [5,9–13]. We aimed to capture host–pathogen interactions in which different groups of individuals within a host population each carry unique resistance alleles that recognize different pathogen epitopes, and pathogens can evade recognition by acquiring “escape” mutations in the corresponding epitopes. In this model, we assume that the host population contains a fraction (1 − fR) of individuals that are fully susceptible to the pathogen, while the remaining fraction fR of the population is resistant and composed of a mixture of n host types in equal frequencies, each of which has a different resistance allele. The efficacy of resistance is assumed to be perfect (we relax this assumption in section S1.2 of S1 Text). Therefore, a pathogen with i escape mutations (i between 0 and n) can infect a fraction (1 − fR) + fRi/n of the total host population.
We further assume that a host infected with a pathogen that does not carry escape mutations transmits at rate b and dies at rate d. Host resistance prevents infection without affecting b or d. Whereas escape mutations allow the pathogen to infect a larger fraction of the host population, they also carry a fitness cost, c which causes pathogens with i escape mutations to reproduce at rate bi = b(1 − c)i. The probability of acquiring an escape mutation is a function of n, the number of resistance alleles in the population, as well as i, the number of escape mutations already encoded by the pathogen. The probability that a pathogen with i escape mutations will acquire an additional one equals ui,n = 1 − (1 − μ)n−i, where μ is the pathogen mutation rate per target site (a target site is a region of the pathogen genome where a point mutation or a deletion may allow escape from recognition by host immunity). This simplifies to ui,n ≈ μ(n − i) when the pathogen mutation rate is assumed to be small (note how the rate of escape mutations increases with (n − i)). For the sake of simplicity, we assume that escape mutations cannot revert to the ancestral types. These reversions are expected to have a negligible effect on the probability of evolutionary emergence when the target site mutation rate remains small [11]. To account for the effect of spatial structure, we assume that when a pathogen is released from an infected host, it will land with probability ϕ on the same type of host (i.e., a host susceptible to this pathogen) and with probability (1 − ϕ) on a random host from the population, which may or not be of the same type.
The expected number of secondary infections caused by a pathogen with i escape mutations in an uninfected host population is given by its basic reproduction ratio:
Ri,n=bidFi,n
(1)
where Fi,n = (ϕ + (1 − ϕ)(fR i/n + (1 − fR))) is the effective fraction of hosts that can be infected by the focal pathogen. A pathogen with n escape mutations has a basic reproduction ratio equal to R0(1 − c)n, where R0 = b/d refers to the basic reproduction ratio of the pathogen with 0 escape mutations in a fully susceptible host population. Note, however, that a pathogen with 0 escape mutations introduced in a diverse host population has a basic reproduction ratio equal to R0,n ≤ R0.
The key question we wish to address with this model is how the composition and structure of the host population determines the ultimate fate of a pathogen (i.e., extinction versus emergence, see S1 and S2 Figs). We detailed in the Materials and methods section the calculation of the probability of emergence, Pi,n, which is the probability that an inoculum of V0 pathogens with i escape mutations will not go extinct when introduced in a host population with n different resistance alleles. To understand the role of pathogen evolution in this process, we also derive the probability of evolutionary emergence, which quantifies the importance of escape mutations to pathogen emergence.
Next, we wanted to experimentally explore the validity of the above predictions. While this is challenging given the paucity of suitable empirical systems that are amenable to experimental manipulations in a timely fashion, we explored whether this could be achieved by studying the evolutionary emergence of “escape” phages against bacteria with a CRISPR–CRISPR-associated (Cas) system. This immune defense provides full protection against a phage infection by adding phage-derived sequences (known as “spacers”) in a CRISPR locus carried by the bacterial host chromosome [14]. This empirical system allowed us to overcome three important technical challenges (see details of the experimental protocols in the Materials and methods section). First, the stochastic nature of extinction requires a large number of replicate populations to measure a probability of emergence, which is possible using bacteria and phages in 96-well plates. Second, by mixing bacteria with different and unique CRISPR resistance alleles, we could manipulate the fraction of resistant hosts and the diversity in resistance alleles without affecting other traits of the host [18]. Third, unlike most other empirical systems, the mechanism of phage adaptation to CRISPR-based immunity is well known: lytic phages “escape” CRISPR resistance through mutation of their target sequence (the “protospacer”) [13,15,18,19,20].
In order to validate the model using this empirical system, we used eight CRISPR-resistant clones (also referred as bacteriophage-insensitive mutants [BIMs]) of the gram-negative Pseudomonas aeruginosa strain UCBPP-PA14, each of which carried a single and distinct spacer targeting the lytic phage DMS3vir. Each of these spacers provides full resistance to infection. For each of these eight CRISPR-resistant clones, the rate at which the phage acquires escape mutations was found to be approximately equal to 2.8*10−7 mutations/locus/replication, as determined using Luria-Delbrück experiments (see section S2.1.6 in S1 Text and S9 Fig). Using one of these BIMs, we first tested the theoretical prediction that the probability of emergence increases with the size of the virus inoculum (V0). To this end, 96 replicate populations, each composed of an equal mix of sensitive bacteria and a CRISPR-resistant clone, were exposed with five different inoculum sizes of the phage (corresponding to a mean V0 of approximately 0.3, 3, 30, 300, and 3,000 phages). After 24 hours, we measured the fraction of phage-infected bacterial populations in which emergence had occurred. Consistent with the model predictions, we observed that the larger the phage inoculum size, the higher the probability of pathogen emergence (Fig 3, dashed line). In addition, we measured the fraction of viral populations in which the phages had evolved to escape CRISPR resistance. Again, in accordance with the theory, we found that larger phage inocula were associated with an increased evolution of phage escape mutations (Fig 3, full line, Kendall, z = 3.416, tau = 0.784, p < 0.001). Furthermore, we obtained very similar results using a different empirical system consisting of the lytic phage 2972 and its gram-positive bacterial host Streptococcus thermophilus DGCC7710. In this experiment, 96 populations composed of sensitive bacteria and a CRISPR-resistant clone were infected with three different inoculum sizes of the phage. As above, we found that a larger phage inoculum led to both a higher probability of emergence and a higher probability of evolutionary emergence (S10 Fig).
Next, we tested the theoretical prediction that the probability of pathogen evolutionary emergence is highest in populations with an intermediate fraction of resistant hosts (Fig 4). For each of the eight BIMs, we generated populations composed of sensitive bacteria and a variable proportion of CRISPR-resistant bacteria, ranging from 0% to 100% in 10% increments. These populations were subsequently infected with V0 = 300 phages, and the fractions of emergence and evolutionary emergence were measured. As expected, pathogen/phage emergence dropped when the proportion of host/bacteria resistance reached a threshold level (S11 Fig). Interestingly, examination of phage evolution among emerging phage populations also confirmed that the probability of observing escape mutations is maximized for intermediate proportions of host resistance (Fig 4). Again, we obtained very consistent results with phage 2972 and S. thermophilus (S10 Fig).
We noticed substantial variation among CRISPR-resistant hosts in the observed frequencies of escape phage evolution (Fig 4). Variations in phage mutation rates are unlikely to explain this variability because, as pointed out above, we failed to detect significant variations in the rate of escape mutations to the different CRISPR-resistant hosts (see S9 Fig). Variations in the fitness cost associated with these mutations could, however, explain the observed variations in the final frequency of escape mutations (see S5 Fig).
Finally, we experimentally explored the effect of resistance allele diversity on evolutionary emergence for a fixed proportion of host resistance (fR = 0.5). To this end, we generated bacterial populations that were composed of sensitive bacteria and an equal mix of one, two, four, or eight CRISPR-resistant clones. In this case, as expected, an inoculum size of 300 phages always led to pathogen emergence, but increasing host diversity had a strong negative effect on the ability of the phage to evolve to escape host resistance (Fig 5). We also found higher probabilities of observing multiple escape mutations in the low diversity treatment (Kendall, z = −4.8771, Tau = −0.3259, p = 1.07*10−6), which further supports the prediction that host diversity hampers the evolution of the phage population.
The emergence and re-emergence of pathogens has far-reaching negative impacts on wildlife, agriculture, and public health. Unfortunately, pathogen emergence events are notoriously difficult to predict and we need good biological models to experimentally explore the interplay between epidemiology and evolution taking place at the early stages of an epidemic. Here, we used a combination of diverse theoretical and experimental analyses to examine how the composition of a host population impacts the probability of pathogen emergence and evolution. Our theory is tailored to the biology of CRISPR–phage interactions, and subsequent validation using this experimental system demonstrates the predictive power of this theoretical framework. However, we suggest that this framework may be suitable for predicting pathogen emergence whenever hosts recognize specific pathogen epitopes and resistance can be overcome by epitope mutations. For instance, the specificity of the host–parasite interaction driven by CRISPR immunity (S12 Fig) is akin to the classical gene-for-gene system described in plant pathosystems [21]. However, host immunity may not always be perfect, which will impact both the dynamics and the evolution of the pathogen population [22–24]. To further generalize our findings, we derived the probability of pathogen emergence when immunity is imperfect (see section S1.2 in S1 Text). Note, however, that this should be considered separately from the more complex epidemiological dynamics that occur when the probability of a successful infection depends on the pathogen dose or when the pathogen causes immunosuppression, both of which can cause emergence to become dependent on the pathogen population density [25,26].
Our framework provides several insights on emergence and re-emergence in both the presence and absence of pathogen evolution. For instance, this model captures how the composition and diversity of the host population impacts the emergence of a nonevolving pathogen. In this context, a larger proportion of resistant hosts decreases pathogen emergence, but this effect is weaker in spatially structured populations in which transmission is more likely to occur between the same host types, which allows for pathogen persistence in sensitive subpopulations. This effect is akin to the effect of the spatial distribution of suitable habitats on extinction thresholds [27–30] and consistent with earlier work that shows that host composition and spatial structure impact the growth rate of bacteriophage ϕ6 [31]. In the context of an evolving pathogen, our theory helps to explain the general observation that evolutionary emergence and the spread of escape mutations is maximal for an intermediate proportion of resistant hosts in the population [32]. Specifically, this is because increasing host resistance in the population has two opposite effects: (i) the influx of new mutations decreases because the ancestral pathogen cannot replicate on resistant hosts, and (ii) selection for escape mutations increases.
Second, our model predicts that diversity in host resistance alleles decreases the probability of evolutionary emergence. Even though larger host diversity increases the number of adaptive mutations for the pathogen (i.e., a larger number of targets of selection), each mutation is associated with a smaller fitness advantage (i.e., a smaller increase in the fraction of the host population that can be infected). The theory presented here therefore helps to explain previous empirical data on the impact of host CRISPR diversity on the evolution of escape phages [18]. The link between host biodiversity and infectious diseases has attracted substantial attention recently [33–43]. Several studies support the “dilution effect” hypothesis, which postulates that host diversity limits disease spread [39,40, 43]. For example, host diversity may limit the spread of a pathogen by increasing the fraction of bad-quality hosts in the population [43]. Indeed, increasing the fraction of resistant hosts (but not the diversity of resistance alleles) decreases the basic reproduction ratio of the wild-type pathogen [44,45]. In addition, host diversity per se may also limit disease spread, and several studies have shown the negative effect of host diversity on the deterministic growth rate of the pathogen under specific patterns of host–parasite specificity [35,46,47].
Notwithstanding these important insights, what sets our theoretical model apart is its ability to understand the factors that impact the initial pathogen emergence, rather than the downstream spread of a pathogen once it has already emerged. Studying this requires stochastic models, which are critical to model the probability of rare events, for example, pathogen spillover across species, including at the human-animal interface [48,49,4,50], the emergence of drug resistance [51,52], the evolution of vaccine resistance [53], and the reversion of live vaccines [54–58]. In all these public health issues, understanding pathogen emergence requires models accounting for the stochastic nature of epidemiological and evolutionary dynamics. The present study focuses on the effect of the diversity of host resistance when each resistant host carries a single resistance allele (i.e., a single spacer in CRISPR). Our joint theoretical and experimental approach could be readily extended to evaluate the impact of the accumulation of multiple resistance alleles in a single host genotype rather than mixing multiple genotypes with a single resistance allele in the host population. The impact of such alternative strategies on the durability of resistance and on disease spread is particularly relevant in agriculture [59,60]. Our work provides a theoretical framework to study these different issues, and our experimental model system can be used to evaluate the ability of different control strategies to limit pathogen adaptation and emergence.
We detail the derivation of the probability of pathogen emergence presented in the main text (the main parameters of the model are listed in S1 Table). We are interested in the ultimate fate (extinction or not) of a single pathogen with i escape mutations dropped into a very large host population with a proportion fR of resistant hosts. This resistant population is composed of an equal frequency of n different resistance genotypes. This free infectious particle first has to infect a host to avoid extinction, and the probability of ultimate extinction of this pathogen is
Qi,n=(1-fR)qi,n+fR(inqi,n+n-in)
(5)
where qi,n is the probability of ultimate extinction of the pathogen when it is currently infecting a host.
Next, we focus on the probability qi,n(t) at time t that a pathogen with i mutations in an infected host will ultimately go extinct. In a small interval of time, dt, four different events may take place. First, the pathogen may transmit to a new host without additional escape mutations. Second, after a mutation event, the pathogen may transmit a pathogen with i + 1 escape mutations to a new host. Third, the infected host (and the pathogen in the host) may die. Fourth, nothing may happen during this interval of time dt. Collecting these different terms allows us to write down recursions for the probability qi,n(t), at time t, as a function of the probability qi,n(t + dt) and qi+1,n(t + dt), at time t + dt:
qi,n(t)=Ai,ndtqi,n(t+dt)qi,n(t+dt)︸reproductionwithoutmutation+Bi,ndtqi,n(t+dt)qi+1,n(t+dt)︸reproductionwithmutation+ddt︸death+qi,n(t+dt)(1−Ai,ndt−Bi,ndt−ddt)︸noevent
(6)
with:
Ai,n = bi(1 − ui,n)Fi,n
Bi,n = biui,nFi+1,n
Fi,n = (ϕ + (1 − ϕ)(fR i/n + (1 − fR))).
The above calculation is based on the assumption that the pathogen never reaches a high prevalence and that the composition of the host population remains constant (i.e., Fi,n is assumed to remain constant). In other words, the probabilities qi,n(t) are assumed to be invariant with time. We can thus set qi,n(t) = qi,n(t + dt) to obtain a recursion equation that allows us to derive qi,n from qi+1,n.
The first term of this recursion gives the probability of extinction, qn,n that a pathogen with n escape mutations (a pathogen fully adapted to the novel host population) will go extinct. The heterogeneity of the environment has no impact on a fully adapted pathogen, and its probability of extinction is simply the extinction probability of the birth–death process:
qn,n=1/(R0(1-c)n)
(7)
Next, to derive qn−1,n from qn,n, we need the recursion equation for qi,n. However, we have to distinguish two different scenarios. First, if Ai,n = 0, for example, the case of a cell infected by a fully maladapted pathogen (i.e., i = 0) in a well-mixed population with no susceptible hosts (i.e., ϕ = 0, fR = 1), we find:
qi,n=dd+Bi,n(1-qi+1,n)
(8)
Second, in the more general scenario, in which Ai,n > 0, we have
qi,n=Ci,n--4dAi,n+Ci,n22Ai,n
(9)
with: Ci,n = Ai,n + Bi,n(1 − qi+1,n) + d.
Knowing qn,n and the above recursion equations, we can derive qn−1,n and next qn−2,n… until we get q0,n. We are particularly interested in q0,n and Q0,n because these quantities measure the probability of extinction of a pathogen with no escape mutations (in an infected host or as an infectious particle, respectively). Ultimately, we obtain the probability of emergence of an inoculum of V0 propagules of pathogen with no escape mutations (when n = 1, this yields Eq 2 in the main text):
P0,n=1-(Q0,n)V0
(10)
We show in Fig 2 how the diversity of host resistance affects the probability of pathogen emergence through a reduction of evolutionary emergence. In S4 Fig, we illustrate the interaction between host diversity and spatial structure in pathogen emergence. We show that more spatial structure decreases the impact of host diversity on evolutionary emergence and increases the overall probability of pathogen emergence.
To study the impact of the host population composition on the probability of evolutionary emergence, we used two different microbial systems: (i) the gram-negative P. aeruginosa and its lytic phage DMS3vir, and (ii) the gram-positive S. thermophilus and its lytic phage 2972. All the resistant bacteria (i.e., BIMs) derived from the phage-sensitive wild-type strains P. aeruginosa UCBPP PA14 and S. thermophilus DGCC7710 rely on CRISPR-Cas immunity for complete resistance against the corresponding phage [14,61].
For all treatments, we performed 96 replicate infections of the corresponding host populations. We manipulated the composition of the host populations by mixing overnight cultures of sensitive bacteria and BIMs in the proportions indicated in the text, figures, and figure legends. Each replicate population was inoculated 1:100 into fresh growth media and infected with a quantity V0 of phages (the inoculum size), as indicated in the text, figures, and figure legends. After 23 hours, we monitored within each population (i) the occurrence of phage epidemics (i.e., an emergence) and (ii) the presence of escape mutants (i.e., an evolutionary emergence). A detailed description of these experiments is provided in section S2 of S1 Text.
|
10.1371/journal.pgen.1000899 | Chromosome 9p21 SNPs Associated with Multiple Disease Phenotypes Correlate with ANRIL Expression | Single nucleotide polymorphisms (SNPs) on chromosome 9p21 are associated with coronary artery disease, diabetes, and multiple cancers. Risk SNPs are mainly non-coding, suggesting that they influence expression and may act in cis. We examined the association between 56 SNPs in this region and peripheral blood expression of the three nearest genes CDKN2A, CDKN2B, and ANRIL using total and allelic expression in two populations of healthy volunteers: 177 British Caucasians and 310 mixed-ancestry South Africans. Total expression of the three genes was correlated (P<0.05), suggesting that they are co-regulated. SNP associations mapped by allelic and total expression were similar (r = 0.97, P = 4.8×10−99), but the power to detect effects was greater for allelic expression. The proportion of expression variance attributable to cis-acting effects was 8% for CDKN2A, 5% for CDKN2B, and 20% for ANRIL. SNP associations were similar in the two populations (r = 0.94, P = 10−72). Multiple SNPs were independently associated with expression of each gene (P<0.05 after correction for multiple testing), suggesting that several sites may modulate disease susceptibility. Individual SNPs correlated with changes in expression up to 1.4-fold for CDKN2A, 1.3-fold for CDKN2B, and 2-fold for ANRIL. Risk SNPs for coronary disease, stroke, diabetes, melanoma, and glioma were all associated with allelic expression of ANRIL (all P<0.05 after correction for multiple testing), while association with the other two genes was only detectable for some risk SNPs. SNPs had an inverse effect on ANRIL and CDKN2B expression, supporting a role of antisense transcription in CDKN2B regulation. Our study suggests that modulation of ANRIL expression mediates susceptibility to several important human diseases.
| Genetic variants on chromosome 9p21 have been associated with several important diseases including coronary artery disease, diabetes, and multiple cancers. Most of the risk variants in this region do not alter any protein sequence and are therefore likely to act by influencing the expression of nearby genes. We investigated whether chromosome 9p21 variants are correlated with expression of the three nearest genes (CDKN2A, CDKN2B, and ANRIL) which might mediate the association with disease. Using two different techniques to study effects on expression in blood from two separate populations of healthy volunteers, we show that variants associated with disease are all correlated with ANRIL expression, but associations with the other two genes are weaker and less consistent. Multiple genetic variants are independently associated with expression of all three genes. Although total expression levels of CDKN2A, CDKN2B, and ANRIL are positively correlated, individual genetic variants influence ANRIL and CDKN2B expression in opposite directions, suggesting a possible role of ANRIL in CDKN2B regulation. Our study suggests that modulation of ANRIL expression mediates susceptibility to several important human diseases.
| The chromosome 9p21.3 region adjacent to the loci encoding the cyclin-dependent kinase inhibitors CDKN2A (ENSG00000147889) and CDKN2B (ENSG00000147883) is an important susceptibility locus for several diseases with a complex genetic background. Recent genome-wide association (GWA) studies have shown that single nucleotide polymorphisms (SNPs) in this region are associated with coronary artery disease (CAD) [1]–[4], ischaemic stroke [5], [6], aortic aneurysm [7], type II diabetes [8],[9], glioma [10], [11], and malignant melanoma [12]. Candidate gene approaches have also reported SNPs in this region to be associated with breast [13], [14], ovarian [15], and pancreatic carcinoma [16], melanoma [17], and acute lymphoblastic leukaemia [18], as well as with poor physical function in the elderly [19]. Variants associated with these diseases are represented in Figure 1. Most of the risk variants in the chromosome 9p21 region identified by GWA studies are in non-coding regions, suggesting that their effects are likely to be mediated by influences on gene expression. Sequence variation can influence expression by cis or trans mechanisms. Trans-acting elements influence transcript levels of both alleles via diffusible factors and are usually located distant to the genes they regulate, whereas cis-acting elements act on genes on the same chromosome in an allele-specific manner and are usually located close to the genes they regulate. Since most reported risk variants in the 9p21 region do not appear in mature transcripts, and there are no known or predicted microRNAs mapping to this region [20]–[23], these variants are unlikely to produce diffusible trans-acting factors and are therefore likely to influence expression of nearby genes in cis. Genes in the region include the cyclin-dependent kinase inhibitors CDKN2A (p16INK4a) including its alternative reading frame (ARF) transcript variant (p19ARF), CDKN2B (p15INK4b), and a recently-discovered non-coding RNA, designated ANRIL (CDKN2BAS, ENSG00000240498), that undergoes splicing and is transcribed from the opposite strand to CDKN2A/B. The ARF/CDKN2A/B proteins are established tumour suppressors deleted in a range of cancers including familial cutaneous malignant melanoma [24]; they block cell cycle progression and influence key physiological processes such as replicative senescence, apoptosis, and stem-cell self-renewal [25]. Cis-acting regulatory elements for these genes have been identified in vitro using reporter assays [26]–[30], but expression levels are also influenced by factors such as age, chemotherapeutic agents, DNA damage by ultraviolet or ionizing radiation, and levels of transcriptional regulators [31], all of which are likely to act in trans. The function of ANRIL is unknown, but other processed non-coding RNAs are involved in the regulation of gene expression through transcriptional and translational control mechanisms [32].
Genetic effects on expression can be assessed by comparing total expression levels in individuals with different genotypes at a putative regulatory locus. This is termed expression quantitative trait locus (eQTL) mapping [33]. This approach utilises information from all members of the population, but reflects the net effect of both cis and trans-acting influences; the sensitivity to detect cis-acting effects is therefore reduced in the presence of significant variation in trans-acting influences such as the environmental factors outlined above. An alternative approach that is specific for mapping cis-acting influences is to measure allelic expression (aeQTL mapping). An unequal amount of transcript arising from each allele in an individual heterozygous for a transcribed polymorphism indicates the presence of cis-acting effects on expression. While traditional eQTL analysis assesses the influence of polymorphisms by comparing expression between samples, allelic expression analysis compares the expression levels of alleles within individual samples, making it much more robust to trans-acting influences that affect both alleles such as age, gender, population stratification, or experimental variability. This maximises the sensitivity for detecting cis-acting effects [34].
Variants associated with CAD span a region greater than 100kb, but the association is accounted for by SNPs within a 53kb interval that define a core risk haplotype [35]. Lead SNPs for CAD and diabetes are in separate LD blocks in Caucasians and are independently associated with the two separate diseases [35]. To date, CAD risk SNPs have shown inconsistent association with CDKN2A, CDKN2B and ANRIL by eQTL mapping. One CAD risk SNP was associated with altered ANRIL expression in blood, but not with CDKN2A or CDKN2B expression [36], whilst a different CAD risk SNP has been associated with reduced expression of all three genes in peripheral blood T-cells [37]. However, the latter study found no association with expression for other CAD risk SNPs [37], and another report also found no association of a lead CAD risk SNP with these genes or with global gene expression in primary vascular tissue and lymphoblastoid cells [38]. Based on evolutionary conservation and effects on expression, individual SNPs (rs10757278 and rs1333045) have been highlighted as potential causal variants for the association with CAD [36], [37]. However, if multiple cis-acting effects are present at a locus, resolving a disease association by fine-mapping may not be possible. Examining gene expression rather than disease phenotype increases the power to map cis-acting effects, and we used this approach to determine whether multiple sites independently influence expression. Caucasian populations have strong linkage disequilibrium (LD) in the chromosome 9p21 region which limits the ability to separate the effects of individual SNPs on expression [35]. Populations of African ancestry have less LD [39], [40], which can be exploited to improve the fine-mapping of functional polymorphisms associated with quantitative traits [41], [42].
We therefore used eQTL and aeQTL mapping to perform detailed fine-mapping of the association of SNPs at the 9p21.3 locus with expression of CDKN2A, CDKN2B and ANRIL using a mixed-ancestry South African (SA) population, as well as a British Caucasian cohort. We identified multiple SNPs independently associated with expression of each gene, suggesting that several sites may modulate disease susceptibility. The markers identified in GWA studies were all associated with allelic expression of ANRIL, but association with the other two genes was only detectable for some of them. Our study suggests that modulation of ANRIL expression mediates susceptibility to a range of important human diseases.
We measured expression of CDKN2A, CDKN2B and ANRIL in peripheral blood from 310 healthy SA individuals (demographic details provided in the Methods section). Allelic expression was assessed for each gene using two transcribed SNPs located within the same exon. We selected 56 SNPs that tag the common variation in the region, specifically including SNPs with previously reported phenotypic associations. The results of allelic expression mapping in this population were compared with conventional mapping using total expression in the same samples; and with allelic expression mapping in a separate population of 177 healthy British Caucasians. Information on the selected SNPs and genotyping data are summarised in Table S1.
Total expression levels showed substantial inter-individual variation for each of the three genes, up to 13.9-fold for CDKN2A, 36.1-fold for CDKN2B, and 25.5-fold for ANRIL. Allelic expression ratios at individual transcribed markers also showed considerable inter-individual variation, up to 5.6-fold for CDKN2A, 2.4-fold for CDKN2B, and 6.8-fold for ANRIL. Plots of the allelic expression ratios at each transcribed SNP in the SA and Caucasian cohorts are shown in Figure S1 and Figure S2 and plots of the normalised total expression Ct values are shown in Figure S3. Standard errors for ANRIL were higher than for the other two genes in both the allelic and total expression assays, which is likely to be due to the fact that peripheral blood expression of ANRIL was lower than for CDKN2A and CDKN2B.
We estimated the proportion of the variance in total expression that can be attributed to cis-acting effects for each transcribed SNP in the three genes, as described in the Methods section. For CDKNA this proportion was 8% when rs3088440 was used to estimate the variance in cis acting effects, and 4% when rs11515 was used. For CDKN2B the corresponding values were 5% (using rs3217992), 5% (using rs1063192) and for ANRIL 20% (using rs10965215), and 19% (using rs564398).
Total expression levels of CDKN2A, CDKN2B and ANRIL were positively correlated (r = 0.24 to 0.30, all P<4×10−5) as shown in Figure S4, suggesting that expression of these genes is co-regulated.
Allelic expression ratios (AER) measured at the two transcribed SNPs in each gene were highly correlated (CDKN2A r = 0.68 P = 1.7×10−3; CDKN2B r = 0.80 P = 1.7×10−12; ANRIL r = 0.90 P = 1.0×10−26; all genes combined r = 0.96 P = 3×10−61) as shown in Figure S5. This was expected since the two transcribed SNPs selected to assess AER in each gene are located in the same exon and the same transcripts. We therefore used the AERs from both transcribed markers in each gene (as described in the Methods section) for the aeQTL analysis. This increased the number of informative heterozygotes at which allelic expression could be assessed for each gene and increased the power to detect significant effects, as shown in Table 1.
Unlike allelic expression ratios, total expression data may be influenced by covariates that influence expression in trans. We therefore corrected total expression values for covariates (age, sex, and ethnicity) and excluded outlying individuals as described in the Methods section. These corrections did not significantly alter the results of the eQTL analysis, as shown in Figure S6. All subsequent analyses are presented using the covariate-corrected eQTL data. We compared cis-acting effects assessed by eQTL and aeQTL mapping, as shown in Figure 2. There was a strong correlation both for the effect size (r = 0.87, P = 4.7×10−51) and significance of association (r = 0.97, P = 4.8×10−99) at each mapping SNP between the two techniques. However, the associations were more significant for allelic expression than for total expression analysis, indicating that allelic expression had greater power for detecting cis-acting effects. This suggests that trans-acting effects make a substantial contribution to the overall variance of expression in these genes, which is consistent with our estimates that cis-acting effects account for only between 4 and 20% of the overall variance in expression of these genes.
We compared aeQTL analysis between the SA and British Caucasian samples. Results of aeQTL mapping were highly correlated between the two populations, both for the significance of the detected association (r = 0.94, P = 10−72) and the estimated magnitude of the effect on expression for each SNP (r = 0.82, P = 2×10−38), as shown in Figure 3. Patterns of LD in the two populations are shown in Figure S7. Minor allele frequency in the SA population was higher (which increases the proportion of informative heterozygotes for allelic expression analysis) for 33 of the 53 SNPs typed in both populations.
In view of the similarity of the effects in the two cohorts, we combined the data in subsequent analyses, increasing the power to detect cis-acting effects of smaller magnitude and enabling us to adjust for the effects of individual SNPs. The significance of associations for individual SNPs in the combined cohort is shown in Figure 4. Subsequent results refer to the combined dataset, with specific discussion of differences between the populations where relevant.
As described in the Methods section, we defined significance thresholds for all SNP associations using the family wise error rate (FWER) where multiple testing was taken into account by using a Bonferroni correction for the 56 SNPs tested. Associations with a FWER threshold of 0.05 (corresponding to a nominal P-value of 8.9×10−4, −log10P of 3.05, and −log10 FWER of 1.3) were regarded as significant. Table S2 shows the −log10 of the nominal P-values and FWER for all SNP associations, and nominal P-values are reported in the text.
The effect of each SNP on AER is also shown in Table S2. The maximum change in allelic expression associated with any SNP was 1.4-fold for CDKN2A, 1.33-fold for CDKN2B, and 1.97-fold for ANRIL. Due to the power of our combined dataset we were able to detect SNP effects on allelic expression as small as 1.05-fold that were significant.
As shown in Figure 4, multiple SNPs were associated with cis-acting influences on expression of CDKN2A, CDKN2B and ANRIL. This could be the result of multiple independent loci influencing expression of each gene, but could also be a reflection of strong LD in the region since associations might be observed for ‘non-functional’ SNPs (that do not directly influence expression) which are in LD with other ‘functional’ polymorphisms. Adjusting for the effect of individual SNPs was used to assess whether multiple SNPs were independently correlated with expression of the three genes, as shown in Figure 5. For each gene stepwise adjustments were made for the effect of the SNP which showed the most significant association with expression, until independent effects could no longer be detected. Associations remained significant after adjusting for the top SNP for CDKN2A and CDKN2B, and the top two SNPs for ANRIL.
Our results indicate that even after adjusting for the effects of the most significant marker, some of the remaining SNPs still showed significant association with ANRIL expression. This could be explained by the presence of more than one functional polymorphism affecting expression, but could also reflect the presence of a functional polymorphism that is in disequilibrium with both markers. However, examination of the allelic expression patterns provides additional support for the presence of multiple sites affecting expression. For example, Figure 6 shows the allelic expression ratios observed at the transcribed SNP rs564398 in ANRIL, grouped according to the genotype at rs10965215. These two SNPs are in strong LD (D′ = 0.98), hence the absence of individuals homozygous for the A allele at rs10965215 that are heterozygous at rs564398. We observe that the G allele of the transcribed SNP (rs564398) is overexpressed (G/A AER values greater than 1), however overexpression is stronger (P = 10−15 using the Mann-Whitney test) for individuals that are also heterozygous at the second polymorphism (rs10965215). This pattern is not consistent with allelic expression being determined by a single biallelic polymorphism acting in cis and suggests that there is more than one functional polymorphism or that this polymorphism is multiallelic. Such patterns were common in our data.
The direction of cis-acting effects on expression was compared between genes for SNPs showing significant associations with expression of each gene, as shown in Table 2. SNP effects for CDKN2A and ANRIL were in the same direction for all 10 SNPs, meaning that alleles associated with overexpression of CDKN2A were also associated with overexpression of ANRIL. By contrast, for all 8 SNPs that were significantly associated with allelic expression of both CDKN2A and CDKN2B, the alleles associated with CDKN2A overexpression were associated with CDKN2B underexpression. Similarly for all 3 SNPs significantly associated with allelic expression of both CDKN2B and ANRIL, alleles associated with overexpression of CDKN2B were associated with ANRIL underexpression. The total expression analysis had insufficient power for similar analyses to be performed.
We investigated whether SNPs within regulatory regions previously identified by in vitro reporter assays were associated with cis-acting effects on expression in vivo. The effect on gene expression and significance of the association for each SNP is summarised in Table S2.
CDKN2A expression was significantly correlated with SNPs in its promoter and the ARF transcript promoter [26]–[29], and with SNPs close to the RDINK4/ARF domain that has been shown to regulate expression of CDKN2A, ARF and CDKN2B in vitro [30].
CDKN2B expression was also significantly correlated with SNPs in the CDKN2A and ARF promoter regions, suggesting that these elements influence expression of both genes. CDKN2B expression was not significantly correlated with the single SNP typed in its promoter (rs2069418) prior to adjustment, but this became significant after adjustment for the most significant SNP in the ARF promoter (rs3218018).
ANRIL expression was strongly associated with SNPs in the CDKN2B promoter (P = 10−72), ARF promoter (P up to10−53) and RDINK4/ARF domain (P = 10−12), as well as with SNPs adjacent to the CDKN2A promoter (rs3731239, P = 10−25).
These data validate in vivo the function of the regulatory elements identified by in vitro transfection studies, and confirm that shared cis-acting elements influence expression of CDKN2A, CDKN2B and ANRIL.
We examined the correlation of allelic expression of CDKN2A, CDKN2B and ANRIL with SNPs reported to confer disease susceptibility. The effect on gene expression and significance of the association for each SNP is summarised in Table 3.
This is the most detailed study to date of cis-acting influences on expression at the chromosome 9p21 locus. We have shown that multiple sites in the 9p21 region independently influence CDKN2A, CDKN2B and ANRIL expression, and demonstrated that SNPs associated with diseases including CAD, diabetes, and cancers are all highly associated with ANRIL expression, suggesting that modulation of ANRIL expression may mediate disease susceptibility. We also report novel methodology for allelic expression analysis that allowed us to combine data from multiple transcribed polymorphisms and to adjust for the effects of particular SNPs. We have demonstrated that this approach has greater power than total expression analysis for mapping cis-acting effects.
Total expression levels of CDKN2A, CDKN2B and ANRIL, which reflect the combined influence of cis and trans-acting factors, were positively correlated. This corroborates other recent data [37], and suggests that expression of these genes is co-regulated. We have shown that trans-acting influences account for the majority of the observed variance in expression of these genes (80–96%), and the correlation in total expression levels is likely to reflect co-regulation of the genes through trans-acting factors. In addition, our allelic expression analysis demonstrated that expression is also influenced by shared cis-acting elements in the region. Despite the positive correlation in total expression levels, cis-acting effects associated with individual SNP alleles may act in opposite directions; the effect of individual SNPs on CDKN2B expression were opposite to effects on CDKN2A and ANRIL expression (which were concordant) in our study. Because cis-acting effects represent only a small proportion of the overall variance in expression of these genes, the effects acting in trans are likely to account for the positive correlation seen in total expression, but this does not diminish the potential biological significance of the cis-acting effects. ANRIL overlaps and is transcribed in antisense with respect to CDKN2B [49]. It is modestly conserved across species [36] and its function is not known, but recent work has demonstrated that antisense transcription from CDKN2B downregulates CDKN2B expression in cis through heterochromatin formation [50]. This is consistent with our observation of an inverse effect of SNPs on ANRIL and CDKN2B expression. By contrast, CDKN2A and ANRIL showed positive correlations for both allelic and total expression in our study. CDKN2A and ANRIL do not overlap, but are transcribed divergently from transcription start sites separated by just 300 base pairs. Although the ANRIL promoter is currently not characterised, it may share promoter elements with CDKN2A and the resulting co-regulation could account for the positive correlation in expression we observed for these genes, similar to that described at other sites [51]. In this context, inhibition of CDKN2B expression by ANRIL would enable a level of crosstalk between CDKN2A and CDKN2B expression, which would be consistent with the inverse cis-acting effect of SNPs on CDKN2A and CDKN2B that we observed. The observation that cis-acting genetic effects played a greater role in expression of ANRIL compared to CDKN2A and CDKN2B (20% compared to less than 8% and 5% respectively) makes it a good candidate for genetic causation mediated through influences on expression.
We compared total expression and allelic expression for the investigation of cis-acting influences on expression. While traditional eQTL analysis assesses the influences of polymorphisms by comparing expression between samples, allelic expression analysis compares the expression levels of alleles within individual samples, making it more robust to influences that affect both alleles such as age, gender or population stratification. This offers an important advantage for dissecting such cis-acting influences on expression, which although of lesser magnitude than trans-acting influences, may be of biological importance and possibly account for the genetic susceptibility observed in recent GWA studies. For aeQTL mapping we used a novel adaptation of our previously reported methodology [52] to combine multiple transcribed SNPs per gene, which increased the number of informative individuals and the power for detecting cis-acting effects. We demonstrated this approach using two transcribed polymorphisms per gene, but our methodology offers the potential for the inclusion of multiple additional transcribed variants. The results obtained by eQTL and aeQTL mapping were similar, consistent with previous work suggesting that the two approaches identify the same cis-acting loci [42]. However, we demonstrated that aeQTL analysis had substantially greater power than the eQTL approach. Adjusting for trans-acting covariates including age, sex and ethnicity in our eQTL analysis did not substantially alter the results. An influence of age on CDNK2A has been reported [53], but there was little variability in the age of our SA cohort (90% of whom were between the ages of 18 and 30 years). The fact that allelic expression is a more efficient way to identify cis-acting influences on expression has implications for future studies investigating the effects of SNPs on expression at other loci, for example for the hundreds of non-coding SNPs correlated with different diseases by recent GWA studies [54].
Allelic expression quantifies the relative contributions of each allele to the mRNA pool irrespective of the absolute mRNA levels, and therefore provides information about transcriptional effects and polymorphisms within the transcript influencing RNA degradation in cis. By contrast, total expression analyses that quantify absolute mRNA levels are also sensitive to post-transcriptional regulatory effects, such as mRNA degradation by microRNAs. In extreme cases tight post-transcriptional regulation could keep total mRNA levels constant irrespective of the contributions of each allele to the total mRNA pool. The fact that the results of eQTL and aeQTL mapping were so similar in our study suggests that the effect of regulation at the post-transcriptional level is limited, although regulation of CDKN2A expression by a microRNA has been described [55]. In general, although allelic expression is a robust method for mapping sites influencing expression in cis, investigation of total expression and other intermediate phenotypes such as protein levels or protein activity will provide complementary information that contributes to fully understanding the phenotypic effects of cis-acting polymorphisms. It would be desirable to determine whether the significant associations with mRNA expression observed for CDKN2A and CDKN2B are confirmed at the protein level.
Although we had hoped to use trans-ethnic fine-mapping to refine the associations with expression, the results of aeQTL mapping were in fact very similar in the SA and Caucasian populations. This replication in a separate cohort strongly supports the validity of our findings and enabled us to perform a combined analysis of the two cohorts. This approach of pooling data from ethnically-divergent populations has been previously shown to increase the power to detect influences on expression that are shared across populations [42], [56]. The principal difference we identified between the two populations was for the SNPs associated with type II diabetes. The lead diabetes SNP rs10811661 was correlated with ANRIL underexpression in the Caucasian cohort, but not in the SA population, despite greater power to detect effects in that cohort. This may reflect differences in LD between the populations, but suggests that rs10811661 may not itself be the causal variant influencing diabetes susceptibility through effects on ANRIL expression. Studies to determine whether this SNP is associated with diabetes in populations of African origin would be of interest.
The power of our analyses to detect differences in expression enabled us to adjust for the effects of individual SNPs. Using this we were able to demonstrate that expression, and therefore probably disease predisposition, is independently influenced by multiple sites and that the observed effects cannot be explained by a single polymorphic site. From our analysis we cannot exclude the existence of rare variants with large effects, but previous resequencing studies in this region did not find rare variants associated with disease phenotypes [2], [3]. We are unable to say whether the individual SNPs for which we found associations are the actual ‘causal’ variants responsible for the effects on expression, or if the association simply reflects linkage disequilibrium between these SNPs and the causative polymorphisms. Although fine mapping studies often purport to identify causal variants, in the context of complex diseases characterising the pathways involved in disease predisposition may be more important. This is of particular interest for these genes where variation in expression is mostly due to trans effects which may be substantially influenced by non-genetic factors, raising the prospect that it may be amenable to therapeutic modulation. The putative causal variants rs10757278 and rs1333045 previously associated with altered ANRIL expression [36], [37] were significantly associated with reduced ANRIL expression in vivo in our analysis, but their effects were relatively modest compared to other SNPs in the region and adjustment for the effect of these SNPs accounted for only a small proportion of the effect observed at other SNPs. The maximum changes in expression associated with individual SNPs were substantial, up to 2-fold for ANRIL, but we were also able to detect effects of much smaller magnitude; the minimum significant effect was associated with just a 1.05-fold change in expression. Although the associations of SNPs with expression that we observed were statistically highly significant, we cannot say what impact such effects on expression have on disease risk. However, even small differences in gene expression due to genetic factors that are present throughout an individual's lifetime could contribute to differences in common late-onset phenotypes such as CAD and diabetes, and the effects may be even greater in tissues related to disease.
We examined in vivo expression in primary cells rather than in transformed cell lines. Although cell lines have been extensively used to investigate cis-acting influences on expression [56], [57], patterns of expression may be altered in immortalised cells, particularly for genes such as these that are associated with senescence and cell-cycle regulation. Furthermore, widely used cell lines are pauciclonal or monoclonal [58], [59] and since a significant proportion of human genes exhibit random patterns of monoallelic expression within single clones of cell lines [60], cis-acting effects in these cells are unlikely to be representative of polyclonal cell populations in vivo. Previous studies have delineated the promoters and other elements regulating CDKN2A/ARF and CDKN2B expression using reporter assays [26]–[30]. Such studies are valuable to identify causative polymorphisms, but since they examine the effects on expression outside of the normal haplotype, chromatin and cellular context their findings require confirmation by in vivo studies [34], [61]. Our analysis confirmed that polymorphisms in upstream regulatory elements identified by in vitro assays were significantly associated with cis-acting effects on expression in vivo, but we also demonstrated that other loci located up and downstream were associated with effects on expression of similar or even larger magnitude. These data highlight the complexity and multiplicity of sites influencing expression in the region. The assays we used to investigate CDKN2A expression also included the ARF transcript variant. This gave the possibility to detect sites influencing expression of both transcripts, and we were able to detect effects of SNPs in both the CDKN2A and ARF promoter regions, although differential effects of loci on individual transcripts cannot be distinguished using this approach.
All of the SNPs in the region associated with disease in GWA studies were associated with influences on ANRIL expression, suggesting that modulation of ANRIL expression may mediate susceptibility to these phenotypes. SNPs in the CAD core risk haplotype region [35] that are most strongly associated with CAD in GWA studies were associated with reduced ANRIL expression, but other SNPs associated with CAD which lie outside of the core risk haplotype showed independent and stronger associations with ANRIL underexpression. This may reflect differences in the relative importance of particular sites in the tissues responsible for the association with CAD. Indeed, the patterns of association we have observed in peripheral blood in healthy individuals may differ from those in primary disease tissues. Similarly, differences in the relative contribution of each SNP to modulation of expression in the tissues crucial for the pathogenesis of the different conditions could explain why particular diseases are associated with different subsets of SNPs that influence ANRIL expression. Recent work also suggests that ANRIL has multiple transcripts, which may be differentially expressed between tissues [36], [38]. Confirmation of our findings in tissues relevant to each disease and for different ANRIL transcripts would therefore be desirable, although for CAD and other complex diseases the cell populations responsible for mediating disease susceptibility are unknown and may be inaccessible. Although tissue specificity of cis-acting influences is well documented, variation in cis-acting effects is primarily explained by genetic variation, with allele-specific expression at most SNPs being the same between tissues in the same individual [62]. Analysis of expression in blood is therefore likely to give biologically relevant information despite the fact that this may not be the tissue in which influences on expression actually mediate disease susceptibility.
Previous genomewide expression analyses using microarrays and immortalised cell lines did not identify association of CDKN2A and CDKN2B expression with markers in this region, although they did not examine ANRIL expression [56], [57]. However, two recent studies specifically examining expression in the chromosome 9p21 region in primary cells reported associations between CAD risk SNPs and gene expression in blood [36], [37]. Jarinova et al found significant association of CAD risk variant rs1333045-C with ANRIL expression, but not with CDKN2A or CDKN2B expression [36]. Liu et al reported that a different CAD risk allele rs10757278-G was associated with reduced expression levels of CDKN2A, CDKN2B, and ANRIL, but in the same study found no correlation for five other SNPs tested, including two additional SNPs associated with CAD (rs518394 and rs564398). They also found no association for two SNPs associated with diabetes (rs10811661 and rs564398), the frailty risk SNP rs2811712, and a melanoma risk SNP (rs11515) [37]. We demonstrated that CAD risk SNPs rs1333045-C and rs10757278-G both correlate with ANRIL underexpression, but found no correlation of these SNPs with CDKN2A or CDKN2B expression. However, we identified highly significant influences on expression associated with other SNPs for which Liu et al found no association (rs10811661 with CDKN2A and ANRIL; rs564398 with ANRIL; rs2811712 with CDKN2B; rs11515 with CDKN2A and CDKN2B). These findings are likely to reflect the greater power of our analysis for detection of cis-acting effects due to the larger sample size and increased sensitivity of our aeQTL mapping approach.
The finding that disease associated SNPs are all associated with ANRIL expression suggests that ANRIL plays a role in influencing disease susceptibility. Although little is known about the targets of ANRIL, its effects may be mediated through antisense transcription regulation of CDKN2B in the tissues critical for the pathogenesis of the different diseases. The observation that the effects of sequence variants acting in cis were stronger for ANRIL than for CDKN2B may reflect selection pressure against variants that have substantial direct effects on the expression of critical genes. CDKN2A, ARF and CDKN2B are cell cycle regulators and are plausible candidates for involvement in the pathogenesis of the diseases for which we found SNP associations with ANRIL. Mutations involving these genes are well documented in glioma [63], [64] and melanoma [49], [65], [66]. Overexpression of CDKN2A and CDKN2B in murine models is associated with pancreatic islet hypoplasia and diabetes [67], [68], and there is also emerging evidence that vascular cell senescence involving these pathways is involved in the pathogenesis of atherosclerosis [69], [70].
Our data show that multiple independent sites in the chromosome 9p21 region influence CDKN2A, CDKN2B and ANRIL expression. SNPs associated with disease in GWA studies are all associated with ANRIL expression, indicating that modulation of ANRIL expression mediates susceptibility to a variety of conditions.
Peripheral blood for DNA and RNA analysis was collected from anonymous adult volunteers in two cohorts: 310 SA blood donors and 177 British Caucasians from north-east England. The self-reported ethnicity of the SA cohort was: 200 Cape mixed-ancestry; 67 African black; 19 Indian; 10 white; 4 other/unknown. 42% were male, with median age 20 years (range 17–60, lower quartile 19, upper quartile 23). In the Caucasian cohort, 50% were male, with median age 63 years (range 25–101, lower quartile 51, upper quartile 69).
The study complies with the principles of the Declaration of Helsinki. Informed consent was obtained from all participants and the study was approved by the Newcastle and North Tyneside Local Research Ethics Committee and the University of Cape Town Faculty of Health Sciences Research Ethics Committee.
For the South African samples DNA was extracted using a phenol/chloroform method from 4ml of peripheral blood in EDTA collected at the time of the RNA sample. For the British samples, DNA was obtained from the RNA solution prior to DNase treatment.
RNA was extracted from 2.5ml of peripheral blood collected using the PAXgene system (Qiagen) following the manufacturer's protocol and was DNase treated using RQ1 RNase-Free DNase (Promega). For AEI measurements, approximately 2µg of total RNA was reverse transcribed and eluted in 20µl, using SuperScript VILO cDNA Synthesis Kit (Invitrogen) for the SA samples and SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen) for the British samples. For real-time PCR measurements, 500ng of total RNA was reverse transcribed using High Capacity RNA-to-cDNA Master Mix (Applied Biosystems) and eluted in 20µl.
Using the NCBI Entrez Gene database (http://www.ncbi.nlm.nih.gov/, 28/01/08), transcribed SNPs with expected heterozygosity >0.2 in the HapMap CEU population were selected as suitable candidates for assessment of allelic expression. Transcribed polymorphisms in ANRIL, which was not annotated in the databases at the time of the design, were identified by comparing the reported mRNA sequence [49] with NCBI dbSNP. Transcribed SNPs selected using these criteria were: rs3088440 and rs11515 in exon 3 of CDKN2A; rs3217992 and rs1063192 in exon 2 of CDKN2B; rs10965215 and rs564398 in exon 2 of ANRIL. The two CDKN2A SNPs are also present in ARF, allowing the assessment of cis-acting influences on both of these transcripts. Another SNP rs10738605 in exon 3 of ANRIL also satisfied these criteria but was subsequently excluded due to poor performance of the assay.
SNPs previously reported to be associated with disease phenotypes were selected for mapping effects on expression [6], [8], [9], [13]–[19], [43], [45]–[48], [71]–[74]. Additional tag SNPs required to capture common variation in a core region of interest (Chr9:21958155–22115505) based on HapMap CEU data were also selected using HaploView 4.0 Tagger software using the following parameters: minimum minor allele frequency 0.01, pairwise tagging, r2 threshold >0.8. SNPs within other functionally important elements such as CDKN2A and CDKN2B promoters [26], [29], [30], [75] or a putative ANRIL promoter region (which we arbitrarily defined as 1kb up and downstream of the transcription start site) were selected if they were reported more than once in NCBI dbSNP, had expected heterozygosity >5%, and were associated with alteration of transcription factor binding sites (using PROMO v.3.0.2) [76], [77]. Details of included SNPs are shown in Table S1.
Multiplex SNP genotyping was performed by primer extension and MALDI-TOF mass spectrometry using iPLEX Gold technology from Sequenom (Sequenom Inc, San Diego, USA). SNP assays were designed using Sequenom's RealSNP (www.RealSNP.com) and MassARRAY Assay Design v3.0 Software (multiplex details and primer sequences available in Table S4). PCR was performed using 20ng of DNA in a 10µl reaction volume for 35 cycles using standard iPLEX methodology. Spectra were analysed using MassARRAY Typer v3.4 Software (Sequenom). Spectra and plots were manually reviewed and auto-calls were adjusted if required. Positive and negative controls were included. Individual samples with low genotype call rates (<80%) and SNP assays with poor quality spectra/cluster plots were excluded. Correspondence to Hardy-Weinberg proportions was checked for each SNP.
PCR primers for the selected transcribed SNPs were designed using Primer3 (v.0.4.0) software [78]. CDKN2A primers span exons 3–4 and include both transcribed SNPs (rs3088440 and rs11515) in the same amplicon. ANRIL primers span exons 1–2 and include both transcribed SNPs (rs10965215 and rs564398) in the same amplicon. For CDKN2B, separate primer pairs for transcribed SNPs rs1063192 and rs3217992 were designed entirely within exon 2 (due to the distance of transcribed SNPs from the exon boundary).
Quantification of the allelic expression ratio was performed by primer extension and MALDI-TOF mass spectrometry using iPLEX Gold with similar parameters to the genotyping assay. Spectra were analysed using MassARRAY Typer v3.4 Software (Sequenom) and allelic ratios were estimated as the ratios of the area under the peak representing allele 1 to that representing allele 2. Measurements were performed in four replicates using 50ng of cDNA template. Results from amplification of genomic DNA were used as an equimolar reference to normalise the cDNA values. Genomic normalisation reactions for CDKN2B used the same PCR primers as used for cDNA, but for CDKN2A and ANRIL (where primers were cDNA-specific) separate assays designed to be as close as possible in size and location to the cDNA primers were used. Primer sequences are shown in Table S3. For some assays the allelic ratios measured in gDNA ratios did deviate from a 1∶1 ratio, as shown in Table S4, confirming that allelic ratios in cDNA required correction for assay bias. However, as expected the gDNA ratios for each assay were relatively homogeneous with little inter-individual variability compared to cDNA ratios (Figure S1 and Figure S2). We compared the results of expression mapping using two different normalisation strategies in the SA cohort: normalising to a mean population normalisation factor versus normalising each individual's cDNA to their own gDNA ratio. There was no difference in the results obtained using these two normalisation strategies, as shown in Figure S8. The mean gDNA ratios for each assay were the same in the SA cohort and a sample of Caucasian individuals (no significant difference using a two sample t-test), and we therefore used the mean gDNA ratios for normalisation of all samples.
The appropriateness of genomic normalisation ratios and linearity of the AER response were checked by mixing PCR products from individuals homozygous for the minor and major alleles in varying ratios (8∶1, 4∶1, 1∶1, 1∶4, 1∶8) and using these as template for the allelic expression assays. These experiments confirmed that allelic expression showed a linear response and that normalisation ratios obtained using allelic expression assays on a 1∶1 mixture of alleles for each SNP correspond to normalisation ratios obtained from genomic DNA (Table S4 and Figure S9).
Allelic expression ratios for the two transcribed markers in each gene were highly correlated (CDKN2A r = 0.68, p = 1.7×10−3; CDKN2B r = 0.80, p = 1.7×10−12; ANRIL r = 0.90, p = 1.0×10−26; all genes combined r = 0.96, p = 3×10−61) as shown in Figure S5; we therefore used a novel approach of combining allelic ratios from the two transcribed markers in each gene to increase the number of informative heterozygotes.
Real-time PCR reactions were performed using TaqMan gene expression gene expression probes and reagents (Applied Biosystems) and run on a 7900HT Real-Time PCR System (Applied Biosystems). Commercially available FAM-labelled TaqMan assays were used for CDKN2A exons 2–3 (Hs00923894_m1) and ANRIL exons 1–2 (Hs01390879_m1). A custom FAM-labelled assay was used for exon 2 of CDKN2B. Commercially available VIC-labelled TaqMan assays were used for three reference genes shown to be suitable for normalisation of expression in peripheral blood [79], [80]: B2M (4326319E), GAPD (4326317E), and HPRT1 (4326321E). TaqMan assays are validated by the manufacturer to have close to 100% amplification efficiency and assays were selected to quantify the same transcripts as the allelic expression assays. PCR was performed according to the manufacturer's protocol using four replicates, 25ng cDNA template per reaction, and the following multiplex combinations: CDKN2A/B2M, CDKN2B/GAPD, and ANRIL/HPRT1.
Relative total expression was analysed using the comparative cycle threshold (Ct) method. Ct values for each target gene were normalised to the mean Ct value of the three reference genes [79]. Standard errors and variances of measurements for allelic and total expression analyses in the SA population are shown in Table S5.
The association between total expression, as measured by real time PCR, and each of the SNPs was assessed using linear regression of the log transformed normalized expression values on the genotype assuming no dominance or interactions between the effects of different SNPs. The effect of including age, sex, and ethnicity as covariates, as well as excluding outlying individuals as determined by visual inspection (highlighted in Figure S4) were investigated. Self reported ethnicity was included as a categorical variable (categorised as “Cape mixed-ancestry”, “black African”, “white”, “Indian”, and “other”). These corrections made no significant difference to the results of eQTL mapping (Figure S6). All analyses were performed using the corrected data. Plots illustrating the associations between genotype and total expression for selected SNPs are shown in Figure S10.
We analysed allelic expression ratios using an extension of the approach we published previously [52]. We restrict ourselves to biallelic markers, and code one arbitrarily chosen allele as 0 and the other as 1. We designate with g the phase-known and with T the phase-unknown genotype of an individual. The latter can be ascertained through genotyping. We assume that the amount of mRNA originating from a single allele follows a lognormal distribution where the variance does not vary between different alleles. The log of the ratio between the expression levels of both alleles, I, can therefore be assumed to be normally distributed.
For an individual that is heterozygous for m transcribed polymorphisms, m ratios can be determined. We designate the vector of the logarithms of these ratios as . Under the assumptions above, the components of I are normally distributed with where the means depend on the genotype but the variance is genotype independent but may depend on the site used to measure the allelic expression ratio. We model the expected value as a linear combination of the influences of the typed polymorphisms:where represents the effect of the ith cis acting markers; and characterizes the phase between transcribed and putative cis acting markers:In order to assess the association between a specific SNP and allelic expression, let us consider a set of individuals. For an individual () we can measure the unphased genotype and a vector representing the log of the allelic expression ratios . Up to a multiplicative constant the likelihood of observing a certain pattern of imbalance in this set of individuals given their genotyping results is:withwhere designates the probability of the phased genotypes g given the genotyping results and describes the density of the distribution of given the genotype g. was estimated using the hap procedure from the R-package gap (as deposited in the CRAN archive http://cran.r-project.org/) to phase the genotypes of the two populations separately.
We assume that the allelic expression ratios measured at different sites are conditionally independent given the genotype. Therefore:where and denotes the density of a normal distribution with the individual expression ratio as variate, a genotype dependent mean and a variance . Therefore depends on (i = 1,…,n) and (k = 1,…,m), and maximisation of this likelihood allows assessment of the effects of single or groups of SNPs and to adjust for the effects of other markers by comparing nested models using likelihood ratio tests.
For both total and allelic expression multiple testing was taken into account by calculating the family wise error rate using a Bonferroni correction for the 56 SNPs tested. Associations with family wise error rate below a threshold of 0.05 (corresponding to a nominal P-value of 8.9×10−4, −log10P of 3.05, and −log10FWER of 1.3) were called significant.
From our allelic and total expression data we also estimated the proportion of total expression variance that is due to cis-acting effects. This assumes that cis and trans-acting factors act in an additive manner, do not interact, are independent and that there is random mating, no segregation distortion, and the locus is not subject to imprinting. Given these assumptions, we estimate the variance due to cis acting effects, , as , where is the allelic expression ratio for individual i, and the proportion of the total variance due to cis acting effects can be estimated as , where is the estimated total variance, i.e. with and represent the total expression level for individual i as determined by real time PCR.
|
10.1371/journal.pcbi.1002764 | Mesoscopic Model of Actin-Based Propulsion | Two theoretical models dominate current understanding of actin-based propulsion: microscopic polymerization ratchet model predicts that growing and writhing actin filaments generate forces and movements, while macroscopic elastic propulsion model suggests that deformation and stress of growing actin gel are responsible for the propulsion. We examine both experimentally and computationally the 2D movement of ellipsoidal beads propelled by actin tails and show that neither of the two models can explain the observed bistability of the orientation of the beads. To explain the data, we develop a 2D hybrid mesoscopic model by reconciling these two models such that individual actin filaments undergoing nucleation, elongation, attachment, detachment and capping are embedded into the boundary of a node-spring viscoelastic network representing the macroscopic actin gel. Stochastic simulations of this ‘in silico’ actin network show that the combined effects of the macroscopic elastic deformation and microscopic ratchets can explain the observed bistable orientation of the actin-propelled ellipsoidal beads. To test the theory further, we analyze observed distribution of the curvatures of the trajectories and show that the hybrid model's predictions fit the data. Finally, we demonstrate that the model can explain both concave-up and concave-down force-velocity relations for growing actin networks depending on the characteristic time scale and network recoil. To summarize, we propose that both microscopic polymerization ratchets and macroscopic stresses of the deformable actin network are responsible for the force and movement generation.
| There are two major ideas about how actin networks generate force against an obstacle: one is that the force comes directly from the elongation and bending of individual actin filaments against the surface of the obstacle; the other is that a growing actin gel can build up stress around the obstacle to squeeze it forward. Neither of the two models can explain why actin-propelled ellipsoidal beads move with equal bias toward long- and short-axes. We propose a hybrid model by combining those two ideas so that individual actin filaments are embedded into the boundary of a deformable actin gel. Simulations of this model show that the combined effects of pushing from individual filaments and squeezing from the actin network explain the observed bi-orientation of ellipsoidal beads as well as the curvature of trajectories of spherical beads and the force-velocity relation of actin networks.
| Cell migration is a fundamental phenomenon underlying wound healing and morphogenesis [1]. The first step of migration is protrusion – actin-based extension of the cell's leading edge [2]. Lamellipodial motility [3] and intracellular motility of the bacterium Listeria monocytogenes [4] are two prominent model systems that in the past decades have added considerably to our understanding of the protrusion based on growth of actin networks. These in vivo systems are complemented by in vitro assays using plastic beads [5] and lipid vesicles [6] that, when coated with actin accessory proteins, move much the same way as the Listeria pathogen.
Here we examine computationally the mechanics of growing actin networks. This problem has a long history starting from applying thermodynamics to understand the origin of a single filament's polymerization force [7]. The notion of polymerization ratchet led to the derivation of an exponential force-velocity relation (Figure S1 in Text S1) for a rigid filament growing against a diffusing obstacle [8]. Then, elastic polymerization ratchet model [9] was proposed for flexible actin filaments. This model evolved into tethered ratchet theory, in which a dynamic balance between surface-pushing growing filaments and motion-resisting attached filaments (Figure 1A) governs the protrusion [10]. These early theories considered independent single filaments. However, actin filaments do not grow individually, but evolve interdependently as a network by branching sideways from each other [11]. Mathematical treatments and computer simulations of branching and nucleation [12], [13] of filaments growing against an opposing force, which treated the dendritic actin network as a mechanically rigid body, predicted various force-velocity relations. Those ranged from concave-down (velocity of protrusion being insensitive to the load up to a threshold and plunging to a stall at a critical opposing force) to concave-up (more or less exponential decrease of the velocity with the growing load) relations (see Figure S1 in Text S1). These theoretical efforts culminated in detailed agent-based three-dimensional (3D) models of growing networks of rigid filaments propelling Listeria pathogen [14], [15].
In parallel to these microscopic theories, macroscopic elastic propulsion model [16], [17] suggested that the curved surface of the pathogen is not merely pushed, but squeezed forward by an elastic stress. This stress is developed from the stretching of the outer layer of actin gel by the growth of the gel near the inner surface (Figure 1B). This model treated the actin network as an isotropic elastic continuum and did not explicitly consider the microscopic mechanism of force generation at the surface. As a result, a concave-up force-velocity relation for the actin-propelled spherical bead was derived [18], predicting an initial rapid decay with opposing force followed by a region of slower decay of velocity. This prediction was confirmed by using a cantilever setup for beads coated with the actin polymerization activator N-WASP and moving in a pure-protein medium [18]. On the other hand, when the force-velocity relation of an actin network growing against a flat surface was measured using the cantilever method, it was found that the growth velocity was constant at small forces but dropped rapidly at higher forces [19] as predicted by some microscopic ratchet theories.
Note that the widely used terminology could be confusing as the elastic propulsion theory is sometimes called mesoscopic rather than macroscopic. Both terms are justified: the macroscopic mechanics is described using continuum theory, but an actin layer of a few microns thin is certainly a mesoscopic system. The model we present is mesoscopic in the sense that it spans from the microscopic level of individual filaments to the macroscopic level of continuous description of an actin gel. The model is also hybrid because it takes into account both local discrete forces and global network stress. We will mostly use the term “hybrid” throughout the paper.
The first simple attempt to use hybrid modeling of the lamellipodial edge was recently made in [20], where the actin network was divided into a semiflexible region near the membrane and a gel-like region at the back. Near the membrane, semiflexible filaments are assumed to produce entropic forces against both the membrane and the gel. In the back, the viscous gel deforms in response to stresses both from frontal filaments and internal contractions, causing retrograde flow. Because the semiflexible region is assumed to be supported by the gel region, the moving speed of the membrane is determined by the coupling between the two regions. This model was able to reproduce both concave-up and concave-down shapes of the force-velocity relation. Since this model considered only a one-dimensional strip of actin gel, it did not address the effects of surface geometry.
Besides the force-velocity relation, the non-zero curvatures of the trajectories of motile objects [21] is another important observable. A pioneering microscopic ratchet-based model, which investigates how randomly distributed actin filaments propel a cigar-shaped pathogen, predicted that the resultant bacterial trajectories have curvature values following a Gaussian distribution with zero mean [22]. This conclusion was challenged by a number of studies. One of them showed helical movements that were explained as a result of a non-vanishing torque that arises from a persistent actin-induced off-center force [23]. Another study did not result in helical paths of beads, but rather showed a highly varying curvature of trajectories which has a Gaussian distribution, albeit with a sharp peak at zero curvature [24]. In contrast, a third study indicated that the distribution of the curvatures of trajectories deviated significantly from Gaussian, which was explained by a cooperative breaking of filaments tethered to the bead [25]. All theories used to explain these experiments were microscopic; elastic propulsion model was never applied to these phenomena.
Below, we describe observations of ellipsoidal, rather than spherical, beads that cannot be explained by either microscopic or macroscopic model. This, as well as the complex force-velocity relation and curvature distribution described above, hints that perhaps a hybrid model with individual actin filaments pushing from the surface of a macroscopic deformable actin gel can explain the experiments better. Recent experiments and theory [26], [27] demonstrated that disassembly and breaking of the actin gel are as important as the elastic deformations in generating propulsion. Therefore, we developed a model of a node-spring viscoelastic network representing the actin gel with individual pushing and pulling filaments embedded into the network boundary. Simulations of this in silico hybrid network showed that the combined effects of the macroscopic viscoelastic deformation and microscopic ratchets can explain both concave-up and concave-down force-velocity relations for growing actin networks, bistable orientation of the actin-propelled ellipsoidal beads, and peculiar curvature distributions for the actin-propelled trajectories of the beads.
We developed a two-dimensional (2D) simplification of a 3D hybrid model (Figure 1C), which incorporates both arrays of dynamic actin filaments at the surface-tail interface and the bulk deformable actin gel behind the interface. Filament arrays are embedded into the boundary of the deformable actin gel, which is coarse-grained into a network of nodes interconnected by elastic springs. Individual filament arrays at the surface-tail interface switch between pushing the obstacle surface and attaching to it. The existing filaments are constantly becoming a part of the network and dynamically expanding the actin gel, while nascent filament arrays are created around the surface via a mixture of nucleation and branching processes. The actin network undergoes disassembly, which is treated by removing the nodes and springs at a constant rate, as well as by rupturing crosslinks at a critical stretching force. The deformations of the network as well as the elastic filament forces cause both translational and rotational motion of the bead. The model reproduces the steady motion of beads propelled by treadmilling actin tails behind the beads (Video S1). Further details about the model assumptions, equations, numerical simulations and model parameters are described in the Materials and Methods and Text S1.
Recently, with our experimental collaborators, we reported observations of the ellipsoidal beads that were uniformly coated with an actin assembly-inducing protein (ActA) [28] and moved in the plane between two parallel coverslips (see the Materials and Methods below). Surprisingly, roughly half of the time the beads moved along their long axes, and another half – along their short axes (Figure 2, A and B), with infrequent switches between these orientations.
To see whether the two existing models of actin propulsion can explain this result, we simulated the motion of actin-propelled ellipsoidal beads as described in the Materials and Methods. Elastic theory predicts that squeezing of an ellipsoidal bead introduces a torque orienting the bead with its long axis parallel to the actin tail (see Figure S2 and Figure S6 in Text S1). In agreement with this prediction, when we decreased the autocatalytic branching of actin and attachment forces, so that the actin gel exerted almost uniform normal stress on the bead surface, the model resulted in a propulsion along the bead's long axis (Video S2). On the other hand, when we simulated a network of rigid branching filaments pushing the bead, the propulsion was always along the short axis, so the bead moved sideways (Video S3). This change in the preferred orientation is caused by a subtle bias in how the actin network spreads along the bead surface: if the bead's orientation is skewed relative to the actin tail's axis, filament branching are more likely to happen near the tail-facing flatter surface where there is a higher number of existing filaments. As a result, more filaments push the bead sideways from the actin tail, shifting the filament-contacting region from the curved surface to the flatter one. Eventually, most filaments branch against the flatter part of the surface, orienting the bead with its long axis normal to the tail axis (see Figure S7 and detailed calculations in Text S1).
Thus, the elastic propulsion model predicts that beads only move along their long axes, while microscopic ratchet model predicts that beads only move along their short axes, and neither model can explain the observation. In contrast, the full hybrid model predicts that the bead can move in both orientations due to the combination of the elastic squeezing and the geometric spreading of actin and switch infrequently between them (Video S1, Figure 2, C and D, Figures S5 and Figure S8 in Text S1), in agreement with the observation (Figure 2, A and B). For more insight into this phenomenon and to generate predictions for experiment, we investigated numerically how the fraction of beads moving with a certain orientation depends on the geometric, mechanical and kinetic parameters.
To further test the hybrid model, we simulated the motion of actin-propelled spherical beads (Figure 3, A and C). We recorded the 2D ‘in silico’ trajectories of the beads and compared them to the experimental observations (see the Materials and Methods). We examined two possible mechanisms for the nucleation of new filaments: autocatalytic branching and spontaneous nucleation. We found that each mechanism alone does not produce the observed motion of the bead (see Video S4 and Video S5). Only a combination of the two mechanisms leads to realistic motion of the bead (see Video S6 and details in Text S1). Note that the trajectories are easy to visualize by looking at the actin tails that represent the most recent parts of the trajectories, see Figure 3, B and D). Our typical simulation results (Figure 3, A and E, Video S7) illustrate that in general the trajectories are mildly curved, as observed in some cases experimentally (Figure 3B). However, in other cases the experimental observations (Figure 3D) show that once in a while the beads stop, get surrounded by a dense actin ‘cloud’, and then break through the cloud and resume movement in a new direction.
Indeed, the model predicts that when the detachment rate of actin filaments becomes low and a greater fraction of filaments is attached to the bead surface, beads start to have pulsatory motion due to temporary entrapment by the actin gel (Figure 3C and Video S8), which occurs frequently in this regime. The explanation is that when filaments detach rapidly and thus do not generate great pulling forces, beads move quickly and can hardly be trapped, but at low detachment rate, beads slow down significantly by the strong pulling forces, which increases their chances to be trapped into the actin gel. Both our simulations and observations from our collaborators show that beads often make sharp turns during their escapement from the surrounding actin gel (Figure 3, C and D), causing the switching between the low- and high-curvature trajectories. As a result, the trajectories show spatially separated segments of low and high curvatures (Figure 3F).
To obtain the distribution of the curvatures of the trajectories, we smoothed the simulated bead's trajectory to remove the high frequency noises and calculated (see Text S1 for details) that the curvature distribution is close to Gaussian (Figure 4A) for fast-moving beads in the wide range of parameters. This indicates that the turning of the fast-moving bead is likely to be driven by random events in the protruding actin network.
When the detachment rate is low, we find that the curvature distribution becomes sharply peaked at zero (Figure 4B), in agreement with both our observation (Figure 4B) and previous results [24]. Since the low- and high-curvature trajectories are typically separated in this regime, this sharp peak near zero is due to bead moving in a rapid-and-smooth fashion, while the slowly decreasing distribution at higher curvatures is caused by bead moving in a slow-and-jagged fashion. Furthermore, we find that the distribution is close to a Gaussian at higher curvature, indicating that the highly curved segments of trajectories are also likely to be caused by the random fluctuations in the actin network.
We found that the predicted characteristic value of the root-mean-square curvature, (Figure 4C), is of the same order of magnitude as our observations (Figure S17 in Text S1) and available measurements [4], [24], [25]. We investigated how the filament attachments affect the value of (Figure 4C) and found that is insensitive to for . However, the curvature increases rapidly with for , consistent with the idea that excessive attached filaments cause frequent trapping of the bead leading to highly curved trajectories.
We also studied how the bead radius, , affects (Figure 4D) and found that decreases as the bead size increases. This result is in agreement with the experimental observations reported in [4], [25]. Interestingly, this results is also consistent with our experimental observation on the orientation-dependent turning of the trajectories of ellipsoidal beads (Figure S17 in Text S1): ellipsoidal beads moving along their long-axes are less likely to keep their current direction of motion comparing to those moving along their short-axes. A possible interpretation is that the former are mostly pushed at their sharp ends where the radius of curvature is low. Similar to a spherical bead with small , this will lead to a high in the trajectory and thus will be less likely for the bead to keep the current direction of motion. Together, the above results can be explained as follows: larger beads are propelled by a greater number of filaments, so relative fluctuations in the actin network go down and thus the beads fluctuate less in their motion. These findings suggest that the fluctuation in the number of actin filaments is likely the factor determining the curvature, so we developed a simple model to understand and test such mechanism.
Two possible mechanisms may contribute to the turning of beads' trajectory: turning induced by elastic and ratchet torque, and turning induced by actin tail-reorientation (see Text S1). Because of the symmetry of the spherical bead, the torque-induced rotation found in the ellipsoidal beads is negligible. Our simulations also confirm that a micron-sized spherical bead rarely rotates about its center during its motion. Therefore, the re-orientation of the tail along the bead surface is likely to be the main cause of the trajectory turning. Thus, we consider a simplistic model in which a bead of radius is propelled by randomly distributed filaments at its rear, so the filament number difference between the left and right sides of the bead is on the order of . In other words, out of filaments tend to push the bead off the current direction by an angle while the rest tend to push along the current direction of motion. The change in the direction of motion is expected to be . The typical time over which the directional bias persists is the turn-over time of the actin network, which we estimate in Text S1. Then, the typical angular velocity of the turning is , and the root-mean-square value of the curvature is One thus expects a linear relation between and with a slope of . To test whether this simple conclusion is correct, we used simulations of the hybrid model to obtain the values of , , and . We plotted the simulation results for as a function of for various values of attachment, detachment, capping and nucleation rates, as well as of actin gel elastic constant, together with the predicted linear relation, and found very good agreement except for low values of the detachment rate (see Figure 4E, Figure S10 and Figure S11 in Text S1). The higher-than-expected values of obtained from the simulations with low detachment rates are caused by the entrapment of beads into the actin gel, as mentioned above. Thus, macroscopic elastic effects influence the trajectory only in the limiting case of too many attached filaments. Otherwise, stochastic microscopic filament-ratchets are responsible for the curvature of trajectories.
Note that in contrast to our results, a non-Gaussian distribution of the curvatures of trajectories of the beads was observed in [25]. According to the model in [25], the torque balance alone determines the turning of the bead, while in our model both torque and redistribution of actin around the bead determine the trajectory. This difference suggests that the redistribution of actin probably does not play an important role in the experiments in [25]. One possibility is that the actin tail always interacts with a fixed side of the bead in these experiments, which can result from an asymmetric coating of the bead surface by the actin-nucleation promoting factors. Also note that the autocorrelation function of the simulated curvature of trajectories always decays rapidly at a sub-micron distance (see Figure S12 and details in Text S1). This result differs from the observed long-range correlation of about [24], which is possibly caused by additional long-ranged bias in the actin network near the bead-tail interface.
We simulated growth of an actin pedestal against flat elastic cantilever and force-clamped spherical bead, as in experiments [18], [19], respectively (Video S9 and Video S10). The hybrid model in these cases was used as described above, with the following differences: 1) We first generated undeformed node-spring pedestal underneath the surface to be pushed. 2) All actin network nodes were free to be positioned by the force balances (the nodes in the network did not become immobile when they were more than a few microns away from the surface) except at the very bottom. The layer of the nodes at the very bottom was immobilized. 3) The motion of the cantilever or bead was determined by the balance between the pushing/pulling forces from the filaments touching the surface and either a) the elastic restoring force from the cantilever proportional to cantilever's deflection, or b) clumped force from the bead. The speed of the cantilever or bead, , was then obtained by dividing the displacement increment of the surface by the time interval. Calibration of the model in these numerical experiments is described in Text S1. Simulation snapshots are shown in Figure 5, A and B and Figure S16 in Text S1.
The simulated force-velocity relation predicted by the hybrid model for the flat cantilever is compared to the experimental data [19] in Figure 5C. We scale the cantilever force by , which is the force at half of the maximum cantilever speed and scale to best match the rest of the data. The prediction agrees very well with the observed concave-down force-velocity relation. To quantitatively understand this result, we develop an analytical 1D theory in Text S1 and find that continuing reduction of the network stiffness due to the network disassembly during a long time of the experiment plays an important role in the shape of the force-velocity relation. A network undergoing significant disassembly in the aged gel sections recoils under a high load, reducing both net protrusion rate of the actin network pushing the cantilever and the maximum force that the network can sustain. These factors cause the rapid downturn in the force-velocity relation. Our 1D analytical result ( can be approximated as in relevant parameter range) is shown in Figure 5C and is in very good agreement with both experimental data and simulation of the 2D hybrid model.
We then used the hybrid model to simulate the force-velocity relation for the force-clamped bead. In this case, the force-velocity relation is concave-up, in good agreement with the observations [18] (Figure 5D, Figure S15 in Text S1). Qualitative explanation for this shape is that the velocities in this experiment were measured on a minute time scale before the network significantly disassembles (over a few minutes). Therefore, the network's recoil is negligible in this case and the force-velocity relation is similar to that of individual filaments. From our 1D calculation for under a constant load (see Text S1), we find , where is proportional to the disassembly rate constant of the network and is the age of the network when is measured in our simulations, and is the average velocity of individual filaments. This analytical result is also shown in Figure 5D, in very good agreement with the simulation results of the hybrid model.
To investigate the effect of the filament attachments to the surface on the force-velocity relations, we varied the value of the attachment rate to change the ratio of the number of attached to the number of pushing filaments, . The simulated force-velocity relations for different ratios are shown in Figure S13 in Text S1. For both cantilever and force-clamped experiment, we find that increasing the fraction of attached filaments decreases both velocity and stall force without changing the qualitative shape of the force-velocity curve, consistent with the idea that attached filaments counteract the pushing filaments. Finally, to confirm that it is the actin dynamics rather than the shape of the surface that determines the force-velocity relation, we swapped the shapes of the flat cantilever and round bead used in the two experiments. We considered two cases: a slow-growing actin network against a curved surface of a cantilever, and a fast-growing actin network against a flat force-clamped object. The simulation results shown in Figure S13 and Figure S15 in Text S1 illustrate that the force-velocity relations in both experiments remain qualitatively the same (concave-down and concave-up, respectively). Therefore, the shape of the surface does not appear to affect the overall shape of the force-velocity relation.
Complexity of the relation between geometry of the curved surface, molecular pathways of actin polymerization against this surface and resulting force [29] indicates that the actin-based force-generation is a multi-scale phenomenon, understanding of which requires a combination of macroscopic and microscopic mechanisms. We developed such hybrid model of the actin network growing and pushing against rigid surfaces, in which actin filaments interacting directly with the surface are treated as tethered-ratchet filaments, while other filaments are considered implicitly as parts of viscoelastic node-spring network.
The elastic propulsion theory predicts that squeezing of the ellipsoidal beads orients them so that motility along the long axes ensues, while geometric effect of spreading of branching actin filaments results in beads moving along their short axes. Separately, the existing theories cannot explain the observed bi-orientation of the beads. Our hybrid model posits that the combination of the elastic squeezing and geometric spreading leads to bi-orientation and reversible switching between two orientations, in agreement with the observations. To test the hybrid theory in the future, we propose to vary the bead geometry and concentrations of actin accessory proteins, thus modulating the network stiffness and interactions with the surface. Our model makes specific, nontrivial and testable predictions (see Figure 2, E–G) for such experiments.
The hybrid model reproduces the observed order of magnitude of curvatures of the trajectories in 2D and suggests that switching between the low- and high-curvature trajectories is caused by the temporary entrapment of the beads in the actin gel. The model predicts a Gaussian distribution of the curvatures for fast-moving beads due to random fluctuations of filament numbers and redistribution of actin around the bead's surface. In agreement with observations, our simulations show an additional sharp peak at zero curvature in the curvature distribution for slowly-moving beads. Importantly, the model suggests that elastic effects have little impact on the distribution of trajectory curvatures for fast-moving beads, while for beads that tend to be trapped in the actin cloud due to frequent filament attachments, the elastic effects are responsible for deviations from Gaussian distributions.
The hybrid model posits that the qualitative difference between two force-velocity measurements [18], [19] stems from the characteristic time difference: when the measurement is made over a long time interval [19], the viscoelastic recoil of the older, aging part of the network near the base of actin pedestal cancels protrusion and causes the concave-down force-velocity relation. On the other hand, when the force is clamped and the experiment is performed over shorter times [18], the concave-up force-velocity relation is predicted. A possible way to test our model is to use fluorescent speckle microscopy to measure the kymograph of material points of the actin network that move with the recoiling network away from the surface being pushed. We predict the resulting curves for two considered experiments in Figure S14 in Text S1. Note, that there are alternative explanations for the result [19]. For example, theory in [30] based on a representation of the actin network as a viscoelastic solid could predict a different kymograph. Finally, the model proposes that the shape of the surface does not qualitatively affect the shape of the force-velocity relation.
In the present form, our model has a number of limitations. The main one is that due to computational time limitations, we simulated the model in 2D as a simplification of a 3D system. So, rigorously speaking, all our results are applicable to cylindrical, rather than spherical objects. In Ref. [28], we already attempted the 3D modeling, albeit of an oversimplified model. Preliminary indications from that attempt are that most of the 2D model predictions survive in 3D. However, there are effects of higher dimension: 3D viscoelastic theory and experiment [27] suggest that ellipsoidal beads break through the actin cloud sideways, while [28] reports the observed lengthwise symmetry breaking of the ellipsoidal beads. This problem remains open, and thus more 3D modeling is necessary. In addition, helical and more complex trajectories of actin-propelled beads that have been observed in 3D environments [23], [24] cannot be captured by our 2D model. Furthermore, our model is coarse-grained and neglects important fine-scale processes such as hydrolysis of ATP bound to polymerized actin [31]–[33], exact actin branching angles [34], indirect synergy between capping and branching [35], molecular details of the nano-scale protrusion [36] and dependence of the branching rate on filament bending [37]. Future incorporation of these details into the model will clarify molecular nature of the mixture of nucleation-based and autocatalytic actin growth posited in the model.
Due to these limitations, our model does not capture some observed effects. Notably, the simulations do not reproduce observed hysteresis in the growth velocity of actin networks under force [19], which likely depends on complex dynamic features of the network [34], [38] that are not incorporated into our model. Similarly, not reproducing deviations from the Gaussian distribution of the curvatures of trajectories [25] likely means that some inhomogeneities in the distribution of actin nucleation promoting-factors not included into the model play an important role. These inhomogeneities and 3D effects also have to be built into the model to reproduce helical trajectories reported in [21], [23].
Another open question is relation of our model to other theories of the actin-based propulsion. Those include microscopic models of propulsion by tethered actin filaments [39], [40] that can in principle be used as boundary conditions for the viscoelastic actin gels and tested by simulations similar to those done here. Two mesoscopic models, very different from ours, were proposed recently. One of them considers excluded volume effects [41], another is a liquid of dendritic clusters model [42]; both of them successfully reproduce the concave-down force-velocity curve. It is likely that subtle physical effects on which these models are based complement elastic deformations and individual filament ratchet forces of our model. In the future, after including interactions of the filaments with cell membrane [43]–[46], contractile myosin effects [47] and more adequate actin rheology [48], our model can be applied to the general problem of cell protrusion.
Motility experiments on ellipsoidal beads were carried out in the lab of J. Theriot as previously described [28]. Briefly, 1- carboxylated polystyrene microspheres (Polysciences, Warrington, PA) were placed in a viscoelastic matrix (6% polyvinyl alcohol), heated to , and stretched uniaxially. The film containing the beads was cooled and dissolved using an isopropanol/water mixture to recover the beads before functionalizing their surfaces with carboxylate. Electron microscopy showed that the beads had average dimensions of , with an average aspect ratio of 2.2. His-tagged ActA was purified and adsorbed on the surface of beads at saturating amounts. ActA-coated beads were then added to Xenopus laevis egg cytoplasmic extract, which was diluted to 40% of the original protein concentration. The slide chamber depth was restricted using 2- silica spherical beads. Note, that the ActA-coated motile beads were contained between two parallel coverslips and restricted from moving perpendicularly to the coverslips, and thus the trajectories of the beads were two-dimensional. All time-lapse sequences taken during the steady-state bead motility were acquired between 2 and 4 h after preparing the slide. Phase-contrast and fluorescence images were acquired as described in [28].
Spherical beads were prepared in the lab of J. Theriot as previously described [5], which is similar to that for ellipsoidal beads except for the stretching treatment. Bead trajectories were recorded at 10 s intervals.
For both experiments, positions and orientations of beads were computed from phase-contrast images and assembled into tracks as described in [28]. Smoothing of the instantaneous angular velocity values of the beads was generated using a weighted average of five nearest neighbors and a cubic equation as described in [28]. The angular velocity fit-in was generated using a seventh-order polynomial function. The curvature was obtained by dividing the resulting angular velocity by the instantaneous linear speed of the bead.
In the hybrid model (Figure 1C), arrays of actin filaments interacting directly with the surface of the bead are treated as effective individual filaments, while other (not in touch with the surface) filaments are not modeled explicitly but rather treated as the network of elastic springs interconnected by nodes. The model is formulated and all simulations are done in 2D, which is a simplification of a 3D system. We assume that new filaments are created around the surface via a mixture of spontaneous nucleation, which has a spatially uniform rate along the bead surface, and autocatalytic branching processes, which has a rate proportional to the local density of existing filaments (not necessarily uniform in space). Separately, either of these processes produces a defective actin tail (see Figure S4 and discussion in Text S1). We also assume that newly created filaments immediately anchor to the network at their pointed ends which become new nodes of the network. In the simulations, this step is achieved by connecting each pointed end with undeformed springs to up to 4 neighboring nodes in the network that are within from the pointed end (see Figure S3 in Text S1). Thus, creation of new filaments dynamically expands the actin network. We treat filaments as elastic springs that are created in an attached and undeformed state. When stretched, attached filaments produce resisting forces that are proportional to their deformations. Attached filaments undergo detachment with a rate that increases exponentially with the load force. After detachment, filaments become free and are able to elongate and produce pushing forces against the obstacle. Free filaments are treated as linear elastic springs with the rest length growing with the polymerization rate. This rate is a function of the load on the barbed end of the filament; the function is given by the individual filament force-velocity relation that follows from the Brownian ratchet theory. The pushing force that a free filament exerts on the surface is computed as follows: at each time step, a virtual ‘penetration’ distance of the barbed end of the rest-length spring, corresponding to the filament, into the bead is computed. The filament is assumed to be deformed by this penetration distance, and respective elastic force is the pushing force. Free filaments can re-attach to the surface and get capped at constant rates.
Once capped, the filament is removed from the simulation, since in reality it will stop growing and cannot attach to the surface to exert pulling forces. However, the node corresponding to the pointed end of the filament remains, so this filament effectively becomes a part of the deformable network. We do not track the orientation of individual pushing filaments, but treat them as coarse-grained clusters of actual filaments that always push perpendicularly to the obstacle surface (see Figure 1D). As filaments exert forces on the obstacle, they also apply opposite forces to the elastic network that they are anchored to, causing network deformations (see Figure 1D). Similarly, the stress in the deformed network is transferred to the bead surface through the interacting filaments.
The deformation of the network is represented by the motion of nodes and springs in the network, which is obtained by moving all the nodes toward their force-equilibrium positions at each time step. For actin-propelled beads, we assume that the nodes in the network become immobile when they are more than a few microns away from the bead surface, representing the adhesion of the actin tail to the substrate. The bead moves and rotates to satisfy the force and torque balances from the filaments. For the force-velocity measurements, we fix the network at the bottom and allow all the rest nodes to move to reach force balance. The network undergoes disassembly, which is treated by removing the nodes and their connected springs from the network randomly with a rate proportional to the number of existing nodes. We have also included the effect of rupture of crosslinks by introducing a critical stretching force, above which the links break and get removed from the network. During the steady motion of beads, the creation and extinction rates of actin networks balance, causing a treadmilling actin tail behind the bead (Video S1). Effective viscoelastic behavior of the actin network emerges from the disassembly and breaking of the network. Further details about the model equations and parameters are described in Text S1.
|
10.1371/journal.ppat.1003521 | Rational Design of a Live Attenuated Dengue Vaccine: 2′-O-Methyltransferase Mutants Are Highly Attenuated and Immunogenic in Mice and Macaques | Dengue virus is transmitted by Aedes mosquitoes and infects at least 100 million people every year. Progressive urbanization in Asia and South-Central America and the geographic expansion of Aedes mosquito habitats have accelerated the global spread of dengue, resulting in a continuously increasing number of cases. A cost-effective, safe vaccine conferring protection with ideally a single injection could stop dengue transmission. Current vaccine candidates require several booster injections or do not provide protection against all four serotypes. Here we demonstrate that dengue virus mutants lacking 2′-O-methyltransferase activity are highly sensitive to type I IFN inhibition. The mutant viruses are attenuated in mice and rhesus monkeys and elicit a strong adaptive immune response. Monkeys immunized with a single dose of 2′-O-methyltransferase mutant virus showed 100% sero-conversion even when a dose as low as 1,000 plaque forming units was administrated. Animals were fully protected against a homologous challenge. Furthermore, mosquitoes feeding on blood containing the mutant virus were not infected, whereas those feeding on blood containing wild-type virus were infected and thus able to transmit it. These results show the potential of 2′-O-methyltransferase mutant virus as a safe, rationally designed dengue vaccine that restrains itself due to the increased susceptibility to the host's innate immune response.
| The four serotypes of dengue virus cause severe outbreaks globally in tropical countries with thousands of patients requiring hospitalization. The health care and indirect economic cost of dengue in endemic countries is huge. Despite this, no clinically approved vaccine or antiviral treatment is currently available. Dengue transmission could be stopped with a vaccine that provides full protection to all serotypes. Dengue afflicts many developing countries and a vaccine should therefore be cost-effective and should provide protection with ideally a single injection. Here we present a novel dengue vaccine approach that harbours mutation(s) in the 2′-O-methyltransferase (MTase), a viral enzyme that methylates viral RNA as a strategy to escape the host immune response. Non-methylated RNA is recognized as “foreign” and triggers an interferon response in the cell. The MTase mutant virus is immediately recognized by the host's immune response and hardly has a chance to spread in the organism while an immune response is efficiently triggered by the initially infected cells. Mice and monkeys infected with the mutant virus developed an immune response that fully protected them from a challenge with wild-type virus. Furthermore, we show that MTase mutant dengue virus cannot infect Aedes mosquitoes. Collectively, the results suggest 2′-O-MTase mutant dengue virus as a safe, highly immunogenic vaccine approach.
| Dengue virus (DENV) is a member of the Flaviviridae family. DENV infection causes dengue fever (DF) and the more severe forms of the disease, dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). DENV has four serotypes (DENV-1 to -4), each of which is capable of causing severe disease. The frequency, severity, and geographical spread of cases have increased over the past decades [1], [2]. Every year, one hundred million new cases of DF and 250,000 DHF/DSS are estimated by the WHO. At present, despite intensive global research efforts, no vaccine or antiviral treatment for dengue infection is available. Vaccine development is complex due to multiple factors. (i) An effective vaccine must consist of a tetravalent formulation protecting against each of the four serotypes because more than one serotype typically circulates in a region. (ii) A sub-protective vaccine potentially increases the risk of vaccinees to develop the more severe forms of dengue during repeated infection because of a known association of pre-existing immunity with severity [3], [4]. (iii) Since most infections occur in developing countries, an ideal vaccine should be affordable and fully protective [5]. Taken together, a vaccine inducing a robust level of immunity ideally with only one inoculation is required.
Live-attenuated vaccines are replication-competent viruses, which can induce an immune response and an immune memory that are comparable to those induced by the wild-type virus. Live-attenuated viruses do not cause disease because of the low level of replication and hence low levels of inflammation. Prominent examples of successful live-attenuated vaccines providing long-term immunity are those against vaccinia virus, poliovirus (Sabin), and two members of the Flaviviridae family, yellow fever virus (YF-17D) and Japanese encephalitis virus (JEV SA14-14-2) [6]. Live-attenuated DENV vaccines have been shown to induce protective neutralizing antibody titers in mice, monkeys, and humans [7]–[9]. In addition, evidence that a balanced T cell response contributes to protection is accumulating, emphasizing the importance of T cell epitopes in a vaccine [8].
Flaviviruses are positive-sense, single-stranded RNA viruses. The flavivirus genome encodes for 3 structural (C, prM, and E) and 7 non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5). NS5 is a multifunctional protein, consisting of the RNA-dependent RNA polymerase [10] and methyltransferase (MTase) activities responsible for 5′ RNA cap formation [11], [12] as well as internal RNA methylation [13]. While N-7-methylation is essential for RNA translation and stability, the function of 2′-O-methylation has remained elusive until recently. We and others demonstrated that while 2′-O-MTase is not essential for viral replication in vitro, viruses bearing mutations in the highly conserved MTase catalytic K-D-K-E tetrad are severely attenuated in the host due to the inability of the virus to shield viral RNA from recognition by host innate immune factors [14], [15]. DENV RNA binds to RIG-I and MDA5 [16], [17], which activates interferon (IFN)-β production via a cascade involving IFN-β promoter stimulator 1 (IPS-1) [17]. IFNs in turn activate IFN stimulated genes (ISGs), which induce antiviral responses in infected and neighbouring cells. IFN-induced proteins with tetratricopeptide repeats (IFITs) are critical for the inhibition of viral infections, although their functions are only partially understood [14], [18]. The human IFIT gene family comprises four members: IFIT1, IFIT2, IFIT3 ( = IFIT4), and IFIT5; whereas mice only express IFIT1, 2 and 3 [19]. Interestingly, IFIT homologs are conserved from amphibians to mammals, suggesting that they play a central role in the innate immune response [19]. IFIT1 and 2 bind to eukaryotic Initiation Factor 3 (eIF3) and inhibit translation [20], whereas IFIT3 amplifies the antiviral signal by connecting IPS-1 and TBK1, resulting in more IFN production [21]. The role of human IFIT5 is less well understood.
Here we demonstrate that DENV strains bearing a mutation in the catalytic site of the 2′-O-MTase replicated to high titres in cell culture whereas these mutant viruses were highly attenuated in mice and rhesus monkeys. The mutation was stable over several passages and reversion to wild-type (WT) was not observed. For further safety improvement, a second mutation in the 2′-O-MTase catalytic tetrad was introduced without affecting the viability of the virus in vitro. A single dose administration to rhesus macaques (RM) conferred protection to homologous DENV challenge. Mice immunized with a single dose of a divalent (DENV-1/2) formulation of the mutant viruses and mice immunized with the monovalent formulation showed comparable antibody responses, demonstrating that there was no interference between two serotypes of the DENV MTase mutants. Moreover, no enhanced infection and increased TNF-α levels were observed in immunized mice upon challenge with heterologous virus. Overexpression of IFITs in HEK-DC-SIGN cells suggested a role for IFIT1 in the attenuation of MTase mutant in human cells. Taken together, these results demonstrate the potential of 2′-O-MTase mutants as a DENV vaccine. To our knowledge, this is the first live-attenuated rational vaccine approach, tailored to optimally activate the innate and adaptive immune response while being severely attenuated due to its susceptibility to the IFN response.
Flaviviruses replicate in the cytoplasm. The cytoplasm-replicating viruses have evolved N7- and 2′-O-methyltransferases (MTase) to methylate their viral mRNA 5′ cap structures [22]. We have previously shown for West Nile virus (WNV) and DENV-1 that mutation of the Asp of the tetrad K-D-K-E completely abolished both N7- and 2′-O-MTase activities and was lethal for viral replication; mutations of the other three residues of the tetrad abolished 2′-O-methylation (with a decrease in N7-methylation), and led to attenuated viruses [14], [23]. Since there are four serotypes of DENV, we introduced the same MTase mutations into DENV-2 to examine whether the same approach was feasible with more than one serotype. A WT recombinant MTase, representing the N-terminal 296 amino acids of the DENV-2 NS5 (strain TSV01), was cloned and expressed. Two mutant MTases containing Ala-substitutions at the K-D-K-E tetrad (Fig. 1A) were prepared: one with a single E217A mutation and another with double K61A+E217A mutations. The mutant enzymes retained 95% and 77% of the WT N7-methylation activity, respectively; neither mutant exhibited any 2′-O-methylation activity (Fig. 1B). BHK-21 cells transfected with equal amounts of WT and mutant (E217A and K61A+E217A) genome-length RNAs of DENV-2 generated equivalent numbers of viral E protein-expressing cells (Fig. 1C). Both WT and mutant RNAs produced infectious viruses (passage 0) with similar plaque morphologies (Fig. 1D). The replication of mutant viruses was attenuated in mammalian Vero and mosquito C3/36 cells (Fig. 1E). Continuous culturing of the mutant viruses on Vero cells or HEK-293 cells expressing DC-SIGN (HEK-DC-SIGN) for ten rounds (3–4 days per round) did not change their plaque morphologies (Fig. 1D and data not shown). The expression of DC-SIGN facilitates DENV infection [24]. Sequencing of the passage 0 and 10 viruses from both Vero and HEK-DC-SIGN cells showed that the engineered mutations were retained (Supplementary Fig. S1a and S1b). Similar results were obtained for DENV-1 containing the E216A (E216 in DENV-1 MTase is equivalent to E217 in DENV-2 MTase) or K61A+E216A mutation in MTase (Supplementary Fig. S2). Collectively, the results demonstrate that the 2′-O-MTase mutant DENV-1 and -2 are slightly attenuated, but stable in cell culture.
We infected AG129 mice with the WT and 2′-O-MTase mutants (called “E216A” for DENV-1 and “E217A” for DENV-2 from this point) to assess viral replication and immunogenicity in vivo. AG129 mice lack the receptors for type I and type II IFNs, and have been used widely for antiviral and vaccine testing [25]–[28]. Mice were intraperitoneally (i.p.) infected with 2.75×105 plaque-forming unit (PFU) of WT or mutant viruses. The viremia result showed that mutating K61A or E216A in DENV-1 and mutating E217A in DENV-2 attenuated the virus compared to the WT virus (Fig. 2A and B).
Next, we examined a combination of two MTase mutants (E216A and E217A) representing DENV-1 and DENV-2 to address a potential competition effect that has been described previously with attenuated strains in humans [29] and in mice [25]. To this end, mice were injected i.p. with 2.75×105 PFU of E216A or E217A or a combination of both (a total of 5.5×105 PFU viruses). At 30 days post immunization, mice were challenged i.p. with 1×106 PFU of WT DENV-1 or 5×106 WT DENV-2. DENV specific IgG titers and viremia were observed. All mice immunized with E216A and/or E217A were protected against homologous challenge (Fig. 2C), demonstrating that the immune response was protective even though the IgG titers in E216A and/or E217A-infected mice were 2 to10 times lower than those in the WT virus-infected mice (Fig. 2D and E).
A general concern for live attenuated vaccines is their theoretical potential to mutate back to WT under pressure of the immune system. To address this in our system, virus from mice infected with mutant DENV1 or DENV2 was isolated at day 3 after infection and the mutations were found to be stable (Supplementary Fig. S1c). To rule out that compensatory mutations were introduced into the viral genome the input and output (day 3 after infection) virus was sequenced using Illumina deep sequencing technology. As summarized in Supplementary Table S1, only the single nucleotide polymorphisms (SNPs) responsible for the E216A or E217A mutation were found when comparing the sequences to wild-type DENV-1 or -2, respectively.
We next compared the neutralization and infection enhancing capacity of serum collected 30 days post immunization (Table 1 and Supplementary Fig. S3) [30]. Mutant viruses cause the same or less antibody-dependent enhancement (ADE) than the respective wild-type viruses in the heterologous setting (0.51±0.16 vs. 0.74±0.2 for DENV-1 immunization and ADE tested against DENV-2 and 0.64±0.22 vs. 0.62±0.14 for DENV-2 immunization and ADE tested against DENV-1) (Table 1). More importantly, we did not observe enhanced infection in vivo (Fig. 2C and see later challenge experiments with a virulent DENV-2 strain). These data suggest that vaccination with the E216A/E217A mutants does not cause ADE during heterologous challenge even though lower neutralizing Ab titers are generated by the mutant strains compared to the wild-type virus.
While antibodies are crucial to reduce the viral load by binding and neutralizing virus particles, T cells are necessary for efficient viral clearance [31], [32]. AG129 mice are not suitable to study T cell responses because of their lack of IFN-γ signaling, which is critical to activate T cells. We therefore used IFNAR mice lacking the receptor for IFN-α/β [33]. IFNAR mice were immunized with 2.75×105 Pfu DENV-2 E217A or DENV-2 WT and spleens were harvested at day 7 for restimulation in vitro and detection of IFN-γ production (Fig. 3A). Mutant and WT virus elicited a strong CD4 and CD8 T cell response after re-stimulation with DENV-2. The CD4 response was weaker in E217A-immunized mice, likely due to the lower total viral load in E217A-immunized mice compared to mice immunized with the WT virus (Fig. 3B). To test for targeted DENV T cell response splenocytes were re-stimulated with a pool of NS4B and NS5 CD8 peptides described by Yauch et al [32]. No significant difference in the NS4B and NS5-specific T cell response was seen between mice immunized with E217A or WT DENV-2 (Fig. 3B). Taken together, DENV 2′-O-MTase mutants induce a T cell response and epitope presentation that is similar to WT infection. Nevertheless, additional studies in mice and monkeys are necessary to assess the T cell response in greater detail and to test its functional contribution to protection.
DENV-1 strain 05K3126 and DENV-2 strain TSV01 do not cause pathology in mice. To test for protection against a more virulent strain we immunized mice with DENV-1 E216A, DENV-2 E217A, a mixture of E216A and E217A, WT DENV-1 (Westpac) or WT DENV-2 (TSV01) or PBS and challenged them with the virulent DENV-2 strain D2Y98P [34] 30 days later (Fig. 4). DENV-2 E217A protected against the homologous challenge (Fig. 4A). Immunization with DENV-1 E216A protected 70% of the mice, showing limited cross-protection after infection with D2Y98P (Fig. 4A and 4B). No enhanced disease was detected after heterologous challenge. Increased TNF-α levels were associated with pathology in the AG129 mouse model in the context of ADE [35]. To further assess the possibility of ADE-associated pathology, we measured TNF-α levels in plasma three days after challenge. High levels of TNF-α were only detected in unimmunized (PBS) mice, showing that TNF-α as a marker of pathology was independent of ADE, and that immunization with E216A did not cause ADE after heterologous challenge. These data demonstrate that immunization with E217A protects mice against challenge with an aggressive, virulent DENV-2 strain that causes 100% mortality in unimmunized mice.
To assess the safety (viremia profile) and efficacy (neutralizing antibody response and protection against challenge) of the 2′-O-MTase mutant DENV vaccine approach in an immunologically competent host, three groups of Rhesus monkeys (RM) were immunized with different doses of E217A. One group received a low dose (1×103 PFU), one group a medium dose (1×104 PFU), and one group a high dose (1×105 PFU) of E217A virus. Viremia was monitored during 10 days after inoculation. The E217A virus was severely attenuated, and no viremia was detected except for one animal (R0105) that had received a high dose (1×105 PFU) and developed a low viremia (Table 2). Virus was extracted for sequencing, and it was confirmed that the E217A mutation was retained in the virus extracted at days 3, 4 and 7 from this animal. Importantly, full virus genome sequencing of the viral RNA recovered at day 7 showed that no compensatory mutations were introduced (data not shown). All immunized monkeys developed neutralizing antibodies to DENV-2 on day 15 after immunization (Table 3). ADE was analyzed in a K562 assay and a similar enhancement pattern was observed for both heterologous and homologous infection in vitro: ADE correlated with the neutralizing titer, ie the higher the NT50 the higher the enhancement (Supplementary Fig. S4). This argues against a physiologically relevant infection enhancement, which would only be expected after heterologous infection. By day 30 after immunization, all monkeys including the ones with low dose immunization developed high titers (GMT≥92) of neutralizing antibodies (Table 3). The monkeys were then challenged with 1×105 PFU of WT DENV-2 on day 64 post-immunization. No viremia was detected in all immunized monkey, whereas all four PBS-immunized controls had a mean peak virus titer of 2.5 log10 PFU/ml and mean viremia duration of 4.8 days (Table 4). In all animals except one (R0055), anamnestic antibody responses were observed after challenge (Table 3). These data demonstrate that live, attenuated DENV MTase mutant virus, even when administrated at low dose (1×103 PFU), can induce protective immunity in non-human primates.
The 2′-O-methylation of the 5′ cap of WNV and coronavirus RNA functions to subvert innate host antiviral response through escape of IFIT-mediated suppression [14], [15]. To assess whether this is true for DENV as well, we pretreated HEK-DC-SIGN cells with an increasing dose of IFN-β for 24 h. While HEK-DC-SIGN cells are susceptible to type I IFN, they do not produce detectable levels of IFN-β after infection with mutant or WT virus (data not shown). The IFN-β-treated cells were infected with WT or mutant E217A DENV-2. The E217A virus was significantly more sensitive to IFN-β pretreatment than the WT virus, as demonstrated by the percentage of infected cells (Fig. 5A) as well as the viral titers in culture supernatants (Fig. 5B). To test the stability of the mutation under IFN pressure and in different cell types we passaged the virus in the presence of 0, 20 and 200 U/ml IFN-β in HEK-DC-SIGN and U937-DC-SIGN. As illustrated in Supplementary Fig. S5, E217A was lost in the presence of IFN, whereas wild-type virus resisted the IFN pressure in both cell lines. E217A isolated from passage three in HEK-DC-SIGN and from passage one in U937-DC-SIGN was isolated for sequencing. The E217A mutation was retained and no compensatory mutations were introduced (data not shown).
To elucidate the molecular mechanism of attenuation, we over-expressed human IFIT1, 2, 3, or 5 in HEK-DC-SIGN cells. The cells were infected with WT or mutant DENV-2 and assessed for the number of infected cells by flow cytometry (Fig. 5C). The WT virus infection was not affected, whereas E217A mutants were significantly inhibited by IFIT1, but not IFIT2, 3, or 5. However, IFIT1 over-expression did not completely block E217A infection nor did it affect virus output from the infected cells (Fig. 5D), suggesting that other IFN-mediated signals are involved in the response against DENV. Both mutant and WT virus show similar growth kinetics in untreated cells (Fig. 5E). We currently don't know why the mutant virus is attenuated in Vero cells but not in HEK-DC-SIGN since both lines are deficient in IFN production. It should be noted that the maximum antiviral effect of IFITs could be underestimated due to the low transfection efficiency (30–50%) of the IFIT-expressing plasmids.
We compared the effect of 2′-O-MTase mutation on viral fitness in mosquito Ae. aegypti, the natural transmission vector for DENV. The mosquitoes were fed with blood containing WT or E217A. After the mosquitoes were fed at a titer of 1×105 PFU/ml, significant differences in oral infection and dissemination between the WT and mutant viruses were observed 15 days post-infection (Table 5). The WT virus infected 29% of mosquitoes at the highest titer (1×105 PFU/ml), but only 1–6% of mosquitoes at lower titers (1×103 and 1×104 PFU/ml). When orally fed with 1×105 PFU/ml WT virus, approximately 10% of mosquitoes were infected after 9; the WT virus disseminated in 24% of the mosquitoes (Table 5). When fed with 1×103 and 1×104 PFU/ml WT virus, the dissemination rates reached 1–4%. In contrast, the mutant virus was unable to infect the Ae. aegypti and, subsequently, no dissemination was observed for all titers (Table 5).
To examine whether the E217A mutant could replicate in vivo, we intra-thoracically inoculated the WT and mutant viruses into Ae. aegypti mosquitoes. Intra-thoracic inoculation bypasses the mosquito midgut, which is the key barrier to establish infection during natural feeding route. Both WT and mutant viruses reached 100% infection rate upon intra-thoracic inoculation. The mean genome copy number reached 4.6×109 and 6.2×109, respectively (Supplementary Fig. S6). The genome copy number of the WT virus was approximately 35% higher than that of the mutant virus (p = 0.1054). Overall, the results demonstrate that the 2′-O-MTase mutant virus is compromised in vector fitness.
Various dengue vaccine strategies are currently under development, including live attenuated virus, subunit vaccines, chimeric viruses, and DNA vaccines [36], [37]. The YFV 17D-based chimeric dengue vaccine developed by Sanofi-Pasteur is the most advanced in clinical testing [38], [39]. The establishment of reverse genetic manipulation of DENV has greatly facilitated the generation of promising vaccine candidates [36], [38]. The recent progress in understanding the mechanism of attenuation of 2′-O-MTase mutant flaviviruses has provided a novel approach for vaccine and antiviral development [40]. Here we show in a proof-of-concept study that MTase mutant E216A DENV-1 and E217A DENV-2 strains are stable in vitro, and safe and immunogenic in vivo. Importantly, enhancement of infection was not observed after heterologous infection of immunized mice. The fear in a clinical setting is that sub-neutralizing titers of antibodies could enhance infections, even though this has so far not happened in the context of vaccine trials in humans [41]. A commonly used approach to address ADE in vitro is to infect K562 cells in the presence of antibodies. Virus alone is not able to infect K562 cells efficiently, whereas virus-antibody immune complexes bind to K562 cells via Fc-γ receptors (FcγR's), assisting the internalization of the virus and infection of the cells. We found that K562 cells could be infected in the presence of serum from immunized mice and monkeys at dilutions that were approximately 50% neutralizing in the U937-DC-SIGN system (Supplementary Fig. S3 and S4). This is in line with a previous report, which found that even strongly neutralizing antibodies are enhancing at concentrations that are close to the 50% neutralizing titer [42]. Clinically relevant ADE would be expected at sub-neutralizing titers and only after heterologous infection, and this was not observed in our experiments. A caveat of the K562 system is that the cells do not express inhibitory FcγRIIb, which is present on human target cells (dendritic cells and macrophages) and which negatively regulates ADE [43], [44]. Physiological amounts of complement, another negative regulator of ADE, are also not taken into account [45]. In summary, while the K562 assays done here did not show more ADE for heterologous infections, we cannot exclude ADE because of the limitations of the assay. Potential ADE will have to be addressed in further monkey studies.
Live attenuated dengue vaccine candidates have several advantages. Importantly, they can induce long lasting humoral and cellular immune responses to both structural and non-structural viral proteins. In this study we show a CD8 response to NS4B and NS5 peptides that is similar in mice immunized with mutant or WT virus, suggesting that the response is qualitatively equivalent. Chimeric viruses using the same backbone for all four DENV serotype glycoproteins would induce a type-specific response restricted to the structural proteins of one DENV serotype [36]. The interdependence of the T and B cell response for the efficient generation of immune memory has been demonstrated in a number of human studies [46], [47]. We speculate that the advantage of an attenuated non-chimeric DENV that includes all naturally occurring T and B cell epitopes could be that only one vaccination is required to confer long-term immunity to re-infection, as seen for natural DENV infections [48], [49]. A single-dose vaccine would facilitate the logistics of a vaccination program and would significantly reduce its cost compared to candidates requiring several booster immunizations. The 2′-O-MTase mutant DENV vaccine approach, with a known mechanism of attenuation, can be readily generated using a reverse genetic system. This is in contrast to the method to develop live attenuated vaccines by passaging of WT viruses in cell lines, leading to the introduction of random mutations. The reverse genetic system-based rational vaccine ensures that the vaccine maintains the attenuated genotype. Additionally, a tetravalent formulation would contain the same attenuating mutation in all four serotype recombinant vaccine strains, making the generation of a more pathogenic virus by intra vaccine-strain recombination impossible. Moreover, recombination in cell culture is hardly observed in flaviviruses, suggesting that flaviviruses are not prone to evolution by recombination [50]. By introducing additional mutations in the K-D-K-E tetrad of 2′-O-MTase, further safety and attenuation can be achieved. Only a virus that has at least two mutations will be acceptable in the clinical setting. Our data demonstrate that the 2′-O-MTase E217A virus is attenuated in mice and monkeys. We cannot explain fully why the 2′-O-MTase mutant virus was attenuated (10-fold lower virus titer compared to WT virus) in AG129 mice, which are unable to respond to IFN-signals. It is likely that pattern recognition receptors and downstream pathways activated by the mutant virus trigger antiviral defense mechanisms in an IFN-dependent and IFN-independent manner.
Whether the balance between low virulence and high immunogenicity is achieved in humans by 2′-O-MTase mutants remains to be elucidated. Our studies in human HEK293 cells show increased susceptibility of DENV2 E217A mutant to IFN-β in vitro, suggesting that DENV E217A mutants will be attenuated in humans as well. In the monkey immunization experiments, one monkey out of four in the high dose group experienced peak viremia of about 100 PFU, which is comparable to other live attenuated vaccine candidates [51]. Indeed, weak replication of the vaccine approach is desirable in order to induce a strong protective cellular immune response. Replication should be restricted enough to preclude onset of illness, whereas sub-clinical symptoms such as mild rash, transient leukopenia, and mildly elevated liver enzyme values are generally accepted [52]–[54]. Furthermore, studies with murine hepatitis virus have shown that MTase mutants are highly attenuated in its natural host, induce IFN, which could further induce the immunogenicity of a vaccine, and are genetically stable in vivo [15]. Moreover, the replication level of WNV 2′-O-MTase mutant in mice was largely decreased in the spleen, serum, or brain in comparison with the WT WNV infection. Intracranial inoculation of 1×105 PFU of 2′-O-MTase mutant WNV did not cause any mortality and morbidity in mice, demonstrating the safety of this vaccine approach [14]. Taken together, these evidences demonstrate the safety and immunogenicity of the MTase-mutant vaccine approach. We are currently working on the tetravalent formulation to develop the strategy towards a clinical application.
All experimental procedures involving Rhesus Monkeys were approved by and carried out in strict accordance with the guidelines of the Animal Experiment Committee of State Key Laboratory of Pathogen and Biosecurity, Beijing, China. All procedures were performed under sodium pentobarbital anesthesia by trained technicians and all efforts were made to ameliorate the welfare and to minimize animal suffering in accordance with the “Weatherall report for the use of non-human primates” recommendations.
The mouse experiments were conducted according to the rules and guidelines of the Agri-Food and Veterinary Authority (AVA) and the National Advisory Committee for Laboratory Animal Research (NACLAR), Singapore. The experiments were reviewed and approved by the Institutional review board of Biological Resource Center, Singapore (IACUC protocols 90474 and 100536).
BHK-21, C6/36, and HEK-293 were purchased from the American type culture collection (http://www.atcc.org). HEK-293 and U937 cells expressing DC-SIGN were obtained by lentiviral transfection and subsequent cell sorting. All cells were maintained in minimal essential medium supplemented with fetal bovine serum (5%–10%).
WT MTases representing the N-terminal 262 and 296 amino acids of DENV-1 and -2 NS5, respectively, were cloned, expressed, and purified as reported previously [11]. Mutagenesis of MTase was performed using QuikChange II XL site-directed mutagenesis kit (Stratagene). The complete sequence of each mutant MTase was verified by DNA sequencing. N7- and 2′-O-methylation assays were performed as described before [11].
Full-length infectious cDNA clones of DENV-1 (Western Pacific 74 strain) and DENV-2 (TSV01 strain) [55], [56] were used to generate WT and mutant viruses. A standard mutagenesis protocol was used to engineer mutations into the MTase region as reported previously [11]. The protocols for in vitro transcription, RNA transfection, IFA, plaque assay, and growth kinetics were reported previously [23]. Strain D2Y98P was described previously [34].
Female or male 6–8 week old IFN α/β/γ receptor deficient mice (AG129) were purchased from B&K Universal Limited with permission from Dr. M. Aguet (ISREC, School of Life Sciences Ecole Polytechnique Fédérale (EPFL)). IFN α/β receptor deficient mice (IFNAR) on a C57BL/6 background were provided by Prof. Ulrich Kalinke [33]. All mice were bred and kept under specific pathogen-free conditions at the Biomedical Resource Centre, Singapore. For immunization, BHK-21 derived mutant and WT viruses were used. For challenge experiments DENV produced in C6/36 cells was used.
Fourteen RMs, weighing from 3.4 to 5.0 kg, were pre-screened negative for IgG antibodies against DENV and JEV by indirect immunofluorescence assay. Animals were randomly divided into four groups and inoculated subcutaneously (s.c.) in the deltoid region of left arm with 0.5 ml of DENV-2 E217A dilutions containing 5.0, 4.0, 3.0 log10 PFU, respectively. Animals in the control group received PBS. Blood was collected from each RM daily post immunization for 10 days to detect viremia. For neutralizing antibody tests, blood was taken before immunization (day −1) and on days 15 and 30 post-immunization. On day 64 post-immunization, all immunized animals including the PBS-treated control animals were challenged by s.c. inoculation of 0.5 ml containing 5 log10 PFU of DENV-2 (TSV-01). For the following 9 days, blood was collected for determination of viremia. Neutralizing antibody levels in serum were measured by the standard 50% plaque reduction neutralization test (PRNT50) on days 15 and 30 post-challenge, respectively.
Viremia in serum samples was determined by plaque assay in BHK-21 cell monolayers in 12-well plates. Undiluted serum or serial 10-fold dilutions of serum were inoculated onto BHK cells. After 1 h of adsorption at 37°C, wells were overlaid with 1 ml of DMEM supplemented with 2% FBS and 1% agarose. Plates were incubated for 4 days at 37°C in 5% CO2. Monolayers were fixed by addition of 1 ml of 4% formalin solution to the overlay medium. After 1 h of fixation at room temperature, the fixative was removed, wells were washed with water, and monolayers were stained with 1% crystal violet in 70% methanol. Plaques were counted, and titers were expressed as PFU/ml.
For determination of dengue virus-neutralizing antibody titers, serial twofold dilutions of serum (starting at a dilution of 1∶10) were mixed with equal volumes of a suspension of ∼500 PFU of DENV-2 TSV01/ml. The serum-virus mixtures were incubated at 37°C for 1 h and tested (0.2 ml/well) for concentration of infectious virus using the plaque assay described above. The neutralization titer was defined as the lowest serum dilution at which the infectious virus concentration was reduced by 50% from the concentration found when virus was incubated with culture medium.
Cells were seeded at 1×105 per well in a 24-well plate and treated 24 hours prior to infection with medium or varying concentrations of human recombinant IFN-β (Immunotools). Cells were then infected at an MOI of 1 with WT or MTase mutant virus (TSV01), respectively, incubated for 72 hours and harvested and processed for flow cytometry as described. Supernatants were collected for plaque assay.
IFIT expression plasmids were a kind gift from A. Pichlmair (14). For IFIT overexpression cells were seeded at 1×106 per well in a 6-well plate. 24 hours later cells were transfected using 293fectin according to manufacturer's protocol. One day post transfection cells were trypsinized and seeded in a 24-well plate at 1×105 per well. After 24 hours of incubation cells were infected and analysed as described previously. Transfection rate was 30–50% judged by parallel experiments with GFP expression plasmid.
To determine the percentage of infected cells, cells were harvested, washed in PBS and fixed and permeabilized with Cytofix/Cytoperm (BD). Intracellular dengue E protein was stained with antibody 4G2 conjugated to Alexa 647 and fluorescent cells were measured by flow cytometry.
For the assessment of ADE, 4G2 or serum/plasma was serially diluted and a constant amount of virus was added. The antibody-virus mixture was incubated at 37°C for 30 min and then 50 µl of the mixture was added to 25'000 K562 cells per 96-plate well (MOI 0.5–1). After 2 h of infection 150 µl RPMI medium containing FCS was added. After 2.5 days of incubation the infected cells were fixed and stained intracellularly with 4G2-Alexa 647. The percentage of infected cells was quantified by flow cytometry.
For the measurement of neutralization, 4G2 or heat-inactivated serum/plasma was serially diluted and a constant amount of virus was added. The antibody-virus mixture was incubated at 37°C for 30 min and then 50 µl of the mixture was added to 200,000 U937 cells (ATCC) stably transfected with human DC-SIGN (MOI 0.1–1). After 2 h of infection 150 µl RPMI medium containing FCS was added. After incubation over night the infected cells were fixed and stained intracellularly with 4G2-Alexa 647. The percentage of infected cells was quantified by flow cytometry and data were analyzed with GraphPad Prism software for the calculation of the NT50.
Spleens were harvested at day 7 after infection and single cell suspensions were incubated with live virus or a pool of the following peptides: NS4B59-66, NS4B99-107 and NS5237-245 [32] overnight. Brefeldin (Biolegend) was added for 5 h before cells were washed and stained with antibodies CD4-APC, CD8-PercPCy5.5 and IFN-γ-PE-Cy7 (Biolegend). Cells were acquired on a FACSCantoII (BectonDickinson) and data were analyzed with FlowJo (Treestar ltd.)
96-well polystyrene plates were coated with concentrated, heat inactivated dengue virus. Plates were incubated overnight at 4°. Before use, plates were washed three times in PBS (pH 7.2) containing 0.05% Tween-20 (PBS-T). Non-specific binding was blocked with 2% non-fat dry milk diluted in PBS (PBS-M) for 2 h at room temperature (RT). After washing, sera were diluted 1∶50 in PBS-M, heat inactivated for 1 hour at 55°C and threefold serial dilutions were added to the wells. Plates were incubated for 1 h at RT, followed by three washes with PBS-T. Peroxidase-conjugated rabbit anti-mouse IgG, in PBS-M was added, followed by 1 h of incubation at RT and three additional washes with PBS-T. TMB was used as the enzyme substrate. The reaction was stopped with 1 M HCl and the optical densities were read at 450 nm using an automatic ELISA plate reader. Endpoint titers were defined as the lowest dilution of plasma in which binding was twofold greater than the mean binding observed with the negative controls.
Vector competence experiments were performed using a colony of Ae. aegypti mosquitoes in which 10% of the population is derived from field obtained eggs each month. Batches of 50–75 female mosquitoes, aged 5–7 days were fed with pig blood containing WT or MTase mutant DENV-2 at titers of 5, 4, and 3 log 10 PFU/ml. Fully engorged mosquitoes were held at 27°C, 80% relative humidity, and 12 h photoperiod for 15 days, after which the abdomen was separated from the thorax and homogenized. Homogenates were inoculated into Vero cell culture. After culturing the inoculated cells for 5 days, viral infection was assayed using an indirect fluorescent antibody test (IFA). Antibody 6B6C-1 against flavivirus group E protein (at 1∶10 dilution; provided by the USA CDC as a mouse hybridoma) and an anti-mouse antibody conjugated with FITC were used as a primary and secondary antibody, respectively. Positive fluorescence determinations were performed manually using an inverted fluorescent microscope (Olympus IX71). Chi-square and contingency table statistical tests were performed to detail heterogeneity in vector competence within/between WT and mutant viruses.
Intra-thoracic inoculation of 0.17 µl of WT DENV-2 and E217A at a titer of 105 PFU/ml was performed using 10 female mosquitoes each. Following inoculation, mosquitoes were held for seven days under the same conditions as described above. Mosquitoes were then killed by freezing and homogenized. Viral RNA was quantified by real-time qRT-PCR using primers and methods reported previously [57]. Briefly, whole mosquito homogenate viral RNA was extracted using QIAamp Viral RNA Mini Kit (Qiagen). qRT-PCR was completed using Invitrogen SuperScript III Platinum One-Step qRT-PCR mix (without ROX) and CFX96 Real-Time PCR Detection System (BioRad). Cycling parameters performed were 50°C for 30 min, 95°C for 2 min, followed by 45 cycles of 95°C for 10 sec, 60°C for 30 sec. A two-tailed unpaired t-test was performed to determine the statistic difference between the mean genomic equivalents calculated for WT and mutant viruses.
Virus was isolated from mouse serum with Qiagen Viral RNA extraction Kit.
Fifty ng of viral RNA were used to prepare cDNA libraries using the Illumina TruSeq RNA sample preparation kit according to manufacturer's protocol. The only protocol modification was the removal of the mRNA enrichment step. The cDNA libraries were sequenced as a multiplex in a single lane of an Illumina HiSeq2000 (Next Generation Sequencing Core facility, Genomic Institute of Singapore). One to 2 million 50 bp paired-end reads were generated for each virus.
Wild-type and mutant virus samples were mapped to their respective reference genomes using Bowtie 2 [58]. Mapping statistics and genotype calls were made with SAMtools [59]. Data analysis was performed in Pipeline Pilot (http://www.accelrys.com). At least two reads with an alternate base at a given position were defined as a SNP.
Statisitical tests were performed with GraphPad Prism software, using students t test, two-way ANOVA or Chi-square and contingency table statistical tests as indicated in the figure legends.
|
10.1371/journal.ppat.1005038 | Tumor Progression Locus 2 Promotes Induction of IFNλ, Interferon Stimulated Genes and Antigen-Specific CD8+ T Cell Responses and Protects against Influenza Virus | Mitogen-activated protein kinase (MAP) cascades are important in antiviral immunity through their regulation of interferon (IFN) production as well as virus replication. Although the serine-threonine MAP kinase tumor progression locus 2 (Tpl2/MAP3K8) has been implicated as a key regulator of Type I (IFNα/β) and Type II (IFNγ) IFNs, remarkably little is known about how Tpl2 might contribute to host defense against viruses. Herein, we investigated the role of Tpl2 in antiviral immune responses against influenza virus. We demonstrate that Tpl2 is an integral component of multiple virus sensing pathways, differentially regulating the induction of IFNα/β and IFNλ in a cell-type specific manner. Although Tpl2 is important in the regulation of both IFNα/β and IFNλ, only IFNλ required Tpl2 for its induction during influenza virus infection both in vitro and in vivo. Further studies revealed an unanticipated function for Tpl2 in transducing Type I IFN signals and promoting expression of interferon-stimulated genes (ISGs). Importantly, Tpl2 signaling in nonhematopoietic cells is necessary to limit early virus replication. In addition to early innate alterations, impaired expansion of virus-specific CD8+ T cells accompanied delayed viral clearance in Tpl2-/- mice at late time points. Consistent with its critical role in facilitating both innate and adaptive antiviral responses, Tpl2 is required for restricting morbidity and mortality associated with influenza virus infection. Collectively, these findings establish an essential role for Tpl2 in antiviral host defense mechanisms.
| Influenza viruses infect millions of people annually causing significant morbidity, mortality and socio-economic burdens. Host immune responses against influenza virus are initiated upon virus recognition by specific intracellular receptors. Signals relayed from these receptors trigger various signaling cascades, which induce an antiviral immune response to control infection. Herein, we identified the serine-threonine kinase tumor progression locus 2 (Tpl2) as an essential component of virus sensing pathways, regulating induction of interferons (IFNs) and IFN-induced antiviral genes that restrict virus replication. We also demonstrate that Tpl2 is necessary for generation of effector CD8+ T cells, which are required for viral clearance from infected lungs. Consistent with the impaired antiviral responses, Tpl2-deficient mice are defective in controlling virus replication and succumb to influenza virus infection with a normally low pathogenicity strain. Thus, our study identifies Tpl2 as a host factor that integrates antiviral innate and adaptive responses to restrict morbidity and mortality during influenza virus infection.
| Mitogen-activated protein kinase (MAP) cascades represent major intracellular signaling pathways activated in response to a variety of external stimuli. Their activation during infection leads to transcriptional induction of immune and inflammatory mediators. Although MAP kinase signaling is important in eliciting host protective responses, many viruses are known to utilize these pathways directly for their replication [1]. Activation of MAP kinases occurs during virus recognition by pattern recognition receptors (PRRs) like toll-like receptors (TLRs) and RIG-I-like RNA helicases (RLH) [2]. Virus sensing by these receptors activates multiple intracellular signaling cascades including NFκB, MAP kinase and IRF pathways that coordinately regulate induction of interferons (IFNs) which are important mediators of antiviral resistance [3]. Among the MAP kinases, tumor progression locus 2 (Tpl2/MAP3K8), a MAP3 kinase, plays an important role in regulating IFN production by promoting the ERK-dependent induction of c-fos, a component of AP-1 heterodimeric transcription factors [4]. While Tpl2 is required for IFNα production by plasmacytoid dendritic cells (pDCs) and IFNγ secretion by CD4+ T cells, it is a potent negative regulator of IFNβ in macrophages and DCs [4, 5]. Despite being identified as a major regulator of both Type I (IFNα/β) and Type II (IFNγ) IFNs, Tpl2 regulation of Type III IFNs (IFNλs) has not been investigated so far.
Tpl2 was initially identified as an oncogene that induces T cell lymphomas in rodents [6], but more recent studies have established its criticality in regulating both innate and adaptive immune responses via its cell type- and stimulus-specific activation of the MEK-ERK MAPK pathway. Tpl2 regulates signal transduction and cellular responses downstream of TLRs, cytokine receptors, antigen receptors and G protein-coupled receptors [4, 7–9]. In addition to IFNs, Tpl2 also regulates the production of other prominent immune mediators like TNFα, IL-1β IL-10, IL-12 and COX-2 [4, 10–12]. Consequently, Tpl2 is essential for mounting effective immune responses during infections, and Tpl2-/- mice are more susceptible to Toxoplasma gondii [5], Listeria monocytogenes [11], Mycobacterium tuberculosis [13] and Group B Streptococcus [14]. Surprisingly, there is still limited and contradictory information about how Tpl2 contributes to host defense against viruses. Early studies reported normal cytotoxic T cell responses against lymphocytic choriomeningitis virus [10] and resistance to mouse cytomegalovirus infection [14]. However, another study delineating the signaling circuitry in virus sensing pathways implicated Tpl2 as a key regulator of both inflammatory and antiviral gene induction in response to model viral ligands [15]. A recent study also reported increased replication of vesicular stomatitis virus in Tpl2-deficient mouse embryonic fibroblasts (MEFs) [16].
We recently demonstrated that among the TLRs implicated in virus sensing (TLRs 3, 7 and 9), Tpl2 plays a prominent role in TLR7 signaling [17]. In this study, we investigated Tpl2’s regulation of antiviral responses using a murine model of influenza virus infection, which relies upon TLR7 for virus sensing [18], ERK MAP kinase for virus replication [19] and where both IFNα/β and IFNλ are host protective [20]. Our experiments demonstrate positive regulation of IFNλ and cell-type specific regulation of IFNα/β production in Tpl2-deficient cells following stimulation with model viral ligands that trigger influenza virus sensing receptors, TLR7 or RIG-I. However, during influenza virus infection, IFNλ uniquely required Tpl2 for its induction. Moreover, Tpl2 is involved in IFN signaling, regulating ERK activation and STAT1ser727 phosphorylation, and is required for proper induction of antiviral IFN-stimulated genes (ISGs). Impaired ISG induction coupled with reduced antigen-specific CD8+ T cells resulted in failure to control virus replication and significant morbidity and mortality of Tpl2-/- mice to an otherwise low pathogenicity strain of influenza virus. Collectively, this study establishes Tpl2 as a host factor that integrates antiviral responses to control influenza virus infection.
To determine whether Tpl2 regulates influenza virus replication, wild type (WT) and Tpl2-/- mice were infected with 104 plaque forming units (pfu) of mouse-adapted influenza virus A/HK-X31(H3N2) (X31), and viral titers in the lungs were evaluated on days 3, 5 and 7 post infection (pi). The average lung viral titers were significantly higher in Tpl2-/- mice compared to WT mice at all time points examined (Fig 1A). Notably, average viral titers were more than ten-fold higher in Tpl2-/- lungs at day 7 pi. This increase in virus replication was also observed in littermate control mice (S1 Fig). In addition to viral titers, early proinflammatory cytokines, except TNFα were significantly higher in the BALF of Tpl2-/- mice compared to WT mice (Fig 1B). Consistent with increased virus replication, total cellular infiltration was also significantly increased in the lungs of Tpl2-/- mice at day 7 pi (Fig 1C). The increased lung viral titers in Tpl2-/- mice early after infection on day 3 suggest a critical role for Tpl2 in limiting virus replication during influenza virus infection.
Airway epithelial cells are the primary targets for influenza virus infection. Early studies after the discovery of Tpl2 demonstrated high levels of Tpl2 expression in the lungs [21]. Moreover, similar to hematopoietic cells, Tpl2 regulation of signal transduction and cytokine gene induction was also demonstrated in airway epithelial cells [22]. To elucidate whether Tpl2 functions in hematopoietic or nonhematopoietic cells to limit virus replication, we assessed lung viral titers in chimeric mice in which WT or Tpl2-/- bone marrow cells were transferred into either WT or Tpl2-/- irradiated recipients. At day 3 pi, average lung viral titers were significantly higher in Tpl2-/- mice reconstituted with WT hematopoietic cells (Fig 1D). In contrast, there was no statistically significant increase in viral titers of WT mice that received Tpl2-/- bone marrow (Fig 1E). These data demonstrate that Tpl2 signaling within radioresistant, nonhematopoietic lung cells is necessary for limiting virus replication early after infection.
Interferons are induced early during infection and are key factors initiating host protective antiviral responses [3]. To determine whether the observed increase in viral titers in Tpl2-/- mice is due to defective induction of IFNs, WT and Tpl2-/- mice were infected with 106 pfu X31 virus, and IFNα/β/λ levels in lung homogenate or BALF were measured at day 1 or day 3 pi. Induction of both IFNα and β were comparable between WT and Tpl2-/- lung homogenates and BALF (Fig 2A). Notably, IFNλ secretion was significantly reduced in Tpl2-/- mice following influenza virus infection (Fig 2B). Surprisingly, while IFNλ was induced to a higher level compared to Type I IFNs in WT mice, there was minimal induction in Tpl2-/- mice in response to infection at both time points. Reduced IFNλ production in Tpl2-/- mice was independent of viral titers which were similar between WT and Tpl2-/- mice at day 1 pi (S2 Fig). Despite differences in IFNλ induction, total cellular infiltration and IFNγ levels in BALF were significantly elevated in Tpl2-/- mice compared to WT mice at day 3 pi (S3 Fig).
The observation that Tpl2 is uniquely required for IFNλ, but not IFNα or IFNβ, production in influenza-infected lungs is especially significant, because IFNλ is regarded as the principal IFN induced during influenza virus infection. Airway epithelial cells and pDCs are considered the major sources of IFNs during respiratory virus infections, including influenza [20, 23]. Although we observed a decrease in IFNλ levels in Tpl2-/- mice at day 1 pi, a more consistent and significant reduction was observed at day 3 pi, which corresponds to the migration of pDCs to infected lungs [23]. Since Tpl2 is required for macrophage and neutrophil migration during acute inflammation [9, 24], we investigated whether Tpl2 similarly regulates the recruitment of pDCs to the infected lung. The reduction in IFNλ levels in influenza-infected Tpl2-/- mice was not due to impaired recruitment of pDCs (S4 Fig). To investigate whether defective IFN induction by pDCs contributes to the reduced IFNλ in BALF from Tpl2-/- mice during influenza infection, bone marrow-derived pDCs (CD11c+B220+CD11b-) from WT and Tpl2-/- mice were infected with influenza virus A/WSN/1933 (H1N1), and the production of IFNα, β and λ was assessed. Consistent with in vivo infections, the levels of both IFNα and IFNβ were comparable between WT and Tpl2-/- cells, whereas IFNλ secretion was significantly less in Tpl2-/- pDCs infected with influenza virus (Fig 2C). A similar reduction in IFNλ induction was also observed in Tpl2-deficient cells infected with X31 influenza virus strain (S5 Fig). Collectively, these data demonstrate the unique requirement for Tpl2 in IFNλ production during influenza infection in vitro and in vivo.
During influenza virus infection, receptors from both TLR and RLR families recognize viral PAMPs and trigger rapid induction of IFNs. Recognition of viral components by PRRs typically occurs in respiratory epithelial cells, alveolar macrophages, DCs and pDCs in a cell type-specific manner [25]. The major receptors involved in recognition of influenza virus are TLR7, which recognizes single-stranded viral RNA, and RIG-I, which recognizes the 5’-triphosphate of single-stranded RNA genomes (5’ppp-RNA). The single-stranded RNA genome is recognized through endosomal TLR7 in pDCs [18] in contrast to epithelial cells and DCs where virus recognition is mediated primarily by the cytosolic sensor RIG-I [26]. We therefore investigated whether differential regulation of IFN production observed during infection is due to differences in Tpl2-mediated sensing by PRRs. MEFs and bone marrow-derived macrophages (BMDMs) from WT and Tpl2-/- mice were either transfected with the RIG-I ligand 5’ppp-RNA or stimulated with the TLR7 ligand R848 [27], and IFNβ production was measured by ELISA. Consistent with previous studies using the TLR4 ligand LPS [4], IFNβ production was significantly increased in Tpl2-/- cells treated with both 5’ppp-RNA and R848 (Fig 3A–3C). This increase in IFNβ correlated with impaired ERK phosphorylation in Tpl2-deficient cells in response to these ligands (S6 Fig). Unlike epithelial cells and DCs, virus recognition in pDCs is mediated via TLRs rather than RLHs, and Type I IFN production occurred normally in RIG-I-deficient pDCs infected with RNA viruses [18, 26]. To determine whether Tpl2 regulates TLR7-mediated IFN production by pDCs, bone marrow-derived pDCs from WT and Tpl2-/- mice were treated with the TLR7 ligand, R848, and IFN levels were quantitated. Consistent with previous studies using the TLR9 ligand CpG [4], and in contrast to BMDMs, secretion of both IFNα and IFNβ were significantly decreased in culture supernatants from Tpl2-/- pDCs treated with R848 (Fig 3D). Notably, IFNλ secretion was also significantly less in Tpl2-/- pDCs compared to WT cells in response to R848 (Fig 3D). Unlike Ifna but similar to NFκB-regulated Il12p40 and Tnfa [28], IFNλ3 (Il28b) transcription occurred early, by 2 hr of stimulation (S7 Fig). Collectively, these data demonstrate that Tpl2 differentially regulates IFN production downstream of PRRs involved in influenza virus sensing in a cell type-specific manner.
The importance of IFNλs in host protection against many viruses is well established, however, the mechanisms that regulate their production are largely unexplored. Common mechanisms have been postulated to regulate Type I and III IFNs during viral infections [29, 30]. Despite their importance in mediating Type I IFN production in pDCs [4, 31], the significance of MAP kinase and PI3 kinase cascades in murine IFNλ production has not been directly investigated. In order to elucidate the potential mechanism by which Tpl2 regulates IFNλ production in pDCs, we evaluated the involvement of ERK and PI3K-mTOR signaling in IFNλ induction. Tpl2 regulation of both ERK and mTOR-Akt signaling in different cell types has been reported previously [8, 32–34]. In addition to the MEK/ERK pathway [4], we demonstrate that Tpl2 also promotes mTOR/Akt signaling in pDCs as determined by a decrease in the proportion of phospho-Akt+ pDCs in the absence of Tpl2 signaling (Fig 4A and 4B). To confirm whether ERK, PI3K or mTOR signaling also contributes to IFNλ production in pDCs, cells were pre-treated with rapamycin (mTOR inhibitor), LY294002 (PI3K inhibitor) or U0126 (MEK inhibitor) 30 min prior to TLR stimulation, and CpG-induced IFNλ secretion was measured by ELISA. CpG was used as the stimulant in these experiments because TLR9 ligation induced higher levels of IFNλ compared to TLR7 stimulation with R848. Pharmacological inhibition of each of these signaling pathways significantly reduced IFNλ secretion to the levels observed in Tpl2-/- cells (Fig 4C). In contrast, only a modest reduction in IFNλ induction was observed in Tpl2-deficient cells treated with rapamycin or U0126 (S8 Fig). These results demonstrate the significance of Tpl2 and both MAPK and PI3 kinase signaling cascades in regulating IFNλ production in pDCs.
Robust production of Type I IFNs in pDCs is dependent upon IRF7 and autocrine IFN signaling, and consequently IFNα secretion is abrogated in both Irf7-/- and Ifnar1-/- pDCs [35]. Similar to IFNα, and as reported previously [20], IFNλ production was abolished in Ifnar1-/- pDCs infected with influenza virus (Fig 5A) demonstrating the absolute requirement for IFNAR signaling in IFNλ secretion by pDCs. Induction of IFNλ in response to direct IFN stimulation has been reported in hepatocyte carcinoma HepG2 cell lines [36]. Although a high dose of IFNβ could induce modest IFNλ secretion, the levels induced were lower than that induced by TLR-stimulation, demonstrating that IFN/IRF7 signaling alone is not sufficient for driving high levels of IFNλ secretion (Fig 5B). Nevertheless, Tpl2 contributed to IFNAR-induced IFNλ production, since significantly less IFNλ was secreted by Tpl2-/- pDCs directly treated with IFNβ (Fig 5B). In addition to demonstrating the role of Tpl2 in IFNAR-mediated IFNλ production, these data also suggest a role for Tpl2 in directly transducing Type I IFN signals.
To determine whether Tpl2 regulates IFNλ production in influenza virus-infected lungs directly via virus sensing pathways or indirectly via IFNAR feedback signaling, we assessed IFNλ levels in lung homogenates from mice that are deficient in both Tpl2 and IFNAR1. Consistent with reduced IFNλ levels in BALF from Tpl2-/- mice day 3 pi (Fig 2B), IFNλ levels were similarly reduced in day 3 lung homogenates (Fig 5C). IFNλ levels were significantly decreased in Ifnar1-/-Tpl2-/- compared to Ifnar1-/- mice, demonstrating that Tpl2 promotes early IFNλ induction independent of Type I IFN signaling (Fig 5D). Notably, the level of IFNλ induction was similar in Tpl2-/- and Ifnar1-/-Tpl2-/- mice (Fig 5C and 5D). In striking contrast to the abrogation of IFNλ production in Ifnar1-/- pDCs (Fig 5A), IFNλ production occurred normally in Ifnar1-/- mice (Fig 5D). Consistent with the critical role of IFNAR signaling in IFNα induction, IFNα levels were significantly diminished in both Ifnar1-/- and Ifnar1-/-Tpl2-/- mice (Fig 5E). These data demonstrate that Tpl2-dependent IFNλ production during influenza virus infection is IFNAR-independent.
Both IFNα/β and IFNλ are known to induce expression of ISGs that establish an antiviral state in infected tissue to prevent virus replication and spread [3, 37]. Because of the observed increase in early virus replication in Tpl2-/- mice (Fig 1A), we questioned whether Tpl2 regulates the induction of ISGs. We first addressed whether Tpl2 regulates IFN signaling. BMDMs from WT and Tpl2-/- mice were stimulated with IFNα or IFNβ, and activation of downstream cascades, especially STAT1, which is the principle regulator of IFN responses, was evaluated by immunoblotting. BMDMs were used in these experiments due to limited availability of pDCs. The phosphorylation of STAT1 Tyr701 and Ser727, which is necessary for maximal STAT1 transcriptional activation, were examined [38]. While phosphorylation of Tyr701 occurred normally in Tpl2-deficient cells in response to stimulation with Type I IFNs, a consistent reduction in Ser727 phosphorylation was observed in Tpl2-/- cells compared to WT cells (Fig 6A–6C). In addition to the classical JAK-STAT pathway, signaling via the Type I IFN receptor also activates other downstream cascades including MAP kinases [39]. Despite the existence of multiple MAP3 kinases, Tpl2 has an essential, non-redundant role in transducing ERK activation signals during TLR, TNF- and IL-1-receptor signaling [7, 8]. We therefore investigated whether Tpl2 is similarly required for ERK activation during Type I IFN signaling, or whether other MAP3Ks could fulfill this role. ERK phosphorylation was strongly induced by both IFNα and IFNβ. Importantly, ERK phosphorylation was absent in Tpl2-/- BMDMs stimulated with IFNα/β demonstrating an absolute requirement for Tpl2 in transducing ERK activation signals in response to Type I IFNs (Fig 6A). Of note, unlike LPS- and TNFα-treated BMDMs and similar to poly I:C-, CpG-, and IL-1β-treated BMDMs [17, 40], no mobility shift (indicative of phosphorylation) or degradation of the p58 isoform of Tpl2 was detected following stimulation with Type I IFNs (Fig 6A). Consistent with our previous studies [41], both Tpl2 protein and mRNA expression were induced upon either Type I IFN stimulation or influenza virus infection (Fig 6A and S9 Fig). Overall, these data demonstrate that Tpl2 contributes to Type I IFN signaling.
Since Tpl2 is known to modulate the antiviral transcriptome [16], we next investigated whether the induction of ISGs in infected lungs is impaired in the absence of Tpl2. The induction of Ifitm3, Isg15 and Oasl2, ISGs known to be important in limiting influenza virus infection [25], were measured. We observed a modest, but statistically significant decrease in Ifitm3 and Oasl2 expression in Tpl2-/- compared to WT mice infected with influenza virus (Fig 6D). A trend towards reduction in Isg15 expression was also noted in Tpl2-/- mice (Fig 6D). In addition to infected lungs, the induction of Oasl2 was significantly reduced in Tpl2-/- BMDMs, while the expression of Ifitm3 and Isg15 was largely unaffected by Tpl2 ablation (S10 Fig). These data demonstrate that Tpl2 promotes the induction of ISGs in influenza-infected lungs to limit virus replication.
Although Tpl2 is important in transducing Type I IFN signals, this would not alone account for the increase in viral titers or reduction in ISGs observed in Tpl2-/- mice, since either Type I or Type III IFN is sufficient for protection during influenza virus infection. This is because both types of IFNs drive redundant amplification loops inducing the expression of similar antiviral genes [42]. To investigate whether IFNAR signaling contributes to the observed increase in virus replication, we next assessed lung viral titers in mice deficient in both Tpl2 and IFNAR1. Consistent with previous studies [20], viral titers were comparable between WT and Ifnar1-/- mice (Fig 6E). Although average lung viral titers were significantly higher in Ifnar1-/-Tpl2-/- mice compared to both WT and Ifnar1-/- mice (Fig 6E), the titers were similar to those observed in Tpl2-/- mice (Fig 1A). These data demonstrate that Tpl2 restricts early virus replication in an IFNAR-independent manner.
Even though the observed reduction in ISGs helps to explain the early increase in viral titers, a more pronounced and biologically significant increase in viral titers was observed at day 7 pi which correlates with the recruitment of influenza-specific CD8+ T cells to the lungs [43]. Since many seminal studies have identified CD8+ T cells as the major mediators of influenza virus clearance from infected lungs [44, 45], we investigated whether virus-specific CD8+ T cell responses are impaired in Tpl2-/- mice. Consistent with defective viral clearance observed in Tpl2-/- mice, induction of protective nucleoprotein (NP)-specific CD8+ T cells [46] was significantly reduced in BAL cells from Tpl2-/- mice compared to WT animals (Fig 7A and 7B). In addition, antigen-specific secretion of IFNγ was also decreased in BAL cells from Tpl2-/- mice (Fig 7C). During the course of this experiment, we unexpectedly observed severe clinical signs in Tpl2-/- mice despite the fact that the mice were infected with the low pathogenicity A/HK-X31(H3N2) (X31) influenza virus. To confirm whether Tpl2 ablation alters the susceptibility to influenza virus infection, WT and Tpl2-/- mice were infected with 104 pfu of X31 virus, and weight loss and clinical symptoms were monitored over a period of 14 days. All Tpl2-/- mice exhibited severe clinical signs and succumbed to infection by day 10 pi, whereas all WT animals survived and returned to pre-infection body weights by day 14 pi (Fig 7D and 7E). Similar to infection with X31 virus, Tpl2-/- mice infected with the virulent PR8 [A/Puerto Rico/8/34 (PR8; H1N1)] strain showed increased disease severity compared to WT mice, although not to the same extent seen with the low pathogenicity virus (S11 Fig). Body weights were collected to day 10 pi, at which time the Tpl2-/- mice met the humane endpoints of the study. At this time point, the body weights were just beginning to show the characteristic switch between the WT and Tpl2-/- mice, such that the Tpl2-/- mice were showing more severe clinical signs of disease. Accordingly, systemic pro-inflammatory cytokine levels were also increased in the Tpl2-/- mice at day 10 pi. Analysis of BAL cells also showed decreased antigen-specific CD8+ T cell responses in Tpl2-/- mice compared to WT mice at this late time point, consistent with the observations with X31 infections. Collectively, these data demonstrate the critical role of Tpl2 in promoting viral clearance and restricting morbidity and mortality associated with influenza virus infection.
Tpl2 is now appreciated to regulate the induction of Type I and Type II IFNs as well as other cytokines that may contribute to antiviral responses. However, there is very limited information on how Tpl2 coordinates antiviral immune responses in vivo. In this study, we demonstrate Tpl2’s obligate role in promoting antiviral responses and viral clearance during influenza virus infection. These findings are important because influenza virus is a ubiquitous seasonal virus that afflicts millions of people annually, causing significant morbidity, mortality and socio-economic burdens [47]. Therefore, understanding the role of host factors like Tpl2 in restricting morbidity and mortality associated with influenza virus infection is critical for developing disease intervention strategies. Mechanistically, Tpl2 promotes the induction of ISGs and virus-specific CD8+ T cells that facilitate viral clearance as shown in the proposed model (Fig 8). Thus, the findings reported here establish an essential role for Tpl2 in host protective innate and adaptive antiviral responses.
Tpl2 deficiency led to cell-type specific alterations in the regulation of Type I IFN production. Specifically, IFNβ production was increased in response to LPS, R848 and the RIG-I ligand, 5’-triphosphate RNA in Tpl2-/- MEFs and BMDMs. In contrast, Type I IFN was significantly reduced in pDCs in response to TLR7 stimulation with R848. This differential regulation of Type I IFN production by Tpl2 in different cell types in response to TLR ligands is consistent with a previous report by O’Garra and colleagues [4]. Importantly, we also demonstrated that Tpl2 similarly functions as a negative regulator of Type I IFN production upon activation of the RIG-I cytosolic sensor with 5’-triphosphate RNA.
One striking observation was the absolute requirement for Tpl2 in the TLR-dependent induction of both Type I (IFNα/β) and Type III IFNs (IFNλ) by pDCs. The fact that pDCs uniquely require Tpl2 for production of both Type I and Type III IFNs suggests that pDCs differ fundamentally from BMDMs and MEFs in their signaling pathways. Indeed, impaired IFN production correlated with reduced activation of the PI3K/Akt signaling pathway in Tpl2-/- pDCs. This finding is also consistent with the observation that the PI3K/Akt pathway appears to be especially important in driving TLR-dependent IFN expression by pDCs [31].
In addition to cell-type specific regulation, Tpl2 also differentially regulates the production of Type I and Type III IFNs during viral infection. Importantly, influenza virus has been reported to utilize the Raf pathway to activate ERK, which explains why Type I IFN induction occurs in a Tpl2-independent manner in mice and pDCs infected with influenza virus [48]. On the contrary, IFNλ production was uniquely dependent upon Tpl2 during the course of influenza infection in vitro and in vivo. Although Type I and Type III IFNs have common regulatory elements in their promoters and are usually co-expressed in response to viruses and TLR ligands [36], selective induction of IFNλ by transcription factors NFκB and IRF1 has been reported [49, 50]. The distinct requirement for Tpl2 in IFNλ induction in virus-infected pDCs likely represents the unique requirement of the IFNλ promoter for an early NFκB-dependent priming event. In support of this, our own data demonstrate that IFNλ induction is rapid and parallels the regulation of NFκB-dependent genes more closely than IFNα (S7 Fig).
With the exception of a recent study reporting that p38, but not ERK, is required for Ifnl1 expression in human cells [49], the roles of MAPK or PI3K pathways in the regulation of IFNλs have not been evaluated. Although the regulation of IFNλ1 by PI3K-mTOR is still unexplored, our data demonstrate a different mechanism of IFNλ3 regulation that relies on the Tpl2-ERK pathway in contrast to the p38-dependent regulation described for IFNλ1. Therefore, in addition to transcription factors [30], diverse signaling cascades also specify induction of different IFNs.
The complexity of the IFN response is not completely understood, since multiple signaling cascades and transcription factors activated during IFN signaling can independently or cooperatively regulate the transcriptional response to IFNs [39]. Importantly, our data demonstrate the involvement of Tpl2 in IFN signaling leading to the phosphorylation of ERK and STAT1Ser727. Previous studies have demonstrated the significance of STAT1Ser727 phosphorylation for full transcriptional activation and induction of ISGs [38, 51]. Conflicting reports exist regarding the identity of the serine kinase responsible for STAT1Ser727 phosphorylation; different kinases including p38, ERK and PKC-δ have been implicated [52–54]. Importantly, an association of ERK with STAT1 and a requirement of ERK activity for expression of ISGs have been demonstrated [55]. Tpl2 regulation of STAT1Ser727 phosphorylation and induction of ISGs might be indirect via its regulation of ERK phosphorylation during IFN signaling. In addition to regulating ISG transcription, Tpl2-ERK signaling also regulates the phosphorylation and dissociation of the translation initiation factor 4E-Bp-eIF4E complex, which is involved in cap-dependent translation of many genes, including ISG15 [34, 56]. Therefore, the Tpl2-ERK pathway regulates the biological effects of IFNs at the transcriptional level and possibly also at the posttranscriptional level.
Although MAP kinase pathways are known to be activated in response to IFNs, the importance of Tpl2 in regulating IFN-inducible effectors has not yet been described. The induction of ISGs is mainly attributed to IFN-stimulated gene factor-3 (ISGF3; consists of STAT1, STAT2 and IRF9). In addition to ISGF3, IRF7 can also act independently to regulate transcription of antiviral genes, and Tpl2 has been shown to promote IRF7-dependent transcription [16]. However, normal induction of IFNα/β during influenza virus infection argues against a major role for IRF7 in the observed phenotype, since IRF7 is regarded as the ‘master regulator’ of Type I IFN induction [35].
To understand the mechanism by which Tpl2 exerts its antiviral effect, we examined the contribution of Tpl2 to virus replication in different cellular compartments and in the context of IFNAR deficiency. Using bone marrow chimeras, we demonstrated that Tpl2 was required within the nonhematopoietic compartment to restrict early virus replication. This likely reflects Tpl2 functions in airway epithelial cells, the primary target of influenza virus. In this regard, Tpl2 is known to be expressed and to regulate inflammation within airway epithelial cells [22]. Studies using Ifnar1-/-Il28ra-/- mice have also demonstrated that interferon responsiveness of these cells is critical for restricting early viral replication [42]. It is well known that abrogation of Type I IFN signaling does not increase influenza virus replication due to the presence of compensatory Type III IFNs [57]. Consistent with this, we observed that Tpl2 ablation promoted virus replication to the same extent on both Ifnar1+/+ and Ifnar1-/- genetic backgrounds. The 50% reduction in IFNλ levels that we observed in Tpl2-/- mice on day 3 pi is unable to explain the increase in virus replication, because compensatory Type I IFNs are induced to normal levels. Furthermore, the presence of IFNs, rather than quantity, is important in driving antiviral responses [42]. One possible explanation for the increased viral replication in Tpl2-/- mice is that Type III IFN signaling is also Tpl2-dependent, like we have demonstrated for Type I IFNs. Additional studies using Il28ra-/- mice are needed to determine the contribution of Tpl2 to Type III IFN signaling.
In addition to antiviral innate responses, we also identified a critical role for Tpl2 in the induction of antigen-specific CD8+ T cell responses. This is in contrast to a recent study reporting a major role for Tpl2 in human, but not murine, CD8+ T cell responses [58]. The impaired induction of virus-specific CD8+ T cells resulting in defective viral clearance and increased mortality in Tpl2-/- mice clearly warrants detailed studies on Tpl2 regulation of effector CD8+ T cell responses.
The increased mortality observed in Tpl2-/- mice infected with X31 virus was surprising because infection with this low pathogenicity virus does not typically cause severe clinical signs or mortality in mice. Even though IFNλ production was impaired in Tpl2-/- mice, this defect is not sufficient to explain their increased morbidity and mortality, because several studies have shown that either Type I or Type III IFN alone is sufficient to limit influenza virus infection [20, 42, 59]. In addition to impaired CD8+ T cell responses [45], the reduction in expression of some ISGs may also contribute to the enhanced pathogenesis, since defective expression of individual antiviral factors, like IFITM3, can alter the course of infection [60]. Early increases in virus replication in Tpl2-deficient lung stromal cells, demonstrated by bone marrow chimera experiments, coupled with defective viral clearance by CD8+ T cells likely potentiate the inflammatory response, which is considered a major factor contributing to morbidity and mortality during pathogenic influenza infection [61].
Overall, our study establishes Tpl2 as a host factor with intrinsic ability to restrict influenza virus replication and also demonstrates immune regulatory functions of Tpl2 within the lungs. The involvement of Tpl2 in major virus sensing pathways as well as antiviral signaling cascades suggests a key role for Tpl2 in integrating antiviral responses. These results are especially significant considering a very recent study demonstrating the requirement of IRF7 as well as Type I and Type III IFNs, all regulated by Tpl2, in protecting humans from life-threatening influenza virus infection [62]. Whether Tpl2 similarly restricts the replication of other classes of viruses requires further investigation. The findings reported here also suggest that therapeutic inhibition of Tpl2 during chronic inflammatory diseases might predispose patients to viral infections.
All animal experiments were performed in accordance to the national guidelines provided by “The Guide for Care and Use of Laboratory Animals” and The University of Georgia Institutional Animal Care and Use Committee (IACUC). The Institutional Animal Care and Use Committee (IACUC) of the University of Georgia approved all animal experiments (Assurance Number A3437-01).
Wild type (WT) C57BL/6J (CD45.2+) mice were purchased from The Jackson Laboratory. Tpl2-/- mice backcrossed to C57B6/J were kindly provided by Dr. Philip Tsichlis (Tufts University) and Thomas Jefferson University. For some experiments, littermate control WT and Tpl2-/- mice were obtained by interbreeding Tpl2+/- mice. Ifnar1-/- mice were kindly provided by Dr. Biao He (University of Georgia). Mice deficient in both IFNAR1 and Tpl2 were generated by interbreeding single knockout animals. To generate chimeric mice, WT or Tpl2-/- recipient mice (both CD45.2+) were lethally irradiated with 1100 rad and reconstituted with donor B6.SJL-PtprcaPepcb/BoyJ (WT CD45.1+ congenic) or Tpl2-/- bone marrow cells. Chimeric mice were maintained for 8 weeks. Animals were housed in sterile microisolator cages in the Central Animal Facility of the College of Veterinary Medicine. Embryonated specific pathogen free (SPF) chicken eggs were purchased from Sunrise Farms, New York. Influenza viruses A/HKX31 (H3N2), A/Puerto Rico/8/34 (PR8; H1N1) and A/WSN/1933 (H1N1) stocks were propagated in the allantoic cavity of 9- to 11-day-old embryonated SPF chicken eggs at 37°C for 72 hr, and viral titers were enumerated by plaque assays [63].
Age-matched, 6- to 8-week-old WT, Tpl2-/-, Ifnar1-/-, Ifnar1-/-Tpl2-/- or chimeric mice were anesthetized with 250 mg/kg Avertin (2,2,2-tribromoethanol) followed by intranasal infection with influenza A/HK-X31 (H3N2) in 50 μl PBS. Control mice were mock-infected with a similar dilution of allantoic fluid. To determine lung viral titers, whole lungs from WT and Tpl2-/- mice infected with 104 pfu of X31 virus were harvested on days 3, 5 and 7 pi. Lungs were placed in 1 ml PBS and dissociated with a bead mill homogenizer (Qiagen), and virus titers were enumerated by plaque assays. To assess susceptibility to influenza infection, WT and Tpl2-/- mice infected with 104 pfu of X31 virus were observed over a period of 14 days. Body weights were recorded daily, and mice exhibiting severe signs of disease or more than 30% weight loss were euthanized. To measure IFN and cytokine secretion, mice infected with 106 or 104 pfu of X31 virus were euthanized 3 or 7 days pi, and bronchoalveolar lavage fluid (BALF) or BAL cells were obtained by washing the lungs twice with 1 mL PBS. Cells were recovered by centrifugation of the lavage fluid for 10 min at 250xg. BALF from the first wash was used for quantitation of cytokine secretion. Cellular recruitment was assessed by quantifying total leukocyte recovery from both washes.
Mice infected with 104 pfu of X31 virus were euthanized on day 10 pi, and cells were obtained by washing the lungs twice with 1 mL PBS. BAL cells were stained with anti-CD4, anti-CD8 (eBiosciences), and H2DbNP366–374 tetramer (NIH Tetramer Core Facility, Emory University, Atlanta, GA) for 30 min at 4°C and fixed in 1% formaldehyde. Cells were acquired on a BD LSRII flow cytometer and analyzed using FlowJo software (Tree Star, Inc.). For IFNγ measurement, BAL cells were stimulated with a cocktail of influenza immunodominant peptides (NP366–374, PA224–233, PB1703–711) (1 μg/mL) for 24 hr at 37°C, and IFNγ levels in culture supernatant was measured by ELISA (eBiosciences).
Bone marrow derived macrophages (BMDMs), pDCs and mouse embryonic fibroblasts (MEFs) were generated from age- and sex-matched mice as described previously [17, 64]. CD11c+CD11b-B220+ pDCs were sorted using a Beckman Coulter MoFlo XDP cell sorter. In some experiments, cells were used on day 10 without sorting (referred as Flt3 ligand-derived DCs). Triggering of RIG-I was accomplished by directly delivering 5’-triphosphate RNA (5’ppp-RNA; 0.5 μg/mL) or a control RNA to the cytosol of BMDMs or MEFs using LyoVec transfection reagent (InvivoGen). 20 μL 5’ppp RNA or control RNA (100 μg/mL) was incubated with 200 μL LyoVec (62.5 μg/mL) at room temperature for 15 min to form complexes. Twenty-five microliters of the complexes were used to stimulate 2.5x105 BMDMs or 0.5x105 MEFs per well for 24 hr. BMDMs at 1x106/mL were also treated with R848 (InvivoGen) (1 μg/mL) for 24 hr. To investigate IFN signaling, BMDMs at 1x106/mL were treated with rmIFNα (2000 IU/mL; R&D Systems), or rhIFNβ (10 ng/mL; Peprotech) for 1–4 hr.
Plasmacytoid DCs at a concentration of 0.5-1x106/mL were left untreated or stimulated with R848 (1 μg/mL), CpG ODN2395 (10 μg/mL) (InvivoGen), 50 ng/mL rhIFNβ (PeproTech) or infected with WSN virus at a MOI of 0.2 for 24 hr. In some experiments, cells were pretreated with LY294002 hydrochloride (20 μM), rapamycin (30 nM) or U0126 (20 μM) (Sigma) for 30 min before stimulating with CpG.
Cytokine levels were measured by ELISA (IFNα, IFNλ and IFNγ, eBioscience; IFNβ, PBL Interferon Source) or bead-based detection assays (Mouse IFNα Flowcytomix simplex, eBioscience; Mouse inflammation cytokine bead array, BD Biosciences).
Cells stimulated with R848 or IFNs were lysed using TRK lysis buffer (Omega Bio-Tek). For in vivo infections, RNA lysates were prepared from tissue after homogenizing whole lungs. RNA was extracted using a Total RNA Kit (Omega Bio-Tek). Real-time PCR was performed after synthesizing cDNA using a High capacity cDNA Reverse Transcription kit (Applied Biosystems). The expression of Irf7 (Mm00516791_g1), Il28b (ifnl3) (Mm00663660_g1), Ifitm3 (Mm00847057_s1), Isg15 (Mm01705338_s1), Oasl2 (Mm00496187-m1), Il12b (Mm00434174_m1), Il6 (Mm00446190_m1), Tnfa (Mm00443258_m1), Ifna (Mm03030145-gH), Ccl5 (Mm01302427-m1) and Actinb (4352341E-1112017) were determined by RT-PCR (Applied Biosystems). RT-PCR reactions were performed in microAmp Fast plates (Applied Biosystems) using SensiFAST Probe Hi-ROX kit (Bioline) and a StepOnePlus RT-PCR machine (Applied Biosystems). Relative gene expression levels were calculated by normalizing the Ct levels of the target gene to both endogenous actin levels and an unstimulated WT control using the ΔΔCt method.
Cell lysates were separated on 4–12% gradient gels (Invitrogen) and were transferred to PVDF membranes using the iBlot Gel Transfer system (Invitrogen). Membranes were probed with various antibodies followed by horseradish peroxidase (HRP)-labeled secondary antibodies. Protein bands were visualized by enhanced chemiluminescent reagent (Lumigen) and Amersham Hyperfilm ECL (GE Healthcare). The following antibodies were used for immunoblotting: Tpl2 (Cot M-20), ERK1, ERK2 and β-actin (Santa Cruz Biotechnology), p-ERK1/2 (Thr202/Tyr204), p-STAT1 (Tyr701), p-STAT1 (Ser727) and STAT1 (Cell Signaling Technology).
Cells harvested after overnight stimulation were fixed, permeabilized with triton buffer (PBS+0.5%triton+0.1%BSA) and stained for p-Akt (Ser473) according to manufacturers’ protocol (Cell Signaling Technology). Samples were acquired on a BD LSRII flow cytometer and analyzed using FlowJo software (Tree Star, Inc.).
Data represent means ± SEM, except where indicated. P-values were determined by Students t-test, and significance was assigned for p-values <0.05. Kaplan-Meier analysis using PRISM software was performed to estimate percentage survival of WT and Tpl2-/- groups infected with influenza virus, and p value was determined using a Mantel-Cox test.
|
10.1371/journal.pcbi.1005766 | C-reactive protein upregulates the whole blood expression of CD59 - an integrative analysis | Elevated C-reactive protein (CRP) concentrations in the blood are associated with acute and chronic infections and inflammation. Nevertheless, the functional role of increased CRP in multiple bacterial and viral infections as well as in chronic inflammatory diseases remains unclear. Here, we studied the relationship between CRP and gene expression levels in the blood in 491 individuals from the Estonian Biobank cohort, to elucidate the role of CRP in these inflammatory mechanisms. As a result, we identified a set of 1,614 genes associated with changes in CRP levels with a high proportion of interferon-stimulated genes. Further, we performed likelihood-based causality model selection and Mendelian randomization analysis to discover causal links between CRP and the expression of CRP-associated genes. Strikingly, our computational analysis and cell culture stimulation assays revealed increased CRP levels to drive the expression of complement regulatory protein CD59, suggesting CRP to have a critical role in protecting blood cells from the adverse effects of the immune defence system. Our results show the benefit of integrative analysis approaches in hypothesis-free uncovering of causal relationships between traits.
| Chronic inflammation is associated with chronic diseases, morbidity and mortality while lower base inflammation levels are thought to be predictive of healthy aging. Thus, to pursue a long and healthy lifespan, it is essential to understand the inflammatory regulatory mechanisms. To that end, we studied the functional role of C-reactive protein (CRP)–an inflammatory biomarker that is used to measure cardiovascular risk in clinical practice. There is evidence for a strong genetic component of elevated CRP levels but it is still unclear if it has a direct impact on the processes that lead to inflammatory diseases. In order to elucidate the function of CRP in the blood, we used statistical methods for causal inference to infer causal relationships between changes in CRP and gene expression levels. Our statistical analysis and cell culture experiments suggest that CRP drives the expression of complement regulatory protein CD59. Thus, CRP can have a functional role in protecting human blood cells from the adverse effects of the immune defence system.
| Increased levels of C-reactive protein (CRP) in the blood are associated with tissue injury, infections and inflammation [1]. In addition to acute bacterial and viral infections, chronically elevated CRP levels are predictive of multiple diseases associated with inflammatory processes, e.g. cardiovascular disease (CVD). Therefore, CRP is used as a biomarker to diagnose CVD and other inflammatory diseases [2–4]. Furthermore, a recent large-scale Mendelian randomization (MR) study has shown a possible causal relationship between CRP and several complex traits, most notably a protective effect against schizophrenia [5]. However, little is known about the mechanisms of the underlying inflammatory processes and the interactions between different risk factors that either prevent or lead to a disease.
In the past years, genome-wide association studies (GWAS) have identified thousands of disease-associated genetic loci, and a GWAS meta-analysis of CRP levels in over 80,000 individuals found a number of allelic variants in genes implicated in pathways related to metabolism and immune system [6]. Altogether, these studies have demonstrated a strong genetic component in chronic inflammatory processes. However, the identification of genetic variants without knowing their functional relevance has not been sufficient to tackle disease-informed genetics and provide intervention measures against complex diseases. Hence it is necessary to integrate different omics data and move beyond associations. Causal inference methods and multi-omics approaches have already been applied successfully in the analysis of complex traits, e.g. obesity, cancer and coronary artery disease [7–10].
Recent methodological approaches in causal inference include finding the best-fitting model from the set of previously defined possible causal models using maximum likelihood [10–14], testing for partial correlation criteria based on the theory of d-separation [15, 16], both of these techniques together [17, 18], and MR [19–22]. Especially MR has been increasingly popular of late but requires thousands of samples to achieve adequate statistical power even at nominal significance level 0.05 [22] and is therefore not feasible for hypothesis-free testing in smaller samples. Model selection-based methods do not necessarily rely on p-values and can have more power but are prone to false positive findings [15, 23]. To overcome this trade-off, we propose a combined approach where we first identify a list of candidate causal relationships using a model selection-based approach and then apply MR on this candidate list to disentangle true positive findings.
Here we combined data on genotype, transcriptome and CRP levels to get further insight into the molecular mechanisms regulating CRP concentration. We hypothesized that understanding the complex genetic architecture of the molecular functions behind CRP levels can be aided by overlapping the genetic basis of CRP and the genetic basis of gene expression variability. To this end, we have performed a multi-step analysis procedure. We identified and described the set of genes whose expression levels are associated with CRP levels and then used genotype data to determine the potential causal structure between the expression of these genes and CRP by maximum likelihood. We ensured that the proposed models satisfied all the necessary partial correlation criteria, and then used MR and independent data to decide on true causal findings (Fig 1A). We identified a causal effect of CRP concentration on CD59 expression in whole blood which we validated experimentally. The main study was conducted on 491 individuals from the Estonian Biobank cohort [24] whose genomes and transcriptomes in whole blood have recently been profiled using whole genome sequencing (WGS) [25] and RNA sequencing (RNA-seq) techniques (Fig 1B).
To find CRP-associated genes, we performed a differential expression analysis using the limma-voom framework [26]. However, instead of dividing continuous CRP values into two or more bins, we simply used the close-to-normally distributed log-transformed CRP values as a continuous predictor. Binning a continuous variable would result in reduced power to detect true associations [27]. The models were adjusted for age, sex, body mass index (BMI), blood composition and principal components (PCs) both from the genotype and gene expression data to account for population structure and hidden batch effects.
Controlling the false discovery rate (FDR) at 0.05, we identified 1,614 genes whose expression values were significantly associated with CRP concentrations in the blood (S1 Table). Of all the CRP-associated genes, 1,108 were positively and 506 negatively correlated with CRP. As expected, we observed a high proportion of interferon-stimulated genes, 738 in total (45.7%), which are known to be induced by infections and inflammatory processes [28] (S1 Fig). By a considerable margin, the most significant CRP-associated gene was FAM20A (adjusted p = 7.5×10−17), followed by UPP1, FCGR1A, LDHA and MTHFD2 (Table 1). Similarly to the CRP gene, FAM20A is most highly expressed in the liver. FAM20A is known to be involved in biomineralisation of teeth and mutations in this gene have previously been linked to dental defects and enamel renal syndrome [29, 30]. Pathway enrichment analysis with g:Profiler [31] showed that CRP-associated genes are overrepresented in immune system processes, particularly in innate immune system and interferon signalling pathways, as well as in NOD-like receptor signalling pathway (S2 Table).
To study the genetics influencing the expression of the CRP-associated genes, we performed an expression quantitative trait locus (eQTL) analysis. We used the same set of covariates as before with the addition of one dummy variable coding for different batches of WGS data. We limited our search to single nucleotide polymorphisms (SNPs) located within 250 kb of the genes. For each pair of SNP and gene expression values, we tested whether an additional minor allele of the SNP has a significant additive effect to the level of gene expression. In total, we performed 1,821,299 tests. We identified 39,507 eQTLs for 470 different genes (S3 Table). To validate our findings, we compared the results against the cis-eQTLs reported in whole blood by the GTEx Consortium (version V6p) [32]. We could replicate at least one eQTL for 313 out of the 470 genes (66.6%), altogether 20,536 SNP-gene pairs. This shows good concordance, despite several differences in the study designs and the relatively low power of both studies. Compared to the eQTLs reported by Westra et al. [33], we replicated at least one eQTL in 273 genes (58.1%), altogether 8,998 SNP-gene pairs. The considerably smaller replication rate here is likely to be due to the differences between array- and sequencing-based expression profiling (e.g. lowly expressed genes are likely not replicable in microarray-based eQTL studies) [34].
In the previous steps, we established genes that are associated with CRP through their expression values and we also identified eQTLs for these genes, creating a set of SNP, gene expression and CRP triplets. Assuming directed acyclic graphs, this leaves only a limited number of possible models that these triplets can be functionally acting by (Fig 1A). To determine the most likely causal structure underlying these triplets, we performed likelihood-based causality model selection [10]. That is, we modelled the joint distribution of all possible triplet models by maximum likelihood and determined which was best supported by our data in terms of minimal values of the Akaike information criterion (AIC). To eliminate the situations where both CRP and gene expression were driven by known confounding factors, we performed the analysis on covariate-adjusted CRP and expression values, using the same set of covariates as before (except for gene expression PCs in the case of CRP). As many of the eQTL SNPs were in high linkage disequilibrium (LD) with each other, we first identified independent eQTLs for each gene using stepwise multiple regression, starting from the strongest cis-eQTL. This is a standard approach for discovering independent loci [35]. In total, we found 536 independent eQTLs for 470 different genes.
For 283 out of 536 triplets tested, the difference in AIC values between the causal and colliding models (ΔAIC) was less than 2, which does not give enough evidence to support one model over the other [36]. Among the remaining 253 triplets, 81 showed stronger evidence for the causal model, 163 for the colliding model and 9 for the independent model. Unsurprisingly, the reactive model never achieved the smallest AIC, due to the selection bias of the SNPs. Altogether, 223 unique genes were represented in the 253 triplets. There were 21 genes with multiple independent eQTLs and 15 of them were supported by a single model, showing good consistency (S4 Table). On average, triplets supported by the colliding model showed higher ΔAIC values (Fig 2A). This indicates that the colliding models are of higher quality in our analysis. Genes best supported by these colliding models were enriched in Gene Ontology terms for response to external stimulus and stress (S5 Table).
We could also observe that more significant association p-values between CRP and gene expression do not necessarily translate to greater ΔAIC values (Fig 2B). This result reinforces that many of the correlations resulting from ordinary differential expression analysis are likely to rise due to unmeasured common confounding and care should be taken when interpreting such results.
To be able to clearly isolate genes whose expression with respect to CRP conforms to either the causal or colliding model (i.e. whether gene expression drives CRP or vice versa), we would expect a clear difference in the AIC values of corresponding triplet models, so we considered only triplets with ΔAIC ≥ 10 as candidates. This ΔAIC threshold corresponds to probability 1 –e-5 > 0.99 that the model with the smaller AIC is more likely [36]. We further required that the models suggested by the maximum likelihood procedure satisfied partial correlation criteria (S4 Table). More specifically, for the causal models we expect to observe at least nominally significant association between the SNP and CRP values, but not if we conditioned on gene expression values. On the other hand, for the colliding models we expect to observe an association between SNP and CRP conditional on gene expression, but not otherwise.
The ΔAIC and partial correlation criteria already provide evidence of causality but unmeasured common confounding can be an issue and lead to overconfident claims. Therefore, we subjected all the best models to MR analysis using published CRP summary statistics [6]. For causal models, we checked whether the eQTL was significantly associated with CRP in the published data. For colliding models, we selected 16 out of 18 CRP-associated SNPs from the CRP meta-analysis [6] (the remaining 2 SNPs had a minor allele frequency of 2.2% and no individual in our sample had two minor alleles of these SNPs) and performed association tests between these SNPs and gene expression in the Estonian data, looking for enrichment of small p-values. To increase power, we also combined the 16 SNPs into a genetic risk score (GRSCRP) using published effect sizes as weights. To estimate the causal effect, we used the two stage least squares (TSLS) method which is standard in MR analysis [37].
Only ten triplets (1 causal, 9 collider) had ΔAIC at least 10 (Tables 2 and 3). Out of those, FADS2 was the only gene best supported by the causal model. The corresponding SNP (rs61897793) was not present in the CRP meta-analysis so we performed summary statistic imputation [38] (with UK10K as reference panel) to infer the CRP-association statistic. It did not reach nominal significance (p = 0.0996). However, SNPs in the FADS2 gene have been associated with circulating phospholipid trans fatty acid and plasma phospholipid n-3 fatty acid levels by the CHARGE Consortium [39–41]. CRP has been shown to bind phospholipids through phosphorylcholine [42] and plasma CRP values have been reported to drop with phospholipid-induced agglutination [43]. FADS2 has a known function in the synthesis of arachidonic acid that is relevant in inflammatory processes and has been associated with both CRP and risk of CVD [44]. Furthermore, SNPs in FADS2 have also been associated with low-density lipoprotein (LDL) and total cholesterol in European populations [45], in addition to weight and BMI in Greenlanders [46]. LDL-cholesterol and BMI have in turn been causally implicated with risk of CVD [47] and CRP [19]. These results are consistent with a mediated causal indirect effect of FADS2 on CRP, even though a recent summary-level MR analysis did not identify FADS2 expression causal to BMI [22]. We could not fully confirm a causal link between FADS2 expression and CRP in this study, but together with evidence from other studies, our results could warrant further analysis with a larger sample size.
The top genes following the colliding model were C3AR1, HIATL1, NRG1, SEMA4A, PLGRKT, CD59, FCGBP, IFITM3 and KREMEN1. Out of these, CD59 was the only gene that showed enrichment of low association p-values with the 16 individual CRP-related SNPs from the CRP meta-analysis (Fig 3A, Kolmogorov-Smirnov test for uniform distribution p = 0.026). The estimate of causal effect from CRP to CD59 expression using all the individual SNPs as instruments (beta = 0.20, SE = 0.06, p = 0.0012) was similar to using only GRSCRP as a single strong instrument (beta = 0.24, SE = 0.08, p = 0.0022), showing a similar positive slope in both cases. None of the SNPs nor GRSCRP were correlated with CD59 expression conditional on CRP values, satisfying the conditional independence assumption of MR. To detect and correct for possible bias from pleiotropy, we calculated the causal effect estimate using each SNP as a single instrument, visualized the individual causal estimates by a funnel plot and performed an MR-Egger test proposed by Bowden et al. [21] (Fig 3B and 3C). There was very little directional pleiotropy present (the intercept coefficient from the MR-Egger test was -0.002) and the MR-Egger-corrected causal effect estimate was similar to the TSLS estimates (beta = 0.21). For the other genes, there is less evidence for a causal effect from CRP to expression and instead, unmeasured confounding might be responsible for the elevated ΔAIC values.
Functionally, CD59 regulates the complement membrane attack complex (MAC) [48] and has been reported to have a protective effect against atherosclerosis by restricting MAC formation [49]. CRP has also been shown to upregulate CD59 in endothelial cells [50] and although some of the findings of this paper were later questioned by the effect of a common additive sodium azide (NaN3), the upregulation of CD59 by CRP was not disproved [51]. To further confirm that the expression of CD59 is upregulated in the blood by elevated CRP levels, we performed cell culture experiments where we stimulated peripheral blood leukocytes with increasing concentrations of CRP (Fig 4). We found a dose-dependent upregulation of CD59 on cell surface by flow cytometry after 48 hours, which importantly was not present when only NaN3 was added to the cell cultures. The dose effect was most prominent in lower doses while reaching a plateau at the concentration of 12.5 μg/ml. A similar trend in increased CD59 surface levels, albeit slightly lower, was also present after 24 hours (S2 Fig). Altogether, our results indicate a causal role of CRP on CD59 expression levels.
We identified altogether 1,614 genes that are associated with CRP by their expression values in the blood. Using pathway analysis, we have shown that these genes are enriched in immune system related functions and thus are good candidates to be directly relevant in biological processes concerning CRP. In agreement with their function in innate immune responses, ca 46% of CRP-associated genes comprised interferon-stimulated genes, which have a wide range of activities ranging from control of bacterial and viral infections, upregulation of chemokines and chemokine receptors and regulating blood cellular homeostasis [52]. However, our results suggest that the most significant CRP-associated genes should not be readily interpreted as the most important in terms of causal effects. To find causal relationships, we integrated gene expression and CRP data with genotype data and used a combined analysis approach. First, we applied integrative genomics techniques to filter out a list of candidate causal relationships and then applied Mendelian randomization to determine the final outcome. We report the expression of CD59 as being causally affected by CRP concentration in the blood, and provide experimental validation of the result.
Our finding of CRP-mediated induction of CD59 suggests a negative feedback mechanism to protect blood cells against potentially damaging complement responses that are upregulated during infections and inflammation. Ubiquitously expressed CD59 is a specific inhibitor of complement membrane attack complex (MAC) formation, which is the main effector of complement-mediated tissue damage and leads to osmotic lysis of targeted cells [48, 53]. Through its inhibitory binding to complement members, CD59 blocks MAC formation and MAC-induced cell lysis. For example, individuals having mutations in CD59 have decreased capacity to inhibit the complement MAC formation and develop an early-onset hemolytic phenotype associated with vascular disease [54]. Thus, our result provides a new insight into the molecular mechanism of CRP function in protecting human blood cells from the adverse effects of the innate immune defence system, albeit the exact interplay between CRP and CD59 needs to be determined in further experiments.
We also found that the expression of FADS2 can be potentially relevant in terms of CRP regulation. There are many known associations between FADS2 genotypes, lipid levels, inflammatory markers and CVD that together are consistent with a mediated causal effect of FADS2 expression on CRP. We did not find conclusive evidence of this causal relationship in this study but suggest further analysis to ascertain the interplay between these traits in terms of inflammation and disease.
Our study has several limitations. By selecting candidate triplets using a stepwise analysis approach, we make an implicit assumption that variation in DNA leads to variation in the phenotype in a linear manner. However, it is reasonable to believe that variation in the phenotype values is determined by the combined variation of many factors in multiple omics layers [55]. Further, the triplet models that we considered (including the assumptions) are likely to be simplistic representations of actual relationships between variables and although we accounted for several known covariates and captured technical variation in the data, it is possible that unmeasured variables are acting as confounders in some cases. These drawbacks can yield false positive findings. However, ΔAIC of at least 10 provides strong evidence that the model with the smaller AIC is considerably better supported by the data [36]. Also, triplet models have shown good promise in distinguishing between competing models [10]. Moreover, we make a causal claim only after comprehensive MR analysis.
A bigger limitation is a lack of statistical power to find more causal relationships, mostly due to our small sample size. Assuming that a genetic instrument (e.g. GRS) describes 5% of the variation of the exposure and a standardized causal effect size between the exposure and the outcome is 0.1, we would need around 15,000 samples to detect an instrument-outcome association with 80% power at nominal significance level 0.05 [56]. We would have to assume a slightly larger causal effect to achieve only 20% power in the Estonian data. It shows that MR is underpowered for hypothesis-free testing in smaller samples. On the other hand, relying on likelihood-based methodology alone can give misleading results due to the number of false positive findings [15, 23]. These can be expensive, time-consuming and difficult to experimentally validate. We think that our approach of combining MR with prior filtering by maximum likelihood modelling can be useful in such cases.
Our analysis strategy could be applied to any trait, but the available sample size in the Estonian Biobank was not sufficient for more complex traits, like BMI and height. A recent MR analysis on summary-level data implicated 68 causal genes for height and 9 for BMI [22], which we attempted to replicate. For height, only 5 genes (out of more than 12,000 tested) cleared the significance threshold at the first analysis step (i.e. significant trait-expression association) in the Estonian data, none of which are among the 68 published causal height genes. Moreover, no gene passed the ΔAIC ≥ 10 threshold filter to indicate a likely causal model. For BMI, only one of the 9 reported causal BMI genes was correlated with BMI in the Estonian data, with ambiguous results in terms of causality. There were again no likely candidates for the causal model. Negative results here can partly be due to different technologies used in quantifying gene expression levels (we used RNA-seq, [22] used microarrays) and partly because BMI and height are far more complex than CRP and require more samples to analyse. Parallel to experimental validation of the CRP-CD59 link, we also performed summary-level MR in an attempt on an alternative validation of this relationship by using all association summary statistics (trans effects based on ~16,000 individuals) between CD59 expression and CRP-associated SNPs (or their proxies due to data availability) provided by the eQTLGen Consortium. We could not detect a causal effect here, probably again due to low power. Using a single SNP as an instrument and assuming a standardized causal exposure-outcome effect of 0.1, we would need close to one million samples to detect an instrument-outcome association with 80% power.
Our filtering approach is conceptually similar to that of Schadt et al. [10] but there is a key difference. Notably, we use gene expression values as the central building blocks to be associated with the phenotype, instead of genotype data. As a result, we have to compute the likelihoods of 4, not 3 possible models, but our approach also has several benefits. First, there are considerably fewer genes than genetic variants, reducing our multiple testing burden. Second, we do not rely on identifying phenotype-associated genetic variants that can be difficult to detect genome-wide due to small effects. The effects of gene expression on the phenotype are likely to be much bigger compared to the effects that individual SNPs have, so in some cases it could be possible to trace the variation in the phenotype back to the SNP only through gene expression [57]. Third, we are able to detect a causal effect from phenotype to gene expression if the colliding model holds. This would not be possible if we required an explicit SNP-phenotype association for the triplets, since under the colliding model that association would not exist. In fact, only our modified approach could identify the CRP-CD59 link.
In summary, we have demonstrated that combining gene expression data with genotype and phenotype data–and importantly using integrated modelling techniques–can give insight to the causal molecular mechanisms underlying trait variation even if the sample size is limited. Using new RNA-seq data from the Estonian Biobank, we have presented genes that are associated with CRP based on the expression values, identified genetic loci that guide this expression, and provided evidence about the direction of causal effects between CRP and a few genes. Most notably, we have shown by statistical analysis and cell culture stimulation assays that CRP upregulates CD59 expression in whole blood and can thus have a role in protecting human blood cells from the adverse effects of the immune defence system. We have also presented suggestive evidence of an indirect causal effect of FADS2 expression to CRP levels. These findings can potentially provide deeper understanding of the functional roles of CRP, but further investigations are required to evaluate these results in terms of chronic inflammation and disease.
All participants have provided an informed consent for the use of their medical records (www.biobank.ee).
This study is based on the Estonian Biobank cohort, developed and maintained by the Estonian Genome Center, University of Tartu (EGCUT). This is a volunteer-based population cohort with close to 52,000 participants, which is around 5% of the Estonian adult population. All participants have donated blood samples, 2,700 of which are characterised by clinical biochemistry measurements including the levels of C-reactive protein (CRP, mg/L), leukocytes, erythrocytes and thrombocytes [24]. Recently, whole genomes were sequenced for 2,244 individuals in the EGCUT cohort [25]. Of those, 1,026 have biochemistry measurements and 586 also have RNA-seq data.
The average CRP value in our data was 2.34 mg/L with standard deviation 3.84. The distribution of CRP measurements was skewed to the right and the maximum CRP value measured was 53.8 mg/L. We have taken a natural logarithm of CRP in this study to bound the effect of slightly outlying values. There was a noteworthy correlation between the levels of CRP and different blood components, most notably white blood cells (p = 2.6×10−14), also BMI (p = 6.9×10−45) and age (p = 6.5×10−13). One individual had missing CRP and blood component values; these were imputed as the corresponding averages of the remaining 1,025 individuals. Four individuals had missing BMI values but all of these originated from follow-up questionnaires so we imputed them from the values given on recruitment a couple of years earlier.
RNA was extracted from thawed Tempus tubes using TRIzol Reagent (Invitrogen) and further purified using RNeasy Mini Kit (Qiagen). Globin mRNA was depleted using GLOBINclear Kit (Invitrogen). RNA quality was checked using an Agilent 2200 TapeStation (Agilent Technologies). Sequencing libraries were prepared using 200 ng of RNA according to the Illumina TruSeq stranded mRNA protocol. RNA sequencing was performed at the Estonian Genome Center Core Facility using Illumina paired-end 50 bp sequencing technology according to manufacturers specification.
We used Trimmomatic (version 0.36) [58] to remove the adapters and leading and trailing bases with a quality score B. Quality control was done by FastQC (version 0.11.2) [59]. We used STAR (version 2.4.2a) [60] to map the reads to a human genome reference version GRCh37.p13. Concurrently, STAR also counted reads that mapped to each genomic feature using the same algorithm as default htseq-count. In this study, only protein coding genes from autosomal chromosomes were used as evidenced by the Ensembl BioMart (genome assembly GRCh37.p13, release 75) database, the rest were filtered out.
The initial pre-processing and quality control of the WGS data was done by EGCUT as reported in [25]. For our purposes, we performed some further filtering steps using Plink 1.9.0 [61]. We excluded chromosomes X and Y from the analysis and only included those individuals with RNA-seq and CRP data (N = 491). From the remaining sample, genetic variants with minor allele frequency below 0.05 or missing call rates exceeding 0.01 were filtered out. We performed identity by descent analysis (prior to that, we excluded SNPs that were in high pairwise linkage disequilibrium: r2 > 0.5 in a sliding window of 50 bases with 5 base increments) which revealed 4 pairs and 1 trio of individuals related to each other (genetic relatedness > 0.1). Only one individual from each group was kept.
As a further quality control measure, we applied MixupMapper [62] to detect and in some cases correct for sample mix-ups. We also performed principal component analysis on the gene expression data and identified a batch of samples with a different gene expression structure compared to other samples. This was discovered to be due to a technical problem during library preparation and affected samples were removed from the analysis.
We also removed non-expressed and lowly expressed genes from the analysis by including only those genes that for at least ten individuals had a count per million (cpm) value greater than 1. After all the filtering steps, the remaining sample size was 491 and the remaining number of genes was 12,619.
RNA-seq count data is heteroscedastic and that remains the case after the log(cpm) transformation. One of the typical approaches in this case is using weighted linear regression where individual gene expression levels are attributed with weights that are inverse proportional to variance. We thus performed the analysis in the limma framework (version 3.26.9) [63] and found gene expression weights by the voom [26] method that has been shown to work well in differential expression analysis; log(CRP) was used as an exploratory variable and gene expression levels as dependent variables. We adjusted for possible confounding effects from age, BMI, sex and blood components (neutrophils, eosinophils, basophils, lymphocytes, monocytes, erythrocytes and thrombocytes). The first four PCs on the genotype data were used to control for population structure (PCs were again calculated on LD-pruned data) as established in [64]. To account for batch effects in the gene expression data, we used the sequencing batch date as a covariate. Raw RNA-seq counts were normalized with the weighted trimmed mean of M-values [65] method in the edgeR package (version 3.12.1) [66]. Logarithm of count per million was used as the final gene expression measure.
Principal component analysis on the gene expression data revealed hidden batch effects despite controlling for the sequencing batch date. To increase power and the reliability of results, we applied a simple algorithm to account for such hidden effects in a similar fashion to surrogate variable analysis [67] and PEER [68]. We tested whether the top PCs were significantly associated with CRP and decided to use the first two PCs as control variables in the further analysis, because we could see strong associations with CRP starting from the third PC. We adjusted the models for confounders such as age, gender and BMI but also the number of different blood cells to account for differences in gene expression in these cells.
We used Benjamini-Hochberg correction to correct for the number of tests and control the FDR at 0.05. Top genes were subjected to enrichment analysis by g:Profiler [31].
With each of the top genes that were significantly associated with CRP by their expression levels, we performed a cis-eQTL analysis. An association between a SNP and a gene was determined only if the SNP resided not farther than 250 kb from the gene. We used the same set of covariates as before, including a batch variable of the WGS data as an additional covariate. The analysis was performed in Plink using ordinary least squares with gene expression measured as log(cpm) as the dependent variable. To control for the number of tests, we used a two-step procedure.
First, we controlled the family wise error rate for each gene by doing 10,000 permutation tests in Plink. However, we did not want to limit our p-values with 1×10−4. For each gene, we pulled the highest t-statistic value of every permutation (10,000 in total) and transformed them to p-values. We used the minimum sample size of tested SNPs in the calculation of degrees of freedom, because SNPs contained a variable amount of missing values (but at most 10%) and the SNP that obtained the highest t-statistic was not specified in the Plink output. For each gene, we transformed the 10,000 extreme p-values by −log10 and then fitted a Gumbel distribution G(μ, β) on them by estimating μ and β. Finally, nominal p-values pnom were transformed to permutation p-values by pperm = P(X > −log10(pnom)) where X ~ G(μ, β). This procedure is conceptually very similar to the one implemented in the FastQTL tool, where Beta distribution is used to model the smallest non-transformed p-values [69].
Second, to control for the number of genes tested, we used the Bonferroni method. All SNP and gene expression pairs with permutation p-values less than 0.05/N (N = 1,614 was the number of unique genes tested) were deemed significant.
We established genes whose expression was associated with CRP and SNPs that were QTL to the expression of those genes. We called these intertwined components triplets. To determine the most likely causal structure underlying these triplets, we performed maximum likelihood modelling in similar fashion to Schadt et al. [10], albeit with some differences discussed above. Assuming directed acyclic graphs and the correlation structure within each of the triplets, the following models are possible (Fig 1B):
These models are not Markov equivalent like E -> CRP and CRP -> E in which case the joint distributions would be equal: P(E, CRP) = P(E)P(CRP|E) = P(CRP)P(E|CRP). This means that by calculating the model likelihoods we can determine, for each triplet, the most likely model and hence identify the most plausible causal direction between E and CRP. We found residual CRP from the model log(CRP) = Xb + e and residual expression values from the model log(cpm) = Xb + e where X includes the confounders (age, sex, blood components, PCs). We then used these residuals in the triplet models. We assumed normal distribution for log(cpm) and log(CRP). Wherever necessary, we also assumed multivariate normal distribution and used the appropriate formulas for conditional distributions. We constructed likelihood functions corresponding to each of the above models and maximized them by numerical optimization (optim function in R). Finally, we chose the model with the minimal AIC as the likeliest for each triplet.
Residual CRP and expression values were used for the analysis of colliding models with MR principles in the Estonian data for consistency of the variables used in the maximum likelihood modelling of triplets. The causal effect between CRP and CD59 expression was estimated using the tsls function in the R sem package. Summary-level MR analysis was performed using the inverse-variance weighted method [20].
Human heparinized peripheral blood was diluted with OpTmizer cell culture medium 1:4. The peripheral blood cells from two independent donors were cultivated in three replicates with five increasing CRP (Sigma) doses (50×4−4, 50×4−3, 50×4−2, 12.5 and 50 μg/ml) for 24 and 48 hours. Separate control experiments with 0.1% NaN3 in three replicates were included. The cell cultures were stained with phycoerythrin-conjugated anti-human CD59 antibody (Biolegend) and treated with Lysing solution (BD Biosciences) to eliminate erythrocytes before analysis by flow cytometer (LSRFortessa) and FACSDiva software. Granulocytes were gated according to their forward and side scatter characteristics and CD59 staining intensity recorded as mean fluorescence index. Approval was obtained from the ethics committee of the University of Tartu.
|
10.1371/journal.pgen.0030063 | Population Stratification of a Common APOBEC Gene Deletion Polymorphism | The APOBEC3 gene family plays a role in innate cellular immunity inhibiting retroviral infection, hepatitis B virus propagation, and the retrotransposition of endogenous elements. We present a detailed sequence and population genetic analysis of a 29.5-kb common human deletion polymorphism that removes the APOBEC3B gene. We developed a PCR-based genotyping assay, characterized 1,277 human diversity samples, and found that the frequency of the deletion allele varies significantly among major continental groups (global FST = 0.2843). The deletion is rare in Africans and Europeans (frequency of 0.9% and 6%), more common in East Asians and Amerindians (36.9% and 57.7%), and almost fixed in Oceanic populations (92.9%). Despite a worldwide frequency of 22.5%, analysis of data from the International HapMap Project reveals that no single existing tag single nucleotide polymorphism may serve as a surrogate for the deletion variant, emphasizing that without careful analysis its phenotypic impact may be overlooked in association studies. Application of haplotype-based tests for selection revealed potential pitfalls in the direct application of existing methods to the analysis of genomic structural variation. These data emphasize the importance of directly genotyping structural variation in association studies and of accurately resolving variant breakpoints before proceeding with more detailed population-genetic analysis.
| Several recent studies have demonstrated that deletions, duplications, and inversions contribute a substantial fraction of the total amount of variation present in the human genome. In this study, we provide a comprehensive population-genetic analysis of a single deletion previously identified by comparing the genome of a single individual against the human genome reference sequence. Complete genomic sequence spanning the deleted region was obtained, allowing us to define the deletion breakpoints and develop a direct genotyping assay. Analysis showed that the deletion removes a member of a gene family involved in the innate immune response against viral pathogens. We genotyped samples from a human diversity panel and found drastic differences in the frequency of the deletion around the world. Using data from the HapMap project and the application of existing analysis techniques, we illustrate the importance of directly genotyping this type of variation and of clearly defining its boundaries. Without this level of detail the potential functional importance of such variation may be missed.
| The APOBEC3 family is known to play a role in innate cellular immunity against retroviral infection. The gene family has undergone an expansion in primates, increasing from a single copy in rodents to at least seven copies in humans [1–3]. Among primates, the APOBEC3 family has been subjected to strong and continuing selective pressures at the amino acid level [3,4]. APOBEC3 proteins defend against retroviruses by deaminating cytosine residues to uracil, resulting in hypermutation and degradation of the viral genome. Members of this gene family contain either one (APOBEC3A and APOBEC3C) or two (APOBEC3B, APOBEC3F, and APOBEC3G) conserved cytosine deamination domains [1,2]. In addition to their role in innate retroviral immunity, some APOBEC3 genes appear to inhibit hepatitis B virus infection [5–8] and the retrotransposition of endogenous elements [9–12]. It is thought that at least part of this activity occurs through a deamination-independent mechanism [12].
Several recent studies have brought increased attention to classes of genomic variation such as deletions, inversions, and copy-number polymorphisms [13–20]. It is thought that these variations contribute substantially to inter-individual genomic, and perhaps, phenotypic variation, but the structure and population characteristics of these variants remain largely unexplored. A deletion in the APOBEC3 gene cluster was recently identified using two different approaches. The deletion was first discovered by mapping end-sequence pairs from a human fosmid library against the human genome reference sequence assembly [16]. A cluster of discordant fosmid clones whose end-sequences mapped further apart than the expected fosmid insert size predicted a deletion of ∼30 kb near the APOBEC3B gene. Later, a second approach confirmed the deletion based on an interrogation of a dense single nucleotide polymorphism (SNP) marker map generated as part of the International HapMap Project [17]. This method discovered deletions by identifying clusters of SNPs that showed apparent non-Mendelian inheritance, deviations from Hardy-Weinberg equilibrium, or evidence of null genotypes. Nevertheless, this variant was not detected in a recent genome-wide screen of structural variation in the HapMap populations using BAC or SNP-based microarrays [20]. The availability of a fosmid clone that captured the deletion event allowed us to sequence the structural variant in its entirety and confirm its presence. Precise sequence definition of the deletion enabled the design of specific genotyping assays across the deletion breakpoints. We present here a sequence-based analysis of this deletion polymorphism, a worldwide population survey of the deletion frequency (1,277 DNA samples), and an analysis of the surrounding haplotype structure. The results suggest this is a functionally important structural variant that is stratified in the human population.
We sequenced the entire insert of one of the fosmid clones whose end-sequence pairs had initially identified the structural variant. Alignment of this sequence with the sequence from the finishing human genome sequence assembly (hg17) confirmed the presence of a deletion overlapping the APOBEC3A and APOBEC3B transcripts (Figure 1). Consistent with non-allelic homologous recombination as the likely mechanism of origin, the deletion breakpoints mapped to two highly identical tracts of sequence: 350 bp in length, 100% sequence identity. In the deleted configuration, a single copy of this sequence exists and the 29.5 kb of sequence between them is removed (position 37,683,131–137,712,716 on Chromosome 22 of hg17). The deletion removes the genomic sequence between the fifth exon of APOBEC3A and the eighth exon of APOBEC3B, leading to a predicted full-length functional hybrid transcript with a predicted amino acid composition identical to APOBEC3A. Thus, individuals possessing this structural variant would lack at least one copy of the unique coding portion of APOBEC3B. Interestingly, the predicted transcript would contain the 3′ UTR from APOBEC3B, but be subject to APOBEC3A upstream regulatory signals.
The availability of the complete sequence of the deletion breakpoints allowed us to design PCR breakpoint assays which distinguished insertion and deletion alleles (Figure 1 and Materials and Methods). We genotyped 1,007 individuals from 51 populations included in the Centre d'Etude du Polymorphisme Humain (CEPH) Human Genome Diversity Panel (HGDP) and found that the deletion frequency was highly variable (Figure 2). The deletion is rare in African and European populations (frequency of 0.9% and 6%, respectively), more common in East Asian and American populations (36.9% and 57.7%), and almost fixed in Oceanic populations (92.9%; Tables S1–S3). As a control against potential SNPs under the PCR primer binding sites leading to an overestimate of the frequency of homozygotes, we reanalyzed all 127 samples initially scored as deletion homozygotes with a second PCR assay targeted to the insertion allele (see Materials and Methods). We reclassified 25 samples (19.6%) as hemizygous (Table 1). In order to rule out further large-scale genotyping error, we calculated estimates of Hardy-Weinberg equilibrium for each population. No significant deviations were observed.
As a measure of population differentiation, we calculated an overall FST value of 0.2843 for the APOBEC3B deletion in the HGDP. Large FST values are consistent with either geographically restricted selection (local adaptation) or demographic history (i.e., population bottlenecks and founder effects). To distinguish between these two possibilities, we calculated an empirical FST distribution using 2,540 autosomal SNPs and 207 small indels genotyped in individuals from the same 51 populations [21,22]. Of the 2,747 loci, 52 had an FST value greater than that obtained for the APOBEC3B deletion, placing this deletion within the top 2% of the empirical distribution. Estimates of FST may be sensitive to allele frequency, so we repeated this comparison by only considering the 635 loci that had a global minor allele frequency between 0.17 and 0.27. A single SNP (rs2250341, located in an intron of the PCP4 gene) of this subset had an FST value greater than the APOBEC3B deletion, placing the deletion in the top 0.16% of the frequency-matched empirical distribution.
A striking feature of the deletion is the clinal increase in frequency as one moves eastward away from Africa (Figure S1). In order to further delineate this pattern, we repeated this analysis using pairwise FST estimates between all possible combinations of the 51 populations (Figure S2). The analysis differentiates Oceanic, Amerindian, and some East Asian populations from other human populations based on the frequency of the deletion variant in comparison to other SNP and indel loci in the same populations. This suggests that this pattern is not solely the result of demographic history.
We genotyped the individuals included in the HapMap project: consisting of samples from the Yoruba people of the Ibidan Peninsula in Nigeria (referred to as YRI), the CEPH project in Utah (CEU), the Han Chinese population of Beijing (CHB), and individuals of Japanese ancestry from the Tokyo area (JPT); and searched for evidence of linkage disequilibrium (LD) between the APOBEC3B deletion and flanking HapMap Phase I SNPs (Tables S4–S5). In contrast to other deletion polymorphisms [17,19], we found no single SNP to be in strong LD (r2 greater than 0.8) with the deletion variant (Figure 3). In the Yoruba sample there was one rare SNP with an r2 value of 0.663. This SNP, rs733107, has a minor allele frequency of 0.025 with two of the three Yoruba deletion chromosomes carrying the minor allele. Interestingly, we also found that no two-marker combination had an r2 value greater than 0.8 (the maximum values are 0.254 for CEU, 0.499 for JPT and CHB, and 0.661 for YRI). However, it remains possible that more sophisticated multi-marker approaches will be able to successfully tag the deletion variant using existing SNPs. Unlike other regions of the human genome, such as those enriched for complex segmental duplications, the SNP density in this region (approximately one SNP every 2.5 kb) is not significantly reduced. Since the deletion frequency shows drastic differences among the HapMap populations, the absence of a suitable single marker tag is likely a consequence of a bias in the ascertainment of SNPs typed in the HapMap panel.
Although no single SNP can act as a reliable proxy for the variant, we noticed that the deletion appears to occur on a common haplotype. Treating the deletion locus as a bi-allelic variant, we constructed phased haplotypes using 21 HapMap Phase II SNPs genotyped in all populations and located within 25 kb of the deletion boundaries and performed a haplotype network analysis (Figures 4 and S3; Tables S6 and S7) [23] (http://fluxus-engineering.com).
We identified 49 distinct haplotypes over this region. Overall, 91% of the deletion events (YRI = 3/3, CEU = 8/8, and 61/68 of the JPT and CHB) lie on a single common haplotype (haplotype 28) which differs from haplotype 31 only by the presence of the deletion variant. Two additional haplotypes are observed only for JPT and CHB deletion chromosomes: haplotype 34 (n = 4 chromosomes) and haplotype 29 (n = 3 chromosomes) differ from haplotype 28 by a single nucleotide difference. The former may represent an independent occurrence of the deletion on haplotype 39 or a recombination event between haplotypes 39 and 29.
In order to further investigate the unusual shared haplotype structure and potential signatures of selection, we assessed the deletion haplotype for evidence of extended homozygosity [24,25]. We calculated extended haplotype homozygosity (EHH), the relative extended haplotype homozygosity (REHH), and the extended haplotype length (EHL) for both the deletion and insertion alleles (Figures 5 and S4; Tables 1 and 2). Without correcting for the decreased size of the deletion haplotype, a potentially strong signal of local adaptation indicated by a high frequency extended haplotype for the deletion allele in Asia is observed (Figure 5B and 5C; Table 3). Accounting for the physical reduction in chromosome size due to the deletion, however, largely eliminates this signal in the Asian population (Table 2). Nevertheless, the haplotype analysis does suggest weak signals of selection, particularly in the Yoruba population.
There are three important results of this examination of the APOBEC3B deletion. First, the deletion occurs between two asymmetric gene structures (APOBEC3B and 3A) and produces a hybrid transcript whose putative coding sequence maintains its frame. Despite the fact that the recombination event occurs between coding exons, the amino acid sequence of the hybrid gene is identical to APOBEC3A with the net effect being complete loss of APOBEC3B and potential altered regulation of APOBEC3A due to juxtaposition of novel 3′ regulatory sequences. Second, the deletion variant shows dramatic population stratification with significantly elevated FST values observed for Eastern Asian, Oceanic, and Amerindian populations. The magnitude of the FST values compared to a set of other genome-wide loci from the same populations suggest that these observed frequency differences are not due to demographic history alone. Third, we observe that no tag-SNP currently exists for this deletion. A sophisticated, multi-marker tagging approach may successfully tag this allele, but this approach is complicated by the observation that the major deletion haplotype (haplotype 28) is identical to another haplotype (haplotype 31), except for the presence of the insertion allele. Thus, despite the fact that nearly 40% of the world's population carries at least one copy of this deletion, a suitable SNP surrogate does not yet exist. In light of its abundance, it is noteworthy that this variant was not detected in a recent genome-wide screen of copy-number variation in the human genome [20].
These data emphasize that the phenotypic impact of this and potentially other structural variants may be overlooked in association studies unless the structural variant is directly genotyped. We also observed that potentially misleading results can be obtained from the direct application of existing haplotype-based methods for detecting selection to structural variants. In our study, adjusting for the length of the haplotype reduced the significance of signals of selection using EHL methods (Figure 5). This highlights the importance of not only directly genotyping this type of variation, but of also clearly resolving the breakpoints of the event. If genome-wide screens for selection based on EHL are performed without controlling for changes in the length of the genome sequence, artifactual signals may be observed.
Our observations of the deletion architecture, the population frequency patterns, and the haplotype structure indicate that variation at this locus may be important. Analysis also revealed a weak, suggestive signal of selection for the deletion. Since it appears that the deletion occurred once and has since spread into other populations, a complete understanding of the history of this locus would require detailed knowledge of the different selective regimes and demographic histories unique to these populations.
Until a more comprehensive population genetic theory is developed and a phenotypic consequence of the deletion event is demonstrated, one must be cautious about placing too much emphasis on a potential selective advantage for the deletion. Nonetheless, several possible scenarios may account for the patterns observed at this locus. First, it is possible that a 29.5-kb deletion of a gene may have altered the properties of genetic recombination on this specific haplotype. Long-range haplotype tests are highly sensitive to variance in recombination rates among different haplotypes [25]. Suppressed recombination could, in principle, retard LD decay resulting in an underestimate of the allele's age leading to erroneous signals of recent selection. If this were the case, one would expect a striking correspondence between long-range haplotype-based signatures of selection [26] and the more than 1,000 deletion polymorphisms that have now been documented for the human genome [13–20]. Such a correspondence has not yet been observed, although few structural variants have been studied at this level of detail.
In a second scenario, the deletion may simply be a genetic marker for another selective event that has occurred on this specific haplotype (haplotype 28). In this model, the deletion event is a genetic hitchhiker as opposed to the causative allele. The APOBEC3 gene cluster has been subject to positive selection at many different time points during human and primate evolution [3,4,26]. Thus, a partial sweep of the deletion variant could occur if individuals carrying this specific haplotype had greater resistance to specific pathogens, perhaps as a result of amino acid changes in adjacent members of the APOBEC gene family. In such a scenario, however, the fitness advantage would have to outweigh loss of the APOBEC3B gene and the potential regulatory changes of APOBEC3A incurred by the disruption to the surrounding genomic region. A similar scenario has been put forward for rearrangements associated with the alpha-globin gene family. In this case, recurrent deletions of Hba1 and Hba2 associated with alpha-thalassemia have risen to high frequency in Mediterranean and Pacific rim populations [27]. Data from Papua New Guinea suggest that homozygous and heterozygous genotypes confer protection against malaria and other infectious disease [28]. A neutral deletion seems unlikely in light of the conservation of this gene in humans and other great ape species ([29] and J. M. Kidd, unpublished data). Moreover, APOBEC3B has recently been shown to inhibit replication of the hepatitis B virus, an observation that may warrant further study in light of the frequency of this deletion [8,30].
In a third scenario, the increased frequency of the deletion could represent a shift in the balance of selective forces impacting this locus. There may be a significant cost associated with the maintenance of active cytidine deaminases due to their mutagenic potential [3,31,32]. It has been shown that APOBEC3B protein is present in the nucleus of cells where it may act to repress retrotransposition in early stages of development and in the germ line [8,9]. When the threat posed by both endogenous and exogenous viral activity is high, the protective properties offered by APOBEC3B may outweigh the risks associated with its own activity. When rates of endogenous viral activity are low and when changes in environment reduce the presence of exogenous virus activity, the detrimental effects associated with APOBEC3B may outweigh its benefit, resulting in strong selective pressure in favor of the deletion. Similar arguments have recently been proposed to account for the presence of an impaired allele of the retroviral defense gene TRIM5α and is consistent with evidence of increased retroviral activity in African but not Asian apes [33,34]. When it comes to innate immune system genes such as APOBEC3B, less truly may be more [35].
A shotgun sequence assembly of the fosmid insert containing the putative deletion was generated as previously described [16]. Prior to sequencing, a fingerprint map with four independent restriction enzymes (EcoRI, HindIII, BglII, and NsiI) confirmed the ∼30-kb deletion [36]. Deletion breakpoints were identified by comparison of the fosmid insert sequence against the human genome reference sequence assembly (hg17) using ClustalW and BLASTN [37,38].
We designed PCR breakpoint assays to distinguish the insertion and deletion alleles based on the following oligonucleotide sequences: Deletion_F: TAGGTGCCACCCCGAT; Deletion_R: TTGAGCATAATCTTACTCTTGTAC; Insertion1_F: TTGGTGCTGCCCCCTC; Insertion1_R: TAGAGACTGAGGCCCAT; and Insertion2_F: TGTCCCTTTTCAGAGTTTGAGTA; Insertion2_R: TGGAGCCAATTAATCACTTCAT. Deletion primers are specific to the deletion sequence configuration and generate a 700-bp PCR product upon amplification. Insertion1 and Insertion2 primers amplify only the insertion configuration and produce 490- and 705-bp products, respectively. Insertion and deletion PCR assays were performed separately, the products pooled, and visualized on a standard 1.5% agarose gel. PCR was performed in 17-μl reactions composed of 0.85 μl of a 10-μM dilution of the forward primer, 0.85 μl of a 10-μM dilution of the reverse primer, 8.5 μl of Qiagen (http://www1.qiagen.com) PCR mastermix, and 50 ng of DNA. The following cycling conditions were used: 5 min at 95 °C, followed by 40 cycles at 95 °C for 1 min, 60 °C for 1 min, and 72 °C for 1 min, followed by 7 min at 72 °C. Each individual from the HapMap was genotyped in replicate with the Deletion and Insertion1 primers while each individual in the HGDP was genotyped one time. In addition, each of the samples, which appeared to be homozygous for the deletion, were genotyped using a second set of oligonucleotides for the insertion (Insertion2)
We genotyped 1,277 DNA samples corresponding to 270 samples from the International Hap Map project and 1,007 individuals from the CEPH HDGP [39,40]. Our analysis from the HGDP includes individuals from 51 different populations and excludes samples previously identified as duplicates [21,41,42]. Eight individuals (numbers 993, 994, 1028, 1030, 1031, 1033, 1034, and 1035) belonging to South African Bantu populations were genotyped (each was homozygous for the insertion) but were not included in the analysis due to small sample size. Individual genotypes are provided in Tables S1 and S4.
Phased SNP genotypes from HapMap Phase I were used for LD and extended haplotype analyses (http://hapmap.org). We excluded SNPs mapping within the deleted region and used PHASE version 2.1 to infer new haplotypes which included the insertion/deletion genotypes (http://www.stat.washington.edu/stephens/software.html) [43,44]. We also constructed haplotypes for insertion/deletion alleles based on the more complete Phase II genotyping data using only SNPs genotyped in all populations (Table S6).
We used an exact test of Hardy-Weinberg equilibrium for two-allele loci as implemented in version 1.2.0 of the R genetics package [45]. LD was measured by r2. FST values were calculated from population allele frequencies using an unbiased estimator [46,47]. The calculated FST values were compared with an empirical distribution defined by a collection of SNPs and small indels genotyped in the same individuals [21,22].
EHH, REHH, and EHL were calculated for the locus using SWEEP (http://www.broad.mit.edu/mpg/sweep/index.html) version 1.0 [24]. EHH is the probability that two randomly chosen chromosomes carrying the same allele at a core region are homozygous for all SNPs to a defined distance (x) from the core. REHH measures the decay of EHH at a given core genotype compared to the decay of other core haplotypes. SNPs mapping within the deleted region (five SNPs in CEU and two SNPs each in JPT and CHB and YRI) were excluded for all analyses of both the insertion and deletion cores. We measured REHH values at a marker H of 0.04, which is a measure of the observed amount of recombination and is roughly equal to a genetic distance of 0.25 cM. Observed REHH values on each side of the core were compared with REHH values calculated for each HapMap Phase I SNP. EHL is operationally defined as the sum of the genetic distance at which EHH falls to 0.5 on either side of the core [48]. This distance is sensitive to the density of markers, so SNP density was controlled by matching to the density around APOBEC3 using SWEEP (approximately one SNP every 2,500 bases). In this analysis, the core haplotype was defined as simply the insertion or deletion genotype. Comparisons were made with an empirical distribution calculated from HapMap Phase I data using two different definitions for the core region. First, we defined the core as the longest non-overlapping haplotypes containing between three and ten SNPs [49]. Secondly, we treated each SNP locus individually as a core. Resulting values were then divided into 20 bins based on the core frequency (intervals of 5%), and the values corresponding to the insertion and deletion alleles for each population were compared.
For the EHL analysis, haplotype length was measured in two ways. For the insertion, the core was placed at the center of the variant region, and EHL was calculated as the sum of the genetic distance at which EHH fell to 0.5 on the proximal and the distal sides of the core. The application of the same procedure to the deletion core results in a misleading haplotype length since the corresponding haplotype on deletion chromosomes is physically shorter due to the deletion. In order to account for this, the haplotype length in the proximal and distal directions was calculated separately by defining each breakpoint as the position of the core. This assures that the length of the extended haplotype is not inflated by the inclusion of the chromosomal segment which is actually deleted.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for the fosmid sequence is AC192820. The sequences of APOBEC3A (NM_145699.2) and APOBEC3B (NM_004900.3) were obtained from GenBank. |
10.1371/journal.pntd.0001289 | Efficacy of a Low-Cost, Inactivated Whole-Cell Oral Cholera Vaccine: Results from 3 Years of Follow-Up of a Randomized, Controlled Trial | Killed oral cholera vaccines (OCVs) have been licensed for use in developing countries, but protection conferred by licensed OCVs beyond two years of follow-up has not been demonstrated in randomized, clinical trials.
We conducted a cluster-randomized, placebo-controlled trial of a two-dose regimen of a low-cost killed whole cell OCV in residents 1 year of age and older living in 3,933 clusters in Kolkata, India. The primary endpoint was culture-proven Vibrio cholerae O1 diarrhea episodes severe enough to require treatment in a health care facility. Of the 66,900 fully dosed individuals (31,932 vaccinees and 34,968 placebo recipients), 38 vaccinees and 128 placebo-recipients developed cholera during three years of follow-up (protective efficacy 66%; one-sided 95%CI lower bound = 53%, p<0.001). Vaccine protection during the third year of follow-up was 65% (one-sided 95%CI lower bound = 44%, p<0.001). Significant protection was evident in the second year of follow-up in children vaccinated at ages 1–4 years and in the third year in older age groups.
The killed whole-cell OCV conferred significant protection that was evident in the second year of follow-up in young children and was sustained for at least three years in older age groups. Continued follow-up will be important to establish the vaccine's duration of protection.
ClinicalTrials.gov NCT00289224.
| New-generation vaccines against cholera are given orally, to stimulate intestinal immunity. An internationally available oral cholera vaccine (OCV) consists of killed vibrio whole cells together with the B subunit of cholera toxin, is safe, and protects vaccinated individuals against cholera for two years, but this vaccine has seen limited use due to its high cost. We developed a simpler, inactivated whole-cell only OCV that can be produced inexpensively and might therefore be attractive for use in developing countries, as well as for travelers from industrialized countries. We tested this new OCV in a randomized, controlled field trial that enrolled 69,328 individuals aged one year and older living in urban slums of Kolkata, India. At three years of follow-up after receiving at two-dose regimen of this OCV, the vaccinated population experienced 66% protection against all episodes of cholera occurring during the three years, and 65% protection against episodes occurring during the third year. Significant protection was evident in the second year in children vaccinated at ages 1–4 years and in the third year in persons vaccinated at ages of five years and older. Follow-up of the study population will continue for five years to ascertain the duration of vaccine protection.
| Cholera is a major global public health problem, causing both epidemic and endemic disease. Although conventional, injectable cholera vaccines have been abandoned as public health tools, modern oral cholera vaccines (OCVs) have been found to be safe and effective [1]. A recently revised World Health Organization (WHO) position paper expands the potential role of vaccination as a preventive tool against both endemic and epidemic cholera [2].
There are two licensed OCVs currently available: one containing cholera toxin B subunit (BS) and killed cholera whole cells (WC), which is licensed in over 50 countries, and the other containing only killed WC, which is licensed in India and Vietnam [1], [3]. A field trial of BS-WC vaccine in Bangladesh found that a three-dose regimen was safe and conferred high grade (85%) short-term protection against cholera; protection was clearly evident throughout the first two years of follow-up, but markedly declined in the third year [4]. An advantage of the WC-only vaccine is its low cost, now at $1.85 per dose to the public sector. We conducted a placebo-controlled, randomized trial to assess the safety and protection conferred by a two-dose regimen of the WC-only vaccine against cholera severe enough to warrant solicitation of medical care. An initial analysis of an ongoing field trial in Kolkata of the WC-only vaccine found a two-dose regimen to be safe and to confer 67% protective efficacy against cholera at two years of follow-up [5]. Here we present results from the third year of follow-up of the Kolkata trial.
Details on the study site, study agents, study procedures, and assembly of subjects for this parallel, randomized trial were previously reported [5].The study was performed in a cholera-endemic area in the slums of Kolkata, encompassing a population of ∼109,000. Residents who were at least one year of age and were not pregnant were eligible to participate in the study.
Each dose of the killed WC OCV (Shanchol™, Shantha Biotechnics), contains inactivated Vibrio cholerae 01 cells representing the El Tor and classical biotypes and the Inaba and Ogawa serotypes, as well as serogroup 0139 cells. Vials containing identical–appearing heat-killed Escherichia coli K12 cells were used as placebo. Single-dose vials were labeled with one of four letter codes, two for vaccine and two for placebo. Project staff and study subjects were unaware of the identities of the codes. Participants were randomly assigned, by residential dwelling, to vaccine or placebo groups. Randomization was done before enrollment by an independent statistician (AD), using a random number table. Dwellings were randomized in blocks of 4, corresponding to the 4 code letters used to label vaccine and placebo, within strata defined by the ward of residence and the number of residents in the dwelling (six strata). Each agent was given as a two-dose regimen with an inter-dose interval of at least 14 days. Enrollment and administration of the pre-assigned agents was performed by dosing teams in vaccination centers serving the population. Codes were kept secretly at Shantha Biotechnics and the International Vaccine Institute by staff who were not involved with the trial. The agents were administered in two rounds in 2006: from July 27 to August 13 and from August 27 to September 10, 2006.
Surveillance was performed in nine community clinics established for the trial and in two hospitals serving the study population. Study physicians completed structured study forms to obtain pertinent clinical information, and fecal specimens were tested for V. cholerae as previously described, including identification of 01 and 0139 serogroups with agglutination tests. Biotype was ascertained for all 01 isolates, and the biotype of the cholera toxin genetically encoded was identified as previously described [5], [6]. Confirmation that the subject had indeed visited the treatment site on the date of the visit was assessed through domiciliary visits for all patients whose samples yielded V. cholerae O1 or O139. A diarrheal visit was defined as having, in the 24 hours before presentation: 3 or more loose or liquid stools; or, at least 1 loose or liquid stool with blood; or, if 1–2 or indeterminate number of loose or liquid stools were reported, the patient must have exhibited at least some evidence of dehydration, using WHO criteria [7]. The onset of a diarrheal visit was the day on which the patient first reported loose or liquid stools. Diarrheal visits for which the date of onset was less than or equal to 7 days from the date of discharge for the previous visit were grouped into the same diarrheal episode. The onset of a diarrheal episode was the onset of the first diarrheal visit of the episode. The primary endpoint, a cholera episode, was defined as a diarrheal episode in which no component visit was described as bloody, in which a fecal specimen yielded V. cholerae O1, and a domiciliary check confirmed that the subject had indeed visited the treatment center for diarrhea on the recorded date of presentation. Demographic surveillance for migrations and deaths among the study population was maintained during the three years of follow-up, and verbal autopsies were done for identified deaths.
The sample size calculation for the trial was previously reported [5]. Prior to the analysis, data were frozen, and a detailed analytic plan was approved by the data and safety monitoring board. The primary analysis was a per-protocol analysis of vaccine protection among subjects who completely ingested two doses of an agent with the assigned treatment code, and included first cholera episodes with onsets between 14 days and 1,095 days after receipt of the second dose. A modified intention-to-treat analysis was done for all individuals who received at least one dose of an agent regardless of the amount ingested, and regardless of whether the agent received was as assigned. First cholera episodes that began from 1 to 1,095 days after the intake of the first dose were included in this analysis. All analyses were conducted and interpreted prior to unblinding of the codes.
Survival analyses were used to calculate vaccine protective efficacy with measurements of the time to the first episode of cholera, censoring the follow-up of individuals who died or migrated out [8]. Kaplan-Meier curves were constructed for descriptive analyses. We also fitted unadjusted and adjusted Cox proportional hazards regression models, after verifying that the proportionality assumptions were fulfilled for all independent variables [9]–[11]. We estimated the hazard ratios by exponentiating the coefficient for the vaccine variable in these models and calculated the vaccine efficacy (PE) as : (1- hazard ratio)×100%. To estimate P values and confidence intervals (CI) for the hazard ratio, we used the standard errors for the coefficients. Robust sandwich variance estimates were used to account for the design effect of cluster randomization, allowing inferences for vaccine efficacy at the individual level [12].
Variables used for stratified randomization as well as baseline variables that were found to be significantly associated with time to event at p<0.10 in bivariate analyses were candidates as independent variables in the final models assessing vaccine efficacy. To avoid overfitting the models, we used a backward elimination algorithm to select independent variables in addition to the vaccination variable. Vaccine efficacy was evaluated in different subgroups that were defined prior to analyses. Heterogeneity of vaccine protection was assessed in these subgroups by analyzing interaction terms in the models. All P values and confidence intervals(CI's) were calculated as one-sided except for assessing heterogeneity of vaccine efficacy in different subgroups, for which stochastic estimates were two-sided. An interim analysis at 2 years of follow-up, using the Haybittle-Peto rule, set the P value for statistical significance for the primary analysis of PE at P<.01 [13]. Because the three-year analysis was the major objective of the trial, all analyses reported in this paper were evaluated at a threshold of P<.0.05, with corresponding one-sided 95% CIs. All statistical analyses were performed using SAS version 9.1. While the initial plan for surveillance was only for three years, follow-up is ongoing to assess the duration of protection up to five years post-vaccination.
The study protocol was approved by the ethics committee of the National Institute of Cholera and Enteric Diseases, the Health Ministry Screening Committee of India and the International Vaccine Institute Institutional Review Board. Written informed consent was obtained from older residents and from the guardians of residents aged 1 to 17 years of age. Additional written assent was obtained from residents aged 12 to 17 years. An independent data and safety monitoring board reviewed the study protocol, assessed serious adverse events, and approved freezing of data and the analytical plan prior to starting the analysis.
The study was prospectively registered at ClinicalTrials.gov (NCT00289224). In the per-protocol analysis, there were 1,721 clusters and 31,932 participants in the vaccine group and 1,757 clusters and 34,968 participants in the placebo group (Figure 1). In the intention-to-treat analysis, there were 1,727 clusters and 33,127 participants in the vaccine group and 1,768 clusters and 36,202 participants in the placebo group. 4,252 and 4,661 participants in the vaccine and placebo groups, respectively, died or migrated out of the study area after the second dose. As previously reported, individual-level and cluster-level baseline characteristics were similar for vaccinees and placebo recipients [5]. There were no substantive imbalances in baseline variables among participants in each arm who were excluded or lost to follow-up.
All detected cholera episodes were due to V. cholerae 01, El Tor biotype that genetically encoded classical biotype cholera toxin. As shown in Table 1, 38 and 128 episodes were detected in per-protocol analysis of the vaccine and placebo groups, respectively (adjusted cumulative protective efficacy 66%, one-sided 95% CI-lower bound = 53%, p<0.001). In the intention-to-treat analysis, there were 49 and 137 cholera episodes in the vaccine and placebo groups (adjusted cumulative PE = 61%, one-sided 95% CI lower bound = 47%, p<0.001). Survival curves for each analysis are presented in Figure 2. Most of the isolates were Ogawa serotype. In the per-protocol analysis of protective efficacy against Ogawa cholera, there were 34 and 118 episodes, respectively, in the vaccine and placebo groups (adjusted PE = 68%, one-sided 95% CI lower bound = 54%, p = <0.001). Inaba serotype was detected in only 4 and 10 episodes in the vaccine and placebo groups, respectively (unadjusted PE = 56%, one-sided 95% CI lower bound = −14%, p = .08). There were no deaths due to cholera identified in the the treatment centers, or due to acute watery diarrhea in verbal autopsies among study participants.
Table 1 presents the per protocol analysis by year of follow-up and by age at vaccination. Cumulative three-year vaccine efficacy was highest for children vaccinated at ages 5–14 years (adjusted PE 88%, one-sided 95% CI lower bound = 71%, P<.001), intermediate for persons vaccinated at older ages (61%, one-sided 95% CI lower bound = 37%, P<.001), and lowest for children vaccinated at ages 1–4 years (adjusted PE 43%, one-sided 95% lower bound = 7%, P = .03), and differed significantly (P = .02, two-sided) among the three age groups. Protection of all age groups was 65% during the third year of follow-up (one-sided 95% CI lower boundary = 44%, P<.001), and showed no evidence of decline over time (P = .24, two-sided, for comparison of PE in years 1, 2, and 3 of follow-up). Variations in vaccine protection for each age group, by year of follow-up, did not reach statistical significance (two-tailed P values for comparison of PE in years 1,2, and 3 of follow-up in the 1–4 year, 5–14 year, and ≥15 year age groups were .11, .87, and .98, respectively). Vaccine protection was significant during the third year in the 5–14 year and ≥15 year age groups, but was significant only in the second year in follow-up of younger persons.
Our findings demonstrate that a two-dose regimen of the killed, WC OCV conferred protection of 66% protection during the three years following vaccination. Vaccine protection was clearly evident in the third year of follow-up in persons vaccinated at ages five years and older and during the second year in children vaccinated at 1–4 years of age. Due to small numbers of outcomes during the third year, however, further follow-up will be required to assess the duration of protection in the youngest age group. Protection was clearly evident against El Tor Ogawa, and suggestive against El Tor Inaba, though the latter analysis was limited by a small number of outcome events. Of note, all episodes of cholera were due to V. cholerae 01 that manifested the El Tor phenotype but genetically encoded classical biotype cholera toxin, a hybrid strain that now accounts for nearly all cholera cases in many parts of both Africa and Asia and that may be associated with cholera of increased severity [6], [14], [15].
An apparently counterintuitive finding was that vaccine protection was lower in the first year of follow-up than in the subsequent two years. However, the most likely explanation for this finding is chance variation, as there were no significant differences in estimates of vaccine protection, either for all age groups combined or for the <5 year , 5–14 year, and ≥15 year age groups individually, during the three years of follow-up.
Comparing the results of different vaccines tested in different trials provides less conclusive evidence of their comparative efficacy than head-to-head comparisons of vaccines in the same trial. Nevertheless, it is of interest to contrast the long-term results for the killed WC OCV studied in this trial with those for the killed BS-WC OCV tested in three doses in Bangladesh in the 1980s, the only evaluation of BS-WC with long-term follow-up [4]. In contrast to the trial of killed WC OCV in Kolkata, which demonstrated efficacy during the third year of follow-up, BS-WC vaccine's protection against cholera in Bangladesh was significant only during the first two years of follow-up. Of interest, protection by BS-WC vaccine in the Bangladesh trial was also lowest and of shortest duration in the under-five age group, an observation that has been attributed to the lower level of pre-existing, anti-cholera immunity in this group, owing to less exposure to natural cholera infections in the youngest group.
It should be emphasized, however, that long-term protection is only one consideration for the use of an OCV. Enhanced short-term protection may be a distinct advantage when considering the use of a vaccine in self-limited outbreaks of cholera. In this respect, the BS-WC OCV has been shown to confer 85% protection lasting 4–6 months after dosing [16], Another advantage of BS-WC OCV is its ability to confer cross-protection against LT-producing enterotoxigenic Escherichia coli diarrhea for several months after dosing [17].
The potential of the killed WC OCV tested in this study for use in control of endemic and epidemic cholera is substantial. However, much remains to be done. The study remains blinded and surveillance will continue to assess the duration of protection provided by the vaccine up to five years after dosing. Increasing access to this vaccine is important, not only in India, where it is currently licensed, but also in other cholera-endemic countries. Access should be increased in the near future by WHO prequalification of the vaccine so that it may be purchased by UN agencies for use in other countries for disease control.
|
10.1371/journal.pgen.1002323 | Positional Cloning of a Type 2 Diabetes Quantitative Trait Locus; Tomosyn-2, a Negative Regulator of Insulin Secretion | We previously mapped a type 2 diabetes (T2D) locus on chromosome 16 (Chr 16) in an F2 intercross from the BTBR T (+) tf (BTBR) Lepob/ob and C57BL/6 (B6) Lepob/ob mouse strains. Introgression of BTBR Chr 16 into B6 mice resulted in a consomic mouse with reduced fasting plasma insulin and elevated glucose levels. We derived a panel of sub-congenic mice and narrowed the diabetes susceptibility locus to a 1.6 Mb region. Introgression of this 1.6 Mb fragment of the BTBR Chr 16 into lean B6 mice (B6.16BT36–38) replicated the phenotypes of the consomic mice. Pancreatic islets from the B6.16BT36–38 mice were defective in the second phase of the insulin secretion, suggesting that the 1.6 Mb region encodes a regulator of insulin secretion. Within this region, syntaxin-binding protein 5-like (Stxbp5l) or tomosyn-2 was the only gene with an expression difference and a non-synonymous coding single nucleotide polymorphism (SNP) between the B6 and BTBR alleles. Overexpression of the b-tomosyn-2 isoform in the pancreatic β-cell line, INS1 (832/13), resulted in an inhibition of insulin secretion in response to 3 mM 8-bromo cAMP at 7 mM glucose. In vitro binding experiments showed that tomosyn-2 binds recombinant syntaxin-1A and syntaxin-4, key proteins that are involved in insulin secretion via formation of the SNARE complex. The B6 form of tomosyn-2 is more susceptible to proteasomal degradation than the BTBR form, establishing a functional role for the coding SNP in tomosyn-2. We conclude that tomosyn-2 is the major gene responsible for the T2D Chr 16 quantitative trait locus (QTL) we mapped in our mouse cross. Our findings suggest that tomosyn-2 is a key negative regulator of insulin secretion.
| Humans carry many genetic variants that confer small effects on metabolic traits relevant to type 2 diabetes. These effects are amplified by environmental stressors like obesity. We used morbid obesity as a sensitizer to identify genes that contribute to the diabetes susceptibility of the BTBR mouse strain. Using mapping and breeding strategies, we were able to narrow a genetic region to one containing just 13 genes. One of these genes, tomosyn-2, emerged as a prime candidate. Our functional studies showed that tomosyn-2 is an inhibitor of insulin secretion, and it binds to the proteins involved in the fusion of insulin containing granules with the plasma membrane. We found a coding mutation and demonstrated that this mutation affects the stability of the protein product. Our work with Tomosyn-2 provides new insights into the regulation of insulin secretion and emphasizes that negative regulation is critical for avoiding insulin-induced hypoglycemia.
| Genetic factors are estimated to contribute approximately 50% towards the risk of developing type 2 diabetes (T2D) [1]. Recent genome-wide association studies have identified a number of “diabetes genes”, gene loci that act in an additive manner and conspire with obesity to augment the risk of T2D. Nearly all of these genes affect β-cell function and/or the maintenance of β-cell mass. Thus, it appears that diet and obesity place demands on β-cells for insulin by causing insulin resistance, but the genetic bottleneck that can lead to T2D involves genes that affect the capacity of pancreatic β-cells to meet the increased insulin demand.
Mouse intercrosses provide high power to detect linkage of gene loci to physiological traits related to obesity and diabetes. The mapping resolution of these intercrosses is typically not fine enough to identify individual genes. However, by generating panels of congenic strains carrying meiotic recombinations within disease loci, it is possible to map the genes underlying those loci with high resolution. We recently reported the positional cloning of a T2D gene, Sorcs1, to sub-genetic resolution using this approach [2]. Several other genes have been identified using a similar approach; Zfp69, which encodes a transcription factor involved in regulating glucose levels, was identified as the gene underlying a diabetes susceptibility locus on mouse chromosome 4 [3]. Similarly, Lisch-like was identified as the gene underlying a T2D locus on chromosome 1 and was shown to be involved in regulating β-cell mass and plasma glucose levels [4].
Borrowing from microbial genetics, mouse genetic studies employ a powerful tool for increasing the sensitivity to detect heritable phenotypes. This involves sensitized screens wherein a severe stressor provokes phenotypes that would otherwise be silent. The stressor need not be a normal feature in human disease pathogenesis to evoke phenotypes of great relevance to disease. For example, the apoE-deficient mouse is the most widely used animal model of atherosclerosis even though apoE deficiency is extremely rare in humans [5]. Similarly, a mutation in the Leptin gene (Lepob) promotes morbid obesity in mice (Lepob/ob) and evokes dysregulation of many pathways, enabling a greater understanding of their regulatory mechanisms [6], [7].
Using the Lepob mutation as a stressor, we found that the BTBR T (+) (BTBR) mouse strain develops severe T2D, whereas the C57BL/6 (B6) strain has moderate hyperglycemia and expands its β-cell mass [8], [9]. In an F2 intercross derived from these two strains, we identified a T2D susceptibility locus on chromosome 16 (Chr 16) [9]. In the present study, we developed a panel of congenic strains that enabled us to narrow this locus to just 0.94 Mb. Lean congenic mice that contain this genomic region derived from the BTBR strain have elevated glucose and reduced insulin levels. Islets from these mice show deficiencies in insulin secretion. Within this small interval, we identified a novel diabetes susceptibility gene, Syntaxin binding protein 5 like (Stxbp5l), also known as Tomosyn-2. We showed that the tomosyn-2 protein is an inhibitor of insulin secretion.
We previously identified a fasting plasma glucose locus on Chr 16 from a Lepob/ob F2 intercross derived from the B6 and BTBR mouse strains [9]. This locus acts in a fully dominant fashion on plasma glucose and a semi-dominant fashion on fasting plasma insulin [9]. The LOD peak on Chr 16 of the fasting glucose locus from the F2 intercross is located at approximately 36–38 Mb (Figure 1A). To determine if the Chr 16 locus could act autonomously to affect T2D susceptibility, we derived a chromosome substitution (i.e. consomic) mouse strain by introgression of Chr 16 from BTBR into B6 ob/ob mice (B6.16BT Lepob/ob). The fasting plasma glucose levels of the resulting consomic mice were significantly elevated at 4, 6, 8, and 10 weeks compared to control (B6.16B6 Lepob/ob) mice and accounted for a major proportion of the plasma glucose phenotype of the parental BTBR strain. Fasting plasma insulin levels were reduced at 8 and 10-weeks in B6.16BT Lepob/ob mice compared to B6.16B6 Lepob/ob mice (Figure 1B and 1C). These data suggested that the hyperglycemia caused by BTBR Chr 16 substitution is due to reduction in insulin levels. The data also indicate that the locus on Chr 16 acts autonomously (i.e. in the absence of BTBR alleles on other chromosomes) to affect glucose and insulin levels.
To assess whether the B6.16BT Lepob/ob mice have a defect in insulin secretion, we isolated pancreatic islets from 10-week old B6.16BT Lepob/ob mice and measured fractional insulin secretion in response to high glucose (16.7 mM). We observed a ∼50% reduction in fractional insulin secretion in the B6.16BT Lepob/ob islets relative to control mice (B6.16B6 Lepob/ob) (Figure 2, left graph).
To avoid the metabolic complexities that are attributed to the leptin mutation in the Lepob/ob mice [10], we performed experiments in lean mice. Islets isolated from the congenic B6.16BT and B6.16B6 lean mice were treated with 8-bromo cAMP (3 mM) at sub-maximal glucose (11.1 mM); this combination of secretagogues was used for phenotyping lean congenic mice because it evoked more insulin secretion from lean islets than glucose alone. We observed ∼40% reduction in fractional insulin secretion in islets isolated from the lean B6.16BT mice relative to the lean control B6.16B6 mice (Figure 2, right graph). The data show that the insulin secretion defect, although initially mapped in a screen of F2 mice sensitized by the Lepob mutation, manifests itself independent of leptin deficiency.
To investigate the region of the BTBR Chr 16 that confers the insulin secretion defect, a panel of lean congenic mouse strains was generated from the B6.16BT mice, each containing a small introgressed region from the BTBR Chr 16 in the B6 background (Figure 3, left panel). The B6/BTBR boundaries for each congenic strain were determined via microsatellite marker, single nucleotide polymorphism (SNP) sequencing or deletion/insertion polymorphism (DIP) sequencing (Dataset S1). By phenotyping each strain, we were able to fine-map the location of the gene responsible for the insulin secretion defect.
Islets were isolated from each lean congenic mouse strain and fractional insulin secretion was determined in the presence of 3 mM 8-bromo cAMP at sub-maximal 11 mM glucose. The islets isolated from the lean congenic mice that carry the 2 Mb Chr 16 derived from the BTBR strain (B6.16BT36–38) were defective in insulin secretion when compared to islets isolated from congenic mice B6.16B6, B6.16BT37–55, or B6.16BT24–37 (Figure 3, right panel). The overlapping region of BTBR Chr 16 represented by congenic strains, B6.16 BT36–38 and B6.16 BT37–55 enabled us to further narrow the region to 1.6 Mb. Islets isolated from congenic mice, B6.16BT24–55, B6.16BT24–47, or B6.16BT24–38, which contained the 1.6 Mb region derived from the BTBR strain, also displayed a reduced level of insulin secretion, indicating that this 1.6 Mb region contains a gene responsible for regulating insulin secretion.
We next determined the effect of introgression of the BT36–38 region of Chr 16 into B6 mice on susceptibility to obesity-induced diabetes. Plasma insulin and glucose levels were determined in random-fed 10-week B6.16BT36–38 and control B6.16B6 Lepob/ob congenic mice. We observed a ∼40% reduction in plasma insulin levels in the B6.16BT36–38 Lepob/ob compared to B6.16B6 Lepob/ob mice (Figure 4A). The reduction in plasma insulin was accompanied by an increase in plasma glucose by ∼100 mg/dL (Figure 4C).
Although not nearly as dramatic as in the Lepob/ob mice, lean congenic mice with the 36–38 Mb BTBR insert had a significant reduction in plasma insulin and increase in plasma glucose (Figure 4B and 4D, respectively). Detection of this very small rise in plasma glucose required a very large sample size (n = 80) to achieve statistical significance. Clearly, we would not have found this modest phenotype in a screen of lean mice, showing that severe stressors like the Lepob mutation are required to identify subtle allelic variation in QTLs that contribute to T2D risk.
Insulin secretion from pancreatic β-cells is biphasic. The first phase represents a brief but rapid secretion from pre-docked insulin granules in response to an initial depolarization of the plasma membrane. Non-nutrient secretagogues like KCl predominantly invoke the first phase of insulin secretion. The second phase of insulin secretion is associated with metabolic signals derived from the metabolism of fuel-based insulin secretagogues like glucose. Glucose affects both the first and second phase of insulin secretion. Briefly, glucose oxidation increases the ATP/ADP ratio, resulting in the closure of ATP-sensitive KATP channels. This causes depolarization of the plasma membrane and influx of Ca2+ via L-type voltage-dependent calcium channels. Glucose promotes the second phase of insulin secretion without causing a further increase in intracellular Ca2+ levels. Diazoxide inhibits closure of the K+ channels, therefore adding this drug along with 16.7 mM glucose will result only in the glucose-induced second phase of insulin secretion and eliminates any glucose induction of the first phase of insulin secretion [11]–[13].
To determine which phase of insulin secretion is defective in our lean congenic B6.16BT36–38 mice, we carried out perifusion studies of isolated islets. Islets were perifused in Krebs-Henseleit Ringer bicarbonate (KRB) buffer at the rate of 1 ml/min. The perfusate was sampled every 30 sec, and the secreted insulin was measured by ELISA. After an initial 60 min equilibration period in KRB containing 1.7 mM glucose, islets were perifused for 10 min in KRB containing 40 mM KCl and 250 µM diazoxide to elicit first phase of insulin secretion. After 10 min, the islets were perifused for an additional 30 min in KRB containing 16.7 mM glucose with 40 mM KCl and 250 µM diazoxide to evoke the second phase of insulin secretion. The peak of the first phase of insulin secretion from B6.16B6 islets was observed within 1–2 min of KCl treatment. Following the first peak, the more sustained second phase of insulin secretion was observed for an additional 30 min, mimicking the well-studied biphasic kinetics of insulin secretion [14].
Islets from B6.16BT36–38 lean mice secreted ∼40% less insulin during the second phase than islets from the B6.16 B6 mice, as determined by calculating the area under the curve (AUC) (Figure 5A, 5B). There was also a small, but statistically significant reduction of first-phase insulin secretion (p = 0.044) (Figure 5A, 5B).
To complement the perifusion experiments, static insulin secretion experiments were performed in islets isolated from the B6.16BT36–38 and control B6.16B6 lean mice. Isolated islets were incubated for 45 min in KRB containing 1.7 mM glucose. Following a 45-min incubation, the islets were treated with 40 mM KCl in KRB containing 1.7 mM glucose. No difference in fractional insulin secretion was observed in response to KCl (Figure 6C, left panel). However, we observed a significant decrease in fractional insulin secretion between islets from the B6.16BT36–38 lean mice and those from the control B6.16B6 lean mice in response to 15 mM arginine, 3 mM 8-bromo cAMP in KRB containing 11 mM glucose, and 16.7 mM glucose alone (Figure 5C, middle and right panels).
To further narrow the 1.6 Mb region of BTBR Chr 16 responsible for the phenotype, we used Agilent's SureSelect Target Enrichment to capture DNA from 35.35 Mb to 38.65 Mb on mouse Chr 16. 55,336 RNA baits were designed using the Agilent eArray and were used to enrich for our region from tail DNA of the B6, BTBR, B6.16BT36–38, B6.16BT24–37and B6.16BT24–38 mice. DNA was sequenced by Next Generation Sequencing using an Illumina GA IIx sequencer at the UW-Madison Biotechnology Center. Using CLC Genomics 4.0.3 Software, we were able to identify 470 SNPs; 3 non-synonymous coding SNPs and 46 DIPs between the B6 and BTBR DNA (Figure 6A). 83 SNPs and 8 DIPs were further confirmed by manual base reading to confirm the accuracy of the software (listed in Dataset S2). Using this sequence and known overlapping regions derived from the BTBR strain in the sub-congenic strains exhibiting normal insulin secretion (B6.16BT24–37, B6.16BT37–55), we were able to narrow the region responsible for the insulin secretion defect to 0.94 Mb containing 13 genes (Figure 6B).
To identify the gene(s) responsible for the insulin secretion defect, each candidate gene in the 0.94 Mb region was scored for the difference in mRNA abundance between the islets B6.16B6 and B6.16BT36–38 islets, the presence of non-synonymous coding SNPs, and similarity to a protein that have a functional role in exocytosis. Tomosyn-2 or Stxbp5l (syntaxin binding protein 5-like) quickly emerged as the top candidate gene. The mRNA abundance of tomosyn-2 was elevated 2.6 fold in the B6.16BT36–38 lean mice compared to control B6.16B6 mice (Figure 7A). Tomosyn-2 has a coding SNP (Ser-912→Leu). Eight other SNPs were also identified in the introns and additional SNPs were identified in the intergenic regions 5′ and 3′ of the gene (Dataset S1, Table S1). The tomosyn-2 protein shares 95% identity in the C-terminal soluble NSF (N-ethylmaleimide-sensitive factor) attachment protein receptor (SNARE) domain with several syntaxin-binding proteins. Finally, a related protein, tomosyn-1, has been shown to inhibit insulin secretion [15].
To understand the role of tomosyn-2 in the regulation of insulin secretion, the expression of tomosyn-2 was determined in key metabolic tissues; islet, liver, brain, cerebellum, kidney, adrenal, adipose (perigonadal), heart, skeletal muscle (gastrocnemius, soleus, and quadriceps) of the lean B6.16BT36–38 and B6.16B6 mice. The mRNA expression of tomosyn-2 in islets of the lean B6.16BT36–38 mice was ∼2.6-fold higher than in islets from the B6.16B6 mice (Figure 7A). No allele-dependent difference in the tomosyn-2 expression was observed in liver, brain, cerebellum, kidney, adrenal, gastrocnemius, adipose, heart, soleus, and quadriceps between the lean B6.16BT36–38 and B6.16B6 mice.
Four tomosyn-2 isoforms have been identified in mice: xb-, b, s, and m-tomosyn-2. We determined the relative expression of the tomosyn-2 isoforms in islets of the lean B6.16BT36–38 and B6.16B6 mice. We found that the b-tomosyn-2 isoform is the most abundant isoform in mouse islets. This was confirmed by RT-PCR with a primer pair that simultaneously amplified all of the tomosyn-2 isoforms (data not shown). The relative expression of b-tomosyn-2 and s-tomosyn-2 mRNA was ∼6-fold higher in islets of the lean B6.16BT36–38 mice than in lean B6.16B6 mice (Figure 7B). We observed no significant difference between the two-congenic mouse strains in the expression of xb- and m-tomosyn-2 isoforms. Together, the data indicate that increased expression of tomosyn-2 may be responsible for the insulin secretion defect observed in the lean B6.16BT36–38 mice.
To investigate the role of tomosyn-2 in insulin secretion, we investigated the effect of overexpressing b-tomosyn-2 in the pancreatic β-cell line, INS1 (832/13). The cells were transfected with GFP or b-tomosyn-2 expression plasmids. After 36 h, the cells were incubated in KRB containing 1.5 mM glucose for 2 h. Following the low glucose incubation, the cells were incubated for additional 10 min or 2 h in 3 mM 8-bromo cAMP at 7 mM glucose. Overexpressing b-tomosyn-2 decreased insulin secretion by ∼40% vs. GFP expressing cells at both 10 min and 2 h (Figure 8A). No inhibition in fractional insulin secretion was observed at low glucose (1.5 mM) (data not shown).
To determine if b-tomosyn-2 binds to syntaxin-1A and syntaxin-4, key t-SNARE proteins involved in the fusion of insulin granules to the plasma membrane, in vitro binding experiments were performed using GST fused syntaxin-1A and syntaxin-4 recombinant proteins (soluble, lacking transmembrane domains) by pull-down assays using glutathione beads. All isoforms of tomosyn-2 bound to GST-syntaxin-1A and GST-syntaxin-4 (Figure 8B). The quantitation for the amount of bound tomosyn-2 isoforms as a fraction of total is shown in Figure 8C. Tomosyn-1 was used as a positive control for binding. The GST tag did not pull down tomosyn-1 or tomosyn-2, confirming that the interaction between tomosyn-2 and syntaixn-1A and -4 is specific. Together, these data suggests that the mechanism by which b-tomosyn-2 inhibits insulin secretion involves binding to the syntaxin proteins. This suggests the possibility that tomosyn-2, like tomosyn-1, inhibits insulin secretion by preventing the binding of VAMP2 to syntaxin-1A and syntaxin-4.
We have shown that tomosyn-2 is a negative regulator of insulin secretion and also binds to syntaxin-1A and syntaxin-4. To investigate the possibility that the serine-912leucine SNP in tomosyn-2 affects its stability, HEK293T cells were transfected with empty vector (mock), b-tomosyn-2 (Serine-912), or b-tomosyn-2 (Leucine-912). After 16 h, the cells were treated with or without the proteasomal inhibitor, MG132 (100 µM) for 6 h. The MG132 treatment rescued the B6 allelic form of the protein, b-tomosyn-2 (serine-912), from proteasomal degradation by ∼50% (Figure 9A and 9B). However, the BTBR allelic form of the protein, b-tomosyn-2 (leucine-912) was not resistant to MG132 treatment, suggesting that an increased stability of the tomosyn-2 protein might be responsible for the attenuation in insulin secretion from islets of the BTBR mice.
Our pursuit of genes conferring susceptibility to obesity-induced T2D focuses on two mouse strains that differ in diabetes susceptibility. BTBR mice, when made obese with the Leptinob mutation, are susceptible to T2D, whereas B6 mice with the same mutation are relatively diabetes resistant [16], [17]. The diabetes susceptibility of the obese BTBR mice has multiple causes, including an insulin secretion defect and a failure to increase β-cell mass. As early as 4 weeks of age, islets from the obese B6, but not BTBR mice have an increase in expression of a module of cell cycle genes whereas the obese BTBR mice fail to induce the expression of this module [16].
Through genetic mapping in an F2 intercross, we identified a strong QTL on Chr 16 wherein the BTBR allele is linked to increased glucose levels. In complex trait genetics, it is often the case that gene loci do not act autonomously, but must act along with specific alleles at other loci to exert their phenotypic effects [18]–[20]. To determine if the Chr 16 locus acts autonomously, we derived a chromosome substitution strain. Substitution of Chr 16 in the B6 strain with Chr 16 from the BTBR strain led to a ∼100 mg/dl increase in glucose and a ∼50% decrease in plasma insulin. This established that a locus on Chr 16 acts autonomously and is sufficient to account for a major part of the diabetes phenotype of the BTBR mouse strain.
QTL mapping in an F2 intercross does not provide the resolution required to identify individual genes. To narrow the interval, we derived a panel of interval-specific congenic strains. The strains were phenotyped on the basis of insulin secretion from isolated islets, a far more robust phenotype than fasting glucose or insulin levels. This enabled us to narrow the position of the QTL to <1 Mb, containing just thirteen genes. Of these thirteen genes, tomosyn-2 was the only gene that had both altered mRNA abundance and a coding SNP (S912L).
Recent studies by Williams et al. show that tomosyn-2 inhibits exocytosis in PC12 cells [21]. Our experiments establish a role for tomosyn-2 in insulin secretion. When we overexpressed tomosyn-2 in the pancreatic β-cell line INS1 (832/13), insulin secretion in response to 8-bromo cAMP at sub-maximal glucose was attenuated. In vitro GST-pull-down experiments showed that tomosyn-2 has the ability to bind t-SNARE proteins, syntaxin-1A and syntaxin-4. Syntaxin-1A is involved in the first phase and syntaxin-4 is involved in regulating both the first and second phase of insulin secretion [22]. These results establish that tomosyn-2, like its homologue, tomosyn-1, inhibits insulin secretion [23]. The fact that allelic variation in tomosyn-2 is sufficient to produce this phenotype suggests that tomosyn-1 cannot compensate for this deficiency, implying that their functions may not completely overlap.
The mRNA abundance of tomosyn-2 was increased in the congenic mouse strains expressing the BTBR allele. It is difficult to determine if this difference in expression level is sufficient to produce the difference in insulin secretion in islets of the two mouse strains because it would require accurately titrating the gene dosage (and the amount of protein product) in a null background.
Sequence analysis revealed a SNP in tomosyn-2 (S912L). We utilized a recent study showing that the proteasome inhibitor MG132 increases the abundance of tomosyn-2 [21]. We found that the S912L SNP abolishes the ability of MG132 to rescue tomosyn-2 from proteasomal degradation, thus establishing a functional role for the S912L SNP. Therefore, the decreased insulin secretion associated with the BTBR allele might be the result of increased stability of the tomosyn-2 protein as a consequence of the SNP at amino acid 912. We also indentified SNPs in the introns and the intergenic regions 5′ and 3′ of the tomosyn-2 gene (Figure 7). The intronic SNPs may regulate the stability of the tomosyn-2 mRNA and the intergenic SNPs might affect the level of transcription of the gene.
To investigate the role of the S912L SNP in pancreatic islets, we conducted perifusion experiments. Our perifusion experiments demonstrated that islets from the B6.16BT36–38 congenic mice were defective in the 2nd phase of insulin secretion. Our results suggest that tomosyn-2 is likely to be responsive to metabolic signals. With static incubation experiments, we tested various insulin secretagogues. Islets with the BTBR allele of tomosyn-2 were clearly less responsive to cAMP or arginine at sub-maximal glucose. These secretagogues are involved in both phases of insulin secretion, but the exact mechanisms by which they stimulate the 2nd phase of insulin secretion are not fully understood. The involvement of tomosyn-2 provides a plausible new target for the actions of these secretagogues.
We show that tomosyn-2, similar to tomosyn-1, binds to syntaxin-1A and syntaxin-4 [24]–[27]. Recent studies suggest that the binding to syntaxin is necessary but not sufficient for tomosyn-2′s inhibition of insulin secretion [21].
Our studies suggest that tomosyn-2 imposes a critical brake on insulin secretion. This is particularly important during fasting when inappropriate insulin secretion could cause life-threatening hypoglycemia. We hypothesize that under fasting conditions when glucose levels are low, tomosyn-2 blocks exocytosis and prevents hypoglycemia.
In mice, the two tomosyn genes, tomosyn-1 and tomosyn-2, encode seven alternatively spliced variants [23]. Tomosyn-1 contains three distinct isoforms (s, m, and b), whereas tomosyn-2 has four different spliced variants (s, m, b, and xb). The spliced exons encode the hypervariable region (HVR), which in tomosyn-1 has been shown to be subject to SUMOylation and PKA-mediated phosphorylation [21], [28]. The amino acid sequences of tomosyn-1 and tomosyn-2 are quite similar in the N-terminal WD40 repeats and C-terminal VAMP-like domain (VLD) [23]. Tomosyn-1 was identified in neurons as a syntaxin-1-binding protein that sequesters t-SNAREs on the plasma membrane by forming a “dead end”, nonfusogenic SNARE complex, resulting in inhibition of the formation of the SNARE complex [24], [26]. Deletion of the tomosyn-1 gene in C.elegans or in mice resulted in enhanced asynchronous neurotransmitter release [29], [30]. Gain of function studies demonstrated that tomosyn-1 is responsible for inhibiting exocytosis of dense core granules in primary adrenal chromaffin cells [27], PC12 cells [25], and pancreatic β-cells [15]. Moreover, in vitro biochemical evidence further supports the conclusion that tomosyn-1 inhibits the formation of the SNARE complex [31], [32].
The Ca2-independent inhibitory effects of the tomosyn-1 have been attributed to the VLD. More recently, Yamamoto et al demonstrated that in the presence of Ca2+, tomosyn-1, via the N-terminal WD40 domain, binds to synaptotagmin and inhibits SNARE complex-mediated neurotransmitter release [26], [30], [33], [34]. Together, the evidence is accumulating for tomosyn-1 as a negative regulator of exocytosis in both the stimulated and unstimulated states.
Insulin resistant animals compensate for their insulin resistance and maintain normal glucose levels by increasing insulin secretion. Our studies show that mutations in tomosyn-2 that increase its inhibitory activity can create a bottleneck and in the presence of obesity-induced insulin resistance, tip the balance towards T2D. However, it is also possible that tomosyn-2 plays an important role in regulating insulin secretion during daily starve/feed cycles by preventing inappropriate insulin secretion during fasting.
Tomosyn-2 may regulate exocytosis by modulating the formation of the SNARE complex in tissues other than islets. We observed significant tomosyn-2 expression in brain, cerebellum, islets, kidney, liver, and gastrocnemius (Figure 7). Therefore it is possible that tomosyn-2, like tomosyn-1, may have an important regulatory role in tissues where regulation of the SNARE complex can be limiting for an important transport process; e.g. insulin-mediated GLUT4 translocation in adipocytes [35] and transport of LDL-derived cholesterol from the trans-Golgi network to the endoplasmic reticulum in hepatocytes [36]. Thus, this tomosyn-2 could be playing a critical role in regulating vesicle trafficking in other tissues.
In summary, we have identified tomosyn-2 as a gene underlying a T2D susceptibility QTL on Chr 16. We show that tomosyn-2 is a negative regulator of insulin secretion. We identified a SNP in tomosyn-2 that affects the stability of the protein and thus suggest a molecular mechanism by which allelic variation in this gene increases diabetes susceptibility. Future studies will focus on the pathways that link nutrient sensing with the role of tomosyn-2 in the regulation of insulin secretion.
The enzymatic glucose reagent was purchased from Thermo Scientific. Insulin in lean mice was measured using a radioimmunoassy kit from Linco Research (St. Charles, MO). In Lepob/ob mice, insulin was measured with an in-house ELISA using an anti-insulin antibody from Fitzgerald Industries (Acton, MA). The mouse anti-myc antibody and Z-Leu-Leu-Leu-al (MG132) were purchased from Sigma-Aldrich, USA. The mouse secondary antibodies were purchase from Cell Signaling Technology (Boston, MA). Glutathione 4B Sepharose beads were purchased from GE Healthcare, USA.
The C57BL/6 (B6) and BTBR T+ tf (BTBR) mice were intercrossed and were crossed to generate F1 mice. The B6.16B6 and B6.16BT mice were created by backcrossing the F1 mice to B6 using microsatellite markers to select for BTBR (or B6 in the case of B6.16B6) homozygosity on mouse chromosome 16 in an otherwise B6 background. The B6.16BT mice were further backcrossed to B6 with marker assisted selection to create congenic strains. Further identification of the B6/BTBR genetic boundaries were determined by SNP sequencing for some of the congenic strains (listed in Dataset S2. The Lepob mutation was introgressed into all strains using Lepob/+ mice as breeders [37]. All mice were maintained at the Department of Biochemistry, University of Wisconsin-Madison animal care facility on a 12 h dark-light cycle (6 PM to 6 AM). The mice were fed Purina Formulab Chow 5008 and water ad libitum. The mice were kept in accordance with the University of Wisconsin-Madison Research Animals Resource Center and the NIH guidelines for care and use of laboratory animals.
For plasma glucose and insulin measurements, blood was taken from the retro orbital sinus from random fed mice at 8 AM or from fasting mice at 12 PM (fasted at 8 AM). For both Lepob/ob and lean mice, glucose was measured via glucose oxidase method (Thermo Scientific). For Lepob/ob mice, insulin was measured via ELISA using a matched rat insulin antibody pair (Fitzgerald Industries International Inc.). For lean mice, insulin was measured by Linco Sensitive rat insulin radioimmunoassay.
Intact pancreatic islets were isolated from mice using a collagenase digestion procedure [38]. Static insulin secretion assays were performed on preparations consisting of three islets incubated with various secretagogues [38]. For perifusion insulin secretion assays, approximately 100 medium sized islets were washed three times, placed in a sterile Petri dish, and incubated overnight in culture media (RPMI 1640, with 11.1mM glucose, antibiotics and 10% heat inactivated fetal bovine serum). The following day, 50 islets were washed and transferred in 100 µl of Krebs Ringer Buffer (KRB) to the Swinnex filter holder (Millipore). The islets were sandwiched between two layers of Bio-Gel P-2 bead (Bio-Rad) solution (200 mg beads/ml in KRB; bottom layer, 150 µl and top layer, 300 µl). The Swinnex filter holder was attached in-line with a Minipuls 3 pump (Gilson) and a FC 204 Fraction Collector (Gilson). Islets were perifused at the rate of 1ml/min and samples were collected at 30 sec intervals. Islet insulin content and secretion were determined by ELISA.
Tail DNA was extracted from B6, BTBR, B6.16BT36–38, B6.16BT24–37and B6.16BT24–38 mice using the QIAGEN Puregene Core Kit. RNA baits were designed using Agilent eArray and used for Agilent SureSelect Target Enrichment to capture sequence from a 35.35 to 38.65 Mb region on mouse chromosome 16. Target enrichment was followed by DNA amplification and confirmation of enrichment using SNP sequencing inside and outside of the target region. DNA was sequenced by Next Generation Sequencing using an Illumina GA IIx sequencer at the University of Wisconsin-Madison Biotechnology Center. CLC Genomics 4.0.3 Software was used to identify SNPs and DIPs between B6 and BTBR sequence. For some SNPs and DIPs an additional visual base calling confirmation step was used to test the accuracy of the software (listed in Dataset S1).
RNA from islets, kidney, and liver was extracted using the QIAGEN RNeasy Plus Kit. RNA from epididymal fat pads, brain, cerebellum, and adrenal glands was extracted using QIAGEN RNeasy Lipid Kit. RNA from heart, soleus, gastrocnemius and quadriceps was extracted using QIAGEN RNeasy Fibrous Tissue Kit. Following extraction, RNA was used for cDNA synthesis (Applied Biosystems). The mRNA abundance was determined by quantitative PCR using FastStart SYBR Green (Roche) and gene expression was represented by comparative ΔCT method.
MMLV-based retroviral vector (RVV, 3051) (gift from Dr. Bill Sugden, University of Wisconsin, Madison) containing a MCS-IRES GFP was used to generate b-tomosyn-2-RVV construct for expression studies. The pcDNA3-m-tomosyn-1, pCR-Script-xb, -b, -m, and s-tomosyn-2 constructs were generously provided by Dr. Alexander Groffen, Virije Universiteit, Netherlands. We corrected a mutation (AG) at nucleotide 3245 of the b-Tomosyn-2 cDNA. The tomosyn-1 or tomosyn-2 cDNA from these vectors were used for subsequent subcloning. The b-tomosyn-2-RVV construct was generated by subcloning the b-tomosyn-2 cDNA with 5′-BspDI and 3′-NotI overhangs into the compatible 5′-BstBI and 3′-NotI ends of the RVV vector. For binding studies, the tomosyn-2-pcDNA/TO/myc-His was generated by subcloning a PCR-amplified tomosyn-2 cDNA in to 5′-BamHI and 3′-XhoI sites of the pcDNA4/TO/myc-His C vector (Invitrogen). The following primers that were used to amplify tomosyn-2 cDNA with the restriction sites for cloning, a partial KOZAK, and a 3′-precision protease cleavage site are: forward (5′-TTAAAGGATCCGCCACCATGAAGAAGTTTAATTTCCG) and reverse (5′-ATATCTCGAGGGGCCCCTGGAACAGAACTTCCAGGAACTGGTACCACTTCTTATCCT). Similar subcloning strategy was used for generating m-tomosyn-1-pcDNA/TO/myc-His construct. The primers used are as follows: forward (5′-CGAGACCGGATCCGCCACCATGAGGAAATTCAACATC) and reverse (5′-ATATCTCGAGCCCCTGGAACAGAACTTCCAGGAACTGGTACCACTTCTTATCTTTG) primes. The pGEX-4T1-syntaxin-4 construct encoding soluble GST-syntaxin-4 (1-273) fusion protein was generated as previously described [39]. The pGEX-2T1-syntaxin construct (1-265) was a generous gift from Dr. Tom Martin, University of Wisconsin, Madison. All constructs were verified by sequencing.
The glucose responsive rat β-cell line, INS1 (832/13, a gift from Dr. Chris Newgard, Duke University) was cultured in RPMI 1640 medium containing 11 mM glucose supplemented with 10% heat inactivated fetal bovine serum, 2 mM L-glutamine, 1 mM sodium pyruvate, 10 mM HEPES, 100 Units/ml of antibiotic-antimycotic, and 50 µM β-mercaptoethanol. Approximately 100,000 cells/well were plated in a 96-well plate. The following day, INS1 (832/13) cells at 80–90% confluency were transfected with 0.4 µg of plasmid DNA using Lipofectamine 2000 (Invitrogen). After 36 h of incubation, cells were washed once with 200 µl and incubated for 2 h in 100 µl of modified Krebs-Henseleit Ringer bicarbonate buffer (KRB: 118.41 mM NaCl, 4.69 mM KCl, 1.18 mM MgSO4, 1.18 mM KH2PO4, 25 mM NaHCO3, 20 mM HEPES, 2.52 mM CaCl2, pH 7.4, and 0.2% BSA) containing 1.5 mM glucose. After 2 h, cells were stimulated for 2 h in 100 µl of KRB buffer containing 7 mM glucose + 3 mM 8-bromo-cAMP. The incubation buffer was collected to determine the amount of insulin secreted under varying conditions. The cells were lysed (lysis buffer: 1 M Tris-HCl, pH 8.0, 1 M NaCl, 0.5 M NaF, 200 mM Na3VO4, 2% NP-40, and protease inhibitor cocktail tablet (Roche)) to determine insulin content. The percent fractional insulin secretion was calculated as the amount of insulin secreted divided by total insulin content. Insulin was determined using ELISA.
The human embryonic kidney 293T cells (HEK293T) were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 25 mM glucose were supplemented with 10% fetal bovine serum, 0.1 mM nonessential amino acid, 6 mM L-glutamine, 1 mM sodium pyruvate, 100 units/ml of penicillin, 100 units/ml of streptomycin, and 500 µg/ml of geneticin. HEK293T cells at 70–80% in 100 mm tissue culture dishes were transfect with plasmid DNA constructs using 40 µl of 1 mg/ml polyethylenimine. Following day, cells were lysed (lysis buffer: 20 mM Tris-HCl (pH 7.5). 150 mM NaCl, 1 mM Na2EDTA, 1mM EGTA, 1% Triton, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM Na3VO4, 1 mM PMSF, and protease inhibitor cocktail tablet (Roche)) and total protein lysates were prepared and the immunoblot was performed as described [40]. For protein stability, 16 h post transfection, cells were treated with or without 100 µM MG132 for 6 h. After 6 h, cells were lysed and whole cell lysates were prepared.
Recombinant proteins encoding GST or GST-fusion proteins with the cytoplasmic domain of syntaxin-1A and syntaxin-4 were expressed in E.Coli strain BL21 (DE3) and were purified using glutathione-affinity chromatography [41]. The concentration and the purity of the fusion proteins were assessed by SDS-PAGE followed by Coomassie-blue staining against BSA standards. The binding studies were preformed by incubating 10 µg of recombinant proteins with 25 µl of 100% Glutathione-Sepharose 4B beads (Amersham Biosciences) with 1 mg of HEK293T lysate overexpressing tomosyn-1 or isoforms of tomosyn-2 (xb, b, m, and s) in 1% Triton-lysis buffer for 2 h at 4°C. After 2 h, the complexes were washed three times with Triton lysis buffer and was eluted in Western loading buffer. The denatured samples were subjected to 10% SDS-PAGE gels followed by transfer to PVDF membrane for immunoblotting. The immunobloting was performed using a standard protocol [40].
Data was expressed as means ± standard error of means. The statistical comparisons were made using Student's t test at p<0.05.
|
10.1371/journal.pgen.1005108 | GAGA Factor Maintains Nucleosome-Free Regions and Has a Role in RNA Polymerase II Recruitment to Promoters | Previous studies have shown that GAGA Factor (GAF) is enriched on promoters with paused RNA Polymerase II (Pol II), but its genome-wide function and mechanism of action remain largely uncharacterized. We assayed the levels of transcriptionally-engaged polymerase using global run-on sequencing (GRO-seq) in control and GAF-RNAi Drosophila S2 cells and found promoter-proximal polymerase was significantly reduced on a large subset of paused promoters where GAF occupancy was reduced by knock down. These promoters show a dramatic increase in nucleosome occupancy upon GAF depletion. These results, in conjunction with previous studies showing that GAF directly interacts with nucleosome remodelers, strongly support a model where GAF directs nucleosome displacement at the promoter and thereby allows the entry Pol II to the promoter and pause sites. This action of GAF on nucleosomes is at least partially independent of paused Pol II because intergenic GAF binding sites with little or no Pol II also show GAF-dependent nucleosome displacement. In addition, the insulator factor BEAF, the BEAF-interacting protein Chriz, and the transcription factor M1BP are strikingly enriched on those GAF-associated genes where pausing is unaffected by knock down, suggesting insulators or the alternative promoter-associated factor M1BP protect a subset of GAF-bound paused genes from GAF knock-down effects. Thus, GAF binding at promoters can lead to the local displacement of nucleosomes, but this activity can be restricted or compensated for when insulator protein or M1BP complexes also reside at GAF bound promoters.
| Transcriptional regulation is critical for proper gene expression in response to environmental changes and developmental programs. Eukaryotes have evolved multiple mechanisms by which transcription factors regulate transcription. One mechanism is the reorganization of chromatin to allow Pol II recruitment. Another is the release of promoter-proximal paused Pol II, where Pol II transcription that is halted 20–60 bases downstream of the transcription start site (TSS) is allowed to enter into productive elongation through the gene body. The Drosophila transcription factor GAF binds to genes that undergo pausing and interacts with nucleosome remodelers and the pausing factor NELF. Thus, GAF can regulate multiple points necessary for transcription, but its mechanistic role is not fully understood genome-wide. We depleted GAF from cells and examined the genome-wide changes in Pol II and nucleosome distributions across genes. We found that GAF depletion reduces polymerase density at genes where GAF binds just upstream of the TSS, and results in nucleosomes moving into the promoter region. Our results show that GAF is important for maintaining the promoter accessibility, allowing Pol II to be recruited to promoters and enter the pause sites downstream of the TSS. Thus, GAF is critical for providing the chromatin environment necessary for the proper control of gene expression.
| Transcription is controlled by transcription factors (TFs) that modulate various steps in the transcription process. Two major points of transcription regulation are recruitment of Pol II to a preinitiation complex (PIC) and promoter-proximal pausing. PICs form when general transcription factors bind to accessible nucleosome-free promoters and recruit Pol II. TFs can change the rate of PIC formation by altering either nucleosome placement on promoters or Pol II recruitment [1]. In addition, many genes are regulated after Pol II recruitment by the controlled release of a stable paused Pol II, which is typically located in the promoter-proximal region 20–60bp downstream of the transcription start site [2]. TFs can stimulate release Pol II from the pause by recruiting, directly or indirectly, P-TEFb kinase that modifies the paused Pol II complex, allowing it to efficiently transcribe across the gene [3].
GAF, encoded by the gene Trithorax-like (Trl), is a Drosophila sequence-specific TF that is associated with the promoters of many genes [4]. GAF was first identified as a regulator of developmental genes and binds GA repeats [5–9]. The GAF DNA binding domain is composed of a basic-rich region followed by a C2–H2 zinc finger that binds DNA sequences as short as GAG or the longer sequence of GAGAG in vitro [5,6,10]. However, in vivo bound regions generally have clusters of GAGA elements [11,12]. In addition to the DNA-binding domain, GAF has a BTB/POZ domain that mediates interactions with other proteins, and allows GAF to homodimerize or heterodimerize with other BTB/POZ-containing factors [13–17]. GAF also has a polyQ domain. Its function is not well-understood, but has been reported to act both as a transcription activator [18,19] and as a multimerization domain that can influence DNA binding [20,21].
Genome-wide studies have identified many genes bound by GAF [4,11,22–24], and GAF binding is enriched on paused genes [4,25,26]. In addition, transgenic reporter genes have transcriptionally-engaged polymerase in their promoter-proximal regions under basal conditions when GAGA elements are present [27,28]. These results suggest that GAF plays a role in establishing paused polymerase.
Several reports support a role of GAF as an anti-repressor for genes [29]. The GAF anti-repressor function is proposed to maintain promoters in an accessible state [30]. GAF can interact with several nucleosome remodelers, including NURF, ISWI, and BPAP, and displace adjacent nucleosomes to make DNA accessible regions [30–33], but this function of GAF has not been investigated in a genome-wide manner.
Here, we examine the role of GAF in transcriptional regulation and nucleosome positioning genome-wide, using global run-on sequencing (GRO-seq) to map transcriptionally-engaged polymerases and MNase-seq to map nucleosome positions in control and GAF-RNAi depleted Drosophila S2 cells. Also, we define GAF binding sites at high resolution and assess their sensitivity to GAF knock-down using ChIP-seq. This allows GAF binding in promoters to be correlated with its effects on transcription and pausing and other factors that function redundantly to GAF or protect genes from the effects of GAF knock-down. Finally, MNase-seq mapping of nucleosomes genome-wide in control and GAF-RNAi cells supports a mechanism by which bound GAF maintains a nearby nucleosome free region at both the promoters of many genes and non-promoter sites.
We initially examined the role of GAF in pausing on the prototypical paused genes, Hsp70. Under basal (non-heat shock, NHS) conditions, GAF is bound to the Hsp70 promoters and Pol II transcribes 20–40 bases downstream from the transcription start site (TSS) and stably pauses. GAF binding was previously implicated in Hsp70 pausing, as a Hsp70 transgene with a mutant GAGA element showed reduced pausing [28]. To test if GAF has a role in pausing on the endogenous Hsp70 genes, we first treated cells with dsRNA targeting all isoforms of GAF, and reduced GAF levels to less than 10% of those in untreated or control cells treated with LacZ dsRNA (Fig. 1A). Chromatin-immunoprecipitation (ChIP) showed that GAF binding on the Hsp70 promoters (-154bp from the TSS) decreased about 4-fold in NHS GAF-RNAi cells (Fig. 1B). We assayed the effect of GAF depletion on the paused polymerase present on NHS Hsp70 using ChIP for the Pol II subunit, Rpb3. In untreated and LacZ-RNAi cells, Pol II levels were high at the 5’ end of Hsp70 (+96bp from the TSS) and decreased in the gene body to near the levels on a non-transcribed (bkgd.) region (Fig. 1C). GAF knock-down resulted in 2-fold reduction in Pol II in the +96 region with no discernible change in the gene body (Fig. 1C). These results show that GAF has a role in maintaining the level of paused Pol II on the 5’ end of NHS Hsp70.
Previous ChIP-chip studies have shown that about 1,500 genes are bound by GAF in S2 cells and these genes are enriched for paused Pol II [4,25,26]. To test the role of GAF in transcription genome-wide, we performed GRO-seq in biological replicates of untreated, LacZ-RNAi, and GAF-RNAi cells to obtain the genome-wide distribution of transcriptionally-engaged polymerases [34]. GRO-seq maps polymerase by affinity purifying and sequencing nascent RNAs after bromo-UTP (BrUTP) incorporation in a nuclear run-on [34]. The density of sequence reads mapped within a region indicates the number of engaged polymerase in the cells from which the nuclei were isolated. In agreement with previous GRO-seq results in Drosophila [35–37] and genome-wide Pol II ChIP data [37,38], the average GRO-seq read profile for genes in each library displayed a peak of engaged polymerase on the 5’ end, and the average Pol II level was not changed by knock-down (Fig. 2A). To examine the polymerase distribution at individual genes, we quantified GRO-seq reads in the promoter-proximal and gene body regions for 9,452 non-overlapping genes [34,37]. We examined the transcription on each gene. Paused genes were defined as genes with significantly higher levels of engaged polymerase in the promoter region than the gene body (Fisher’s exact p-value <0.01). Transcriptionally active genes were defined as genes with significantly higher density of engaged polymerase in their gene body compared to 1% of mapped reads distributed uniformly across the Drosophila genome, the estimated level of background reads (p-value <0.01) [34]. We found that about half were significantly paused, and 60% of genes were actively transcribed. Notably, paused genes were highly enriched among those that were transcriptionally active (72% of transcribed genes were paused, and over 90% of paused genes were transcribed; Table 1).
The GRO-seq biological replicates were used to identify genes that significantly change between control and GAF-RNAi treatments. The biological replicates gave reproducible results: the promoter and gene body GRO-seq read counts for all biological replicates were highly correlated, with Pearson’s correlation coefficients (r) between 0.907–0.968 (S1 Table). Consistent with the similarity between the average GRO-seq read distribution across genes (Fig. 2A), the read counts for the combined replicates correlated well between the untreated and LacZ-RNAi libraries (promoters r = 0.984, gene bodies r = 0.997). We used edgeR to identify statistically significant changes with a false discovery rate corrected threshold q<0.01 in GRO-seq read counts separately in the promoter and gene body regions [39]. There were no genes with significantly different promoter read counts between the untreated and LacZ-RNAi libraries (S1A Fig), and only 5 genes had significantly different gene body read counts (S1B Fig, orange points). In contrast, there were 141 genes with significantly different read counts in the promoter-proximal region in GAF-RNAi and all but one was reduced (Fig. 2B, red points). The GAF-RNAi library had only 84 genes with gene body read levels significantly different from LacZ-RNAi. The majority of these were decreased (68 decreased and 16 increased) (Fig. 2C, orange points), and this bias for decreased reads following GAF-RNAi was highly statistically significant (p = 4.27x10-9, binomial test). These results support a role for GAF, beyond Hsp70, in maintaining levels of Pol II on the 5’ end of genes.
A reduction in recruitment and entry of Pol II into the pause site can lead to a decrease in elongating (gene body) polymerase. Indeed, recent studies have shown that disrupting initiation reduces both pausing and elongating polymerase [40,41]. Following GAF-RNAi, changes in polymerase density were more dramatic in the pause region than the gene body (S1E Fig and S1F Fig), and as a result many genes observed to have significant changes in the pause peak were not called statistically significant in the gene body by edgeR. We hypothesized that the lack of genome-wide statistical significance at many of these genes was because we were underpowered to identify smaller changes using only two biological replicates. To address this possibility, we asked whether genes that show a significant decrease in paused Pol II also show a significant bias for having a decrease in gene body Pol II. We found that genes with significantly reduced promoter GRO-seq reads upon GAF knock-down were also enriched for reductions in gene body reads (Fig. 2D, red points; p = 4.44x10-16, binomial test), demonstrating that, as a group, gene body Pol II decreased along with promoter proximal Pol II. These changes suggest that GAF plays a role early in the transcription cycle, allowing Pol II to initiate transcription and establish pausing at certain genes, which in turn influence the level of Pol II that progresses into the gene body.
To assess if the effects on promoter-proximal polymerase levels are likely to be a direct effect of GAF knock down, we used ChIP-seq to analyze GAF binding sites and the sensitivity of GAF binding at each site to the reduced GAF protein levels in GAF-RNAi cells (S2A Fig). ChIP was performed with an affinity purified GAF antibody in both untreated and GAF-RNAi NHS S2 cells. The combined biological replicates of untreated control material identified 12,583 individual peaks and knock-down reduced binding on the large majority of sites (S2B Fig, S3 Table). The levels of control ChIP-seq reads within each peak correlated well with the previous ChIP-chip data [42] (S2C Fig, r = 0.887) and ChIP-qPCR for GAF at selected sites (S2D Fig, r = 0.718).
To evaluate if GAF is preferentially associated with promoter-proximal pausing, we first determined all the genes that have GAF ChIP-seq peaks within the promoter (within 500bp upstream of the TSS) and gene body. In our set of 9,452 non-overlapping genes, GAF was bound to 1,939 (S2 Table). The majority of these genes had at least one peak within their promoter (1,221; 63%). GAF-bound genes were significantly enriched for actively transcribed genes compared to all other genes (Fisher’s exact test, p < 2.2x10-16) and for paused genes (Fisher’s exact test, p < 2.2x10-16) or all other transcribed genes (Fisher’s exact test, p = 5.41x10-5), which is consistent with previous reports [4,25] (Table 2).
The majority of genes with significantly reduced promoter GRO-seq reads in GAF RNAi-treated cells show GAF binding in untreated cells. Of the 140 genes with significant reduction in promoter GRO-seq reads between the GAF-RNAi and LacZ-RNAi libraries (reduced promoter), all of them were paused and 134 (95.7%) were bound by GAF (Fig. 3A). This suggests that changes in polymerase levels after GAF depletion are a primary effect of the knock-down and the effects of RNAi on levels of pausing are mediated through GAF acting locally at the gene, and not over a large chromatin domain.
Promoter-bound GAF cannot be the sole determinant for pausing because less than 14% of paused genes with GAF-bound promoters had significant reductions in promoter GRO-seq reads upon GAF knock-down. To investigate this further, we divided genes into two sets: paused genes with GAF-bound promoters that had significant reductions in promoter GRO-seq reads (hereafter referred to as Pause Reduced) and the other paused genes with GAF-bound promoters (hereafter referred to as Pause Unchanged). Then we looked for molecular signatures at GAF binding sites that correlated with the magnitude of pausing change after knock-down. The level of GAF binding on promoters was significantly higher in Pause Reduced genes than Pause Unchanged (Fig. 3B, black versus maroon line), even though GAF-binding was reduced by a similar fraction on Pause Reduced and Pause Unchanged promoters (Fig. 3B, gray versus black line and red versus maroon line).
The lower levels of GAF binding on Pause Unchanged genes raised the concern that these peaks could be an artifact and not bona fide GAF binding sites. To identify a high confidence subset of GAF peaks, we selected peaks with 2 additional criteria: they must overlap a peak region called in a dataset from an independent GAF antibody and they must contain a GAGA element. We used the modENCODE GAF ChIP-chip as the independent antibody dataset [42], and found that 9808 of our ChIP-seq peaks overlap with ChIP-chip enriched regions (S3 Table). GAGA elements were called using the position-weight matrix from the JASPAR database (Trl) [43] and 4,397 peaks had a GAGA element (defined using a p-value cutoff <1x10-4, S3 Table). Applying both criteria to our ChIP-seq peaks resulted in 3622 high-confidence GAF (hcGAF) peaks (S3 Table). Although hcGAF peaks were enriched on the Pause Reduced promoters as compared to Pause Unchanged promoters (Fig. 3A, Fisher’s exact test p = 4.542x10-7), 39% of Pause Unchanged promoters had hcGAF peaks, indicating many Pause Unchanged genes are likely truly bound by GAF.
To identify the basis of the differential effects of GAF knock-down, we assessed whether other characteristics of paused genes with GAF-bound promoters correlate with the reduction in pausing. Individual labs and the modENCODE consortium have determined the genome-wide binding profiles for many chromatin-bound factors and histone modifications. We used this information to investigate if any of the factors with genome-wide data in S2 cells correlate with the GAF-RNAi effects on pausing (S3A Fig). Several factors were enriched on Pause Unchanged genes, but the most striking association was seen with BEAF32 [42,44], Chriz [42], and Motif-1-binding protein [45] binding levels (Fig. 4) and more modestly for other insulator factors (S3B-M Fig). Interestingly, BEAF, other insulators, and Chriz all colocalize at chromatin boundaries [46], and these proteins may insulate nearby promoters from the actions of locally bound GAF, making paused Pol II less sensitive to GAF knock-down.
Motif-1 Binding Protein (M1BP) is a transcription factor recently shown to be enriched on a set of paused genes, largely distinct from GAF-bound paused genes, and is believed to function analogously to GAF in Pol II pausing [45]. The striking enrichment of M1BP at Pause Unchanged genes suggests bound M1BP, and possibly other yet to be identified factors, provide functions redundant with GAF. We propose that pause-inducing redundant factors and insulator proteins conspire to render Pause Unchanged promoters insensitive to GAF.
GAF has been shown to affect promoter accessibility through interactions with nucleosome remodelers [30,32,33,47]. To investigate whether the differential effects of GAF knock-down are due to changes genome-wide in promoter accessibility, we performed MNase-seq experiments in LacZ-RNAi and GAF-RNAi cells. Replicates within each treatment correlated well (S4 Table), and the combined replicates for both treatments had the same average level and expected distribution across genes grouped by transcriptional status (S4A-D Fig). We examined nucleosome-sized (120–180bp) MNase-seq reads across GAF-bound paused promoters. Pause Unchanged promoters had low levels of nucleosomes within the promoter increasing to a peak around 135bp downstream of the TSS in the control LacZ-RNAi condition, similar to that of typical transcribed genes (Fig. 5A, black line). Intriguingly, even with the normal levels of GAF in the LacZ-RNAi control, the Pause Reduced promoters had higher nucleosomes around their TSS and the nucleosomes were more disordered downstream (Fig. 5A, maroon line). When GAF was knocked-down, there was only a slight change in the distribution of nucleosomes in the Pause Unchanged promoters (Fig. 5A, gray line), but nucleosomes dramatically increased on the Pause Reduced promoters (Fig. 5A, red line). Heatmaps confirmed that individual promoters in each of these gene sets have changes that are consistent with the average profiles for each class (Fig. 5B). We used edgeR to determine the promoters with significant changes in MNase-seq reads and found that the Pause Reduced promoters were enriched for significantly increased MNase-seq reads (Fig. 5B, right panel). These results indicate that Pause Reduced promoters fill in with nucleosomes upon GAF knock-down.
The nucleosome-sized MNase-seq reads used may not necessarily be produced by a nucleosome. Therefore, to further validate these results, we immunoprecipitated the promoter-enriched histone variant H2AvD from the MNase-seq material. As expected, H2AvD levels were highest at the -1 and +1 nucleosomes bordering promoters of actively transcribed genes and these nucleosomes were not changed genome-wide by GAF knock-down (S4E-H Fig). Similar to the LacZ-RNAi MNase-seq results, the Pause Unchanged genes had higher levels of H2AvD and a more positioned +1 H2AvD-containing nucleosome than the Pause Reduced genes in the LacZ-RNAi control libraries (Fig. 5C, the black versus the maroon line). The Pause Unchanged H2AvD levels or position were not altered by GAF knock-down, but the Pause Reduced genes showed a dramatic increase in H2AvD in their promoter and the H2AvD levels were relatively even across the entire region (Fig. 5C the gray versus the red line, S5A Fig). Indeed, the Pause Reduced promoters were enriched for significant increases in H2AvD reads (S5A Fig, right panel). Thus, GAF is enabling these promoters to adopt a nucleosome-free conformation that may in turn allow polymerase to initiate, and indirectly, to establish a promoter-proximal pause state.
Recently, it was shown that the paused Pol II itself was important for preventing nucleosome encroachment into the promoter [48]. Therefore, the increases in promoter nucleosomes upon GAF knock-down could possibly be due to the reduction in paused polymerase by some GAF-dependent mechanism that is distinct from our proposed function of GAF in maintaining the nucleosome-free conformation. To test whether GAF can directly maintain nearby regions in a nucleosome-free conformation, we examined intergenic GAF-bound regions away from paused polymerases. Indeed, these regions had dramatically lower average levels of transcriptionally-engaged polymerase nearby (S6A Fig). We looked at 611 intergenic hcGAF peaks oriented based on the strand of the GAGA elements within them. We found the LacZ-RNAi control MNase-seq reads were higher on one side of the GAF peaks, suggesting a DNA sequence specific directionality to nucleosome placement (Fig. 5D, gray line). Moreover, MNase-seq reads dramatically increased in GAF knock-down library (Fig. 5D, red line), and this increase was most evident on the GAF peaks with largest ChIP-seq reduction in GAF binding upon GAF-RNAi (S5B Fig). Additionally, the levels of transcriptionally-engaged polymerase were similar between all hcGAF intergenic peaks and still dramatically lower than the promoter regions with paused Pol II, independent of reduction in GAF binding (S6B Fig). These results indicate GAF itself can direct the maintenance of a nucleosome-free region.
In this study, we examine the role of GAF in transcription and pausing genome-wide using GRO-seq to map transcriptionally-engaged polymerases in Drosophila S2 cells depleted for GAF. Almost all of the 140 paused genes with significant reductions in promoter-proximal polymerase levels upon GAF depletion had GAF bound in their promoters. This result indicates that these reductions were direct effects of GAF knock-down and GAF functioned locally at these genes to maintain paused Pol II levels. Moreover, we demonstrate that GAF has a prominent role in creating a chromatin accessible promoter for the recruitment and initiation of Pol II transcription. This opening of chromatin can be seen at GAF binding sites in promoters, but also at intergenic sites that are far from promoters. These results provide strong in vivo and genome-wide support for the hypothesis that GAF can mediate nucleosome displacement proximal to its binding site, as was proposed from in vitro studies that examined the interplay of GAF binding and an ATP-utilizing remodeler (NURF) on the Hsp70 promoter [30].
Several points in the transcription cycle can be targeted to regulate the level of promoter-paused Pol II [1,49]. A TF may contribute to the recruitment of Pol II to the pause by acting at steps upstream of pausing to allow recruitment, initiation, and entry to the pause site (e.g. ERα) [50], or a TF can contribute more directly by creating or stabilizing the paused Pol II (e.g., Spt5/Spt4 and NELF) [51,52]. A TF can also accelerate the release of paused Pol II into productive elongation and thereby reduce the level of paused Pol II. The expectation for disrupting a factor that aids the steps in either recruitment or initiation is that the level of both paused Pol II and Pol II transcribing the gene body will decrease. For example, inhibition of the helicase TFIIH results in the decay of both paused Pol II and Pol II elongating across genes [40,41]. Our results indicate that GAF knock-down reduces levels of transcriptionally-engaged polymerase on the genes where promoter polymerase levels are significantly reduced. This suggests that these genes are dependent on GAF to allow recruitment and initiation providing the Pol II that will subsequently the pause, and thereby, indirectly helping to establish pausing.
Previous studies have shown that GAF can interact with several nucleosome remodelers and maintain promoters in a transcription-competent conformation, but these results have been limited to a few specific genes [31–33]. We found that nucleosome levels dramatically increase on the genes with significantly reduced promoter-proximal polymerase. Interestingly, we found that these genes had higher levels of nucleosomes on their promoter before GAF knock-down, suggesting there is already a competition between nucleosomes and paused Pol II on these promoters under normal conditions. Interpretation of the nucleosome increase upon GAF knock down at these genes is complicated by the recent report indicating that paused Pol II can keep some promoters open [48]. It is possible that GAF contributes directly to Pol II pausing and it is the loss of paused Pol II in GAF knockdowns that leads to increases in nucleosome occupancy. However, knockdown of GAF leads to dramatically increased nucleosome occupancy at GAF sites that are intergenic and away from paused promoters. Thus, GAF appears to be critical to opening chromatin structure at many sites independent of whether or not paused Pol II is present.
Collectively, our analyses demonstrate how TFs can work together to regulate the expression of target genes. We find that only a subset of the paused genes with GAF-bound promoters had reductions in promoter polymerase levels upon GAF knockdown. GAF levels were higher on these genes, and this may reflect that stable binding of GAF is necessary to maintain the chromatin in an open conformation. Interestingly, the set of GAF bound genes whose promoter polymerase levels are insensitive to GAF knockdown are enriched for the transcription factor M1BP, the insulator protein BEAF, and the BEAF-interacting protein Chriz. M1BP was recently found to be enriched on paused genes that are mostly distinct from the group bound by GAF [45], suggesting that this TF might independently facilitate Pol II recruitment and initiation and partially compensate for the loss of GAF in the knockdown. In support of this, multiple mammalian TFs were shown to stimulate formation of paused Pol II without greatly affecting escape of paused Pol II into productive elongation [53,54]. Insulators might also act by unknown mechanisms to compensate for the loss of GAF, or the insulator may be blocking GAF’s action on promoters and allowing other factors like M1BP to independently cause Pol II to generate promoter-paused Pol II. Therefore, these results indicate that many of the genes lacking a significant effect following GAF knockdown are explained by the combinatorial patterns of factor binding and their interplay at the target promoter region.
GAF may function to specify pausing on bound genes by altering multiple steps in the transcription cycle. As we have shown, GAF can indirectly help to establish pausing by binding the promoter and maintaining nucleosome-free promoter regions that allow recruitment and initiation by Pol II. GAF may also have a direct role in initiation, as others have shown that GAF can itself act as an activator through its poly-glutamine domain [18,19] or may promote initiation through interactions with the TAF3 subunit of TFIID [55]. GAF can also interact with NELF to focus pausing in vitro on Hsp70 more proximal to the TSS [56], although these changes in the position of the pause could not be picked up by the GRO-seq assay used here.
While GAF may act by more than one mechanism to generate and maintain paused Pol II, our results provide strong support for the hypothesis that GAF functions genome-wide to keep adjacent regions of chromatin nucleosome free. Our hypothesis is also consistent with previous reports that support a role of GAF as an anti-repressor for genes [29]. GAF might be simply competing with nucleosomes to promote chromatin accessibility; however, GAF is known to interact with several nucleosome remodelers: NURF, ISWI and BPAP, and displace adjacent nucleosomes to make DNA accessible regions [30–33]. We propose that the nucleosome landscape and Pol II occupancy at a subset of promoters is regulated by GAF’s recruitment of nucleosome remodelers and other factors, allowing Pol II entry and pausing.
Drosophila S2 cells were grown in M3+BPYE+10% serum to a density between 3–5x106cells/ml. After splitting to 1x106cells/ml in serum-free M3 media (at least a 1:3 split), the desired volume of cells were mixed with 10μg/ml double-stranded RNA (dsRNA), incubated at 25°C for 45 minutes, and then, an equal volume of M3+BPYE+20% serum was added. After 5 days, the cells were harvested for the experiments. The dsRNAs were generated from a PCR template with T7 promoters on each end, targeting either a region conserved in all GAF isoforms or a region of B-galactosidase (LacZ) gene serving as a control.
GAF Forward: GAATTAATACGACTCACTATAGGGATGGTTATGTTGGCTGGCGTCAA
GAF-Reverse: GAATTAATACGACTCACTATAGGGATCTTTACGCGTGGTTTGCGT
LacZ-Forward: GAATTAATACGACTCACTATAGGGATCGTCAGTAGAAGAGCACCGAGT
LacZ-Reverse: GAATTAATACGACTCACTATAGGGAGAGATATCCTGCTGATGAAGC
ChIP was performed as it was previously [57]. Briefly, after RNAi treatment, Drosophila S2 cell cultures were cross-linked for 2 minutes with formaldehyde at a 1% final concentration, and the cross-linking was quenched with glycine at a 125mM final concentration. The cell pellets were suspended to 1x108cells/ml in sonication buffer (20mM Tris-Cl pH 8.0, 2mM EDTA, 0.5mM EGTA, 0.5% SDS, 0.5mM PMSF, protease inhibitor cocktail [Roche catalog no. 05 056 489 001]). The cells were sonicated 12 times for 20 seconds each time with a 1 minute rest in between at 4°C using a Bioruptor sonicator (Diagenode) on the highest setting.
The sonicated material was centrifuged at 20,000xg for 10 min at 4°C, and the supernatant was saved for the immunoprecipitation (IP). For each IP, 25μl of cleared sonication material was mixed with 1ml IP buffer (20mM Tris-Cl pH 8.0, 150mM NaCl, 2mM EDTA, 10% glycerol, 0.5% TritonX-100) with the antisera (10μl affinity purified Anti-GAF antibody [58] or 4μl of rabbit anti-Rpb3 antisera [59]) at 4°C overnight. For ChIP-qPCR, a standard curve of 10%, 1%, 0.1%, and 0.01% of input DNA and the immunoprecipitated DNA were quantified using a Roche LightCycler 480, and the standard curve was used to determine the amount of DNA immunoprecipitated.
For the ChIP-seq, two replicates of chromatin immunoprecipitation (ChIP) were carried out for each condition, as previously described [60], and sequenced using Illumina GAIIx sequencer.
GRO-seq libraries were constructed using previous methods [57]. Briefly, nuclei were isolated from RNAi-treated cells. Each nuclear run-on was performed for 10 minutes at 30°C with 2x107 nuclei in run-on buffer (10mM Tris-Cl pH 8.0, 5mM MgCl2, 300mM KCl, 500μM ATP, 500uM GTP, 2μM CTP (cold), 1mCi/ml 32P-CTP (100μCi/ run-on), 500μM Br-UTP, 0.4 units Superase-In, 1mM DTT, 40 units Superase-In (Ambion), 0.6% N-lauroyl-sarcosine), and stopped with 1.5ml Trizol and 200μl chloroform. After extraction with acid phenol:chloroform and chloroform, the precipitated RNAs were resuspended in 20μl DEPC-treated ddH2O, and hydrolyzed in 200mM NaOH on ice for 18 minutes. The hydrolyzed RNAs purified by three bead bindings to Anti-Br-dUTP beads (blocked with 0.1% polyvinylpyrrolidone and 1μg/ml BSA). The beads were washed once with 500μl binding buffer, once with 500μl Low salt buffer (0.2x SSPE, 1mM EDTA, 0.05% Tween-20), once with 500μl High salt wash (0.25x SSPE, 1mM EDTA, 137.5mM NaCl, 0.05% Tween-20), and twice with 500μl TET wash (10mM Tris-Cl pH 7.5, 1mM EDTA, 0.05% Tween-20). After elution with elution buffer (50mM Tris-Cl pH 7.5, 150mM NaCl, 1mM EDTA, 0.1% SDS, 20mM DTT), the precipitated RNAs were resuspended in 20μl DEPC-treated ddH20. After the first bead binding, RNAs are treated with T4 polynucleotide kinase (PNK) without ATP to create a 3’ hydroxyl group. Illumina linkers were added using polyadenylation with E. coli polyA polymerase and reverse transcription from a poly(dT)-3’linker covalently attached to the 5’ linker with a 18 carbon spacer, as previously used [37,61]. Each library was made in biological replicates, and bar-coded using specific reverse transcription primers (INOO3: 5’-pTAGAGATCGTCGGACTGTAGAACTCT-iSp18-CAAGCAGAAGACGGCATACGATTTTTTTTTTTTTTTTTTTTVN, INOO4: 5’-pTGATGATCGTCGGACTGTAGAACTCT-iSp18-CAAGCAGAAGACGGCATACGATTTTTTTTTTTTTTTTTTTTVN). The cDNA was circularized using Circligase (Epicentre catalog # CL4111K) to connect the 5’ linker to the 5’ end of the cDNA. After PCR amplification, the libraries were gel purified away from the primers, each replicate library was combined in equal amounts, and sequenced for 50 bases on one lane of an Illumina GIIAx sequencer.
MNase-seq material was created similar to previous studies [48]. Briefly, RNAi-treated cells were cross-linked identically to the ChIP protocol. Nuclei were isolated from the cross-linked cells, and digested so that 80% of DNA was mononucleosome size. For H2AvD nucleosomes, 75ul of material was immunoprecipitated with 4ul Anti-H2AvD antisera (Glaser lab). After reversal of cross-links, Illumina paired-end TruSeq adapter were ligated to 50ng of DNA using standard protocols, and amplified for 10 cycles. The DNA was size selected for inserts between 80–280bp in length, and paired-end sequencing for 50 bases (each end) was performed on an Illumina Hi-seq sequencer. Reads were aligned to the Drosophila dm3 genome using bowtie2 (—no-mixed—no-discordant). Mononucleosome-sized reads between 120 and 180 bases were selected computationally. The heatmaps and composite profiles used the whole reads mapped to the genome, and the centers of each read were used to calculate significance of changes in read counts with edgeR.
MACS1.4 was used to initially call peaks using the combined replicate data compared against pre-immune IP data (MACS parameters: effective genome size = 1.65e+08, band width = 150, model fold = 10,10000, p-value cutoff = 1.00e-04, Range for calculating regional lambda is: 1000 bps and 10000 bps). Closely clustered subpeaks, within broad regions of MACS-identified peaks, were deconvoluted by using the Subpeaks tool contributed to MACS [62].
We defined high-confidence GAF peaks based on overlap with peaks in an independent GAF dataset and the presence of a GAGA element within our peak. We selected our untreated ChIP-seq peaks that overlapped with the “Regions_of_sig_enrichment” in the modENCODE GAF ChIP-chip GFF3 file [42]. We identified GAGA elements with FIMO (p-value threshold 1x10-4) using the JASPAR Trl motif [43,63].
All mapping, quantification, and transcriptional status determinations were performed as in previous studies [34]. The reads for each treatment were normalized to total mapped reads. To validate that GAF-RNAi was not changing the amount of transcriptionally-engaged Pol II genome-wide, we also used the total number of mapped Pol I and Pol III reads to normalize with little change in results.
We identified paused genes based on higher levels of engaged polymerase in the promoter region than the gene body compared to the number of reads in each region when the reads are uniformly distributed across the gene (Fisher’s exact test p-value <0.01). Transcriptional activity was defined exactly as previously [34]. We calculated the probability that the observed gene body read counts were generated from a Poisson distribution, with a mean equal to the observed background density (1% of mapped reads uniformly distributed) times the number of mappable bases in the gene body. Genes with more reads than expected under the background null model (p< 0.01) were considered transcriptionally active.
Regions with significant changes in GRO-seq reads between Untreated or LacZ-RNAi and GAF-RNAi were called using the edgeR package (v.1.4.1) setting a false discovery rate threshold of q = 0.01 [39]. The MNase-seq and H2AvD read centers between 100bp upstream and 50bp downstream of each TSS or 100bp around each intergenic hcGAF peak were used in edgeR to call significantly changed promoters.
The Fisher’s exact test showing GAF-bound genes are enriched for transcriptional activity compared the number of transcriptionally active genes for GAF-bound genes (1580 out of 1939) to all genes (4102 out of 7513). Because GAF-bound genes are dramatically enriched for actively transcribed genes, the Fisher’s exact test showing GAF-bound genes are enriched for pausing compared the number of paused genes for GAF-bound genes (1484 out of 1939) to all other genes (3074 out of 7519). The Fisher’s exact test showing Pause Reduced promoters are more likely to have hcGAF peaks than Pause Unchanged promoters compared the number of hcGAF-bound Pause Reduced promoters (87 out of 134) to hcGAF-bound Pause Unchanged promoters (320 out of 1078).
A binomial test was used to show that genes with statistically significant gene body changes are more likely to be down-regulated than up-regulated, assuming that gene body changes are equally likely to be up- or down-regulated. A binomial test was also used to test whether there is a correlation between changes at the promoter and in the gene body, at genes with significant changes in promoter read counts. To address this, we asked whether genes with significantly reduced promoter GRO-seq reads are more likely to have a positive or negative gene-body log fold-change than expected by chance, under an equal probability for reduced and increased reads.
Quantifications used in the graphs were obtained from the genome-wide datasets for factors/modifications created from S2 cells. For the Pearson correlation in S3 Fig, the level of various factors, histones, and histone modifications was calculated within 500bp of each TSS for GAF-bound promoters and compared to the ratio of promoter GRO-seq reads in GAF-RNAi and LacZ-RNAi libraries. For composite profiles, factor intensity was calculated at each base, unless otherwise indicated. The composite profiles in Fig. 4, S3 Fig, and S4 Fig are the median from 1000 samplings of 10% of genes, and the shaded areas in Fig. 4 and S3 Fig indicate the 10% and 90% confidence intervals. For the enrichment barplots in Fig. 4 and S3 Fig, the GFF3 files for each modENCODE factor and MACS peak bed files for each ChIP-seq datasets were used to identify enriched regions in the ChIP datasets that overlap the TSS.
The genomic data in this work is deposited in the Gene Expression Omnibus under the accession numbers: GSE58957 and GSE40646.
|
10.1371/journal.ppat.1006350 | NLRX1 negatively modulates type I IFN to facilitate KSHV reactivation from latency | Kaposi’s sarcoma-associated herpesvirus (KSHV) is a herpesvirus that is linked to Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman’s disease (MCD). KSHV establishes persistent latent infection in the human host. KSHV undergoes periods of spontaneous reactivation where it can enter the lytic replication phase of its lifecycle. During KSHV reactivation, host innate immune responses are activated to restrict viral replication. Here, we report that NLRX1, a negative regulator of the type I interferon response, is important for optimal KSHV reactivation from latency. Depletion of NLRX1 in either iSLK.219 or BCBL-1 cells significantly suppressed global viral transcription levels compared to the control group. Concomitantly, fewer viral particles were present in either cells or supernatant from NLRX1 depleted cells. Further analysis revealed that upon NLRX1 depletion, higher IFNβ transcription levels were observed, which was also associated with a transcriptional upregulation of JAK/STAT pathway related genes in both cell lines. To investigate whether IFNβ contributes to NLRX1’s role in KSHV reactivation, we treated control and NLRX1 depleted cells with a TBK1 inhibitor (BX795) or TBK1 siRNA to block IFNβ production. Upon BX795 or TBK1 siRNA treatment, NLRX1 depletion exhibited less inhibitory effects on reactivation and infectious virion production, suggesting that NLRX1 facilitates KSHV lytic replication by negatively regulating IFNβ responses. Our data suggests that NLRX1 plays a positive role in KSHV lytic replication by suppressing the IFNβ response during the process of KSHV reactivation, which might serve as a potential target for restricting KSHV replication and transmission.
| Kaposi’s sarcoma-associated herpesvirus (KSHV) is linked to a number of different human cancers, including Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman’s disease (MCD). KSHV predominantly establishes life-long latency in the infected host. Lytic reactivation from latency is critical for KSHV survival and replication. During KSHV reactivation from latency, host innate immune responses are activated to restrict viral replication. Here, we report that NLRX1, a negative regulator of the type I interferon response, is important for optimal KSHV reactivation and subsequent lytic replication.
| Kaposi’s sarcoma-associated herpesvirus (KSHV), also known as human herpesvirus 8 (HHV-8), is a linear double-stranded DNA virus. It is causally linked with Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman’s disease (MCD) [1, 2]. In the majority of KS lesion cells or PEL cells, KSHV maintains latent infection, where KSHV lytic gene expression can only be detected in a small subset of cells [3]. While latent infection is important for maintaining the viral reservoir, reactivation of KSHV from latent infection is important for viral particle production and transmission of KSHV [4, 5]. The balance between lytic and latent infection is critical for KSHV pathogenicity. During reactivation, many cellular signaling pathways are activated and regulated by both host and viral proteins. For example, KSHV infection has been shown to trigger innate immune responses through pattern recognition receptors (PRRs) that recognize pathogen-associated molecular patterns (PAMPs), which results in interferon and pro-inflammatory cytokine production (reviewed in [6]). KSHV encodes multiple proteins that modulate host immune responses to facilitate viral replication [7–11]. However, a better understanding of the role of cellular genes in viral reactivation is also critical for understanding the pathogenesis associated with KSHV.
NLRX1 (also known as CLR11.3 and NOD9) is a member of nucleotide-binding domain, leucine-rich repeat-containing protein family (also known as NOD-like receptors, NLRs) [12–14]. NLRs were originally widely studied as sensors or receptors of inflammasome signaling, but were recently reported to regulate type I interferon production as well (reviewed in [15, 16]). NLRX1 was identified as a negative regulator of RIG-I like receptor (RLR) dependent type I interferon production [12]. NLRX1 localizes to the mitochondria and associates with MAVS (also known as Cardif, VISA and IPS-1), a mitochondria-localized adaptor downstream of RIG-I, to disrupt cytosolic RNA induced type I interferon responses. Depletion of NLRX1 led to enhanced antiviral responses [12]. Notably, NLRX1 was also identified to facilitate HIV-1 primary infection through negative regulation of the cytosolic DNA sensing pathway. NLRX1 was shown to sequester the DNA-sensing adaptor STING from interaction with TBK-1, and therefore block sufficient type I interferon production. In a similar vein, NLRX1 deficient mice also exhibited stronger innate immune responses and thus reduced HSV1 replication [17]. In correlation with these findings, in SIV infection of rhesus monkeys, expression of NLRX1 was negatively correlated with a type I interferon signature [18]. Therefore, in the context of viral infection, NLRX1 might be a key factor in determining the battle between host and virus.
KSHV reactivation has been previously reported to activate MAVS-dependent cytosolic RNA sensing pathways. Blocking MAVS was shown to downregulate type I interferon and thus facilitate KSHV lytic replication [19]. Therefore, as a MAVS negative regulator, it is plausible that NLRX1 plays an important role in KSHV reactivation, and functions as a potential target for restricting KSHV transmission.
To investigate the role of NLRX1 as a potential restriction factor against KSHV lytic reactivation from latency, we utilized KSHV latently infected epithelial cells, iSLK.219 cells [20]. We used control siRNA or siRNA against NLRX1 to specifically deplete NLRX1 in iSLK.219 cells before reactivation. The iSLK.219 cells harbor a latent BAC16-KSHV with a constitutive GFP marker, a doxycycline (Dox)-inducible RTA, which is necessary and sufficient to induce lytic reactivation, and a RFP marker driven by the promoter of the KSHV lytic gene, PAN, which is activated during KSHV lytic replication. Therefore, GFP serves as a marker for infected cells while RFP serves as a marker for KSHV lytic replication. As shown in Fig 1A, while Dox successfully induced KSHV reactivation and lytic replication in both non-specific (NS) siRNA and NLRX1 siRNA transfected cells at each time point, there were lower percentages of RFP positive cells in the NLRX1 depleted samples. Moreover, while we did not observe significant GFP variation, RFP intensity quantitation showed significant inhibition of RFP intensity in the NLRX1 siRNA transfected samples compared to the control NS siRNA transfected samples. This suggests that KSHV lytic replication is suppressed in NLRX1 depleted cells (Fig 1B and 1C). We also tested a second NLRX1 siRNA (siNLRX1 #2) and examined its effect on viral reactivation by investigating RFP fluorescence. The results with the second NLRX1 siRNA was similar to the first siRNA (S2A and S2B Fig). We next monitored KSHV viral particles in the supernatant as well as in infected iSLK.219 cells from each group harvested at 24, 48 and 72 hours post reactivation. Consistent with RFP expression levels, we detected significantly fewer KSHV genomes in the NLRX1 siRNA transfected samples compared to control NS siRNA transfected samples as determined by KSHV genome copy number (Fig 1D and 1E). NLRX1 knockdown efficiency was checked by qRT-PCR (Fig 2A).
To determine the impact of NLRX1 on viral genes, we examined KSHV lytic gene expression. Knockdown efficiency of NLRX1 was monitored by qRT-PCR (Fig 2A, S2E Fig). NLRX1 deficient cells showed a reduced ability to induce lytic gene transcription (such as ORF57, vIRF1 and K8.1) than the control NS siRNA transfected cells (Fig 2B and 2C; S2C and S2D Fig). Both ORF45 and K8alpha lytic proteins were downregulated at 24, 48 and 72 hours post reactivation in the NLRX1 siRNA group (Fig 2D). Knockdown efficiency of NLRX1 was also monitored by immunoblot analysis (Fig 2D). To broadly profile KSHV viral gene expression during reactivation in control versus NLRX1 depleted samples, we also performed KSHV whole genome transcriptional profiling. As seen in Fig 2E, Dox treatment successfully induced KSHV gene expression, and depletion of NLRX1 led to a suppression and delay of viral genome transcription at each time point that we tested, which correlates well with our qRT-PCR data (Fig 2E). We also monitored viral gene transcription at later time points, and we observed a smaller difference between the siNS and siNLRX1 groups at later time points (S3A–S3C Fig).
We next probed for the mechanism by which NLRX1 depletion restricts KSHV lytic reactivation from latency. In order to rule out the possibility that siNLRX1 affects the induction of RTA, we tested if siNLRX1 would affect RTA mRNA transcription induced by Dox. Because RTA transcription in iSLK.219 cells can be both affected by Dox and KSHV reactivation, we used iSLK.RTA cells without KSHV infection to test if siNLRX1 would affect Dox induced RTA. As shown in S3D and S3E Fig, knockdown of NLRX1 did not attenuate RTA mRNA transcription.
Since NLRX1 has been reported to negatively regulate type I interferon, we then investigated if NLRX1 altered type I interferon responses upon KSHV reactivation. As seen in Fig 3A, loss of NLRX1 did result in stronger ifnb transcriptional activity. IFNβ is known to activate the JAK/STAT pathway to induce ISGs for effective antiviral responses against viruses. To investigate this further, we next tested whether the upregulated ifnb transcription level in NLRX1 deficient cells led to the activation of the JAK/STAT pathway. We performed microarray analysis of genes in the JAK/STAT pathway using KSHV infected cells that were transfected with either control or NLRX1 siRNAs at timepoints of 0 hour, 24 hours and 48 hours post reactivation. We compared JAK/STAT responsive genes that were activated in cells transfected with NLRX1 and NS siRNA at each time point. At 0, 24 and 48 hours, many JAK/STAT pathway genes were induced at least 2 fold higher in NLRX1 depleted cells compared to NS siRNA transfected cells, indicating a higher potential for JAK/STAT pathway upregulation upon NLRX1 depletion (Fig 3B–3D). Moreover, as shown in Fig 3E and 3F, at 24 hours post Dox treatment, 16 genes were upregulated by 2 fold in the NS siRNA control samples while 48 genes were upregulated at least 2 fold in the in NLRX1 siRNA transfected samples. At 48 hours post Dox treatment, 20 genes were upregulated by more than 2 fold in the NS siRNA control samples while 46 genes were upregulated at least 2 fold in the NLRX1 siRNA transfected samples (Fig 3G and 3H). In summary, NLRX1 deficient cells induce a much wider variety of JAK/STAT pathway related genes at each time point tested in KSHV reactivated cells compared to the control cells. Fig 3I and S1 Table depicts the overall gene expression of all 84 JAK/STAT related genes tested. As shown, most genes exhibit a higher expression level in NLRX1 deficient cells compared to control cells at each time point tested. We noticed that many JAK/STAT related genes were upregulated in the absence of KSHV lytic reactivation. To eliminate that this is not due to an off target effect of the NLRX1 siRNA, we also tested an alternative NLRX1 siRNA (siNLRX1 #2) to monitor the activation of the JAK/STAT pathway before KSHV was reactivated. As shown in S4A–S4D Fig, we have also observed many upregulated genes. Specifically, of 57 upregulated genes in siNLRX1 #2 group, 54 of the genes were also upregulated in siNLRX1 group, compared to the siNS group. We also transiently overexpressed NLRX1 in iSLK.219 cells to explore the role of NLRX1 in KSHV reactivation. Consistent with the siRNA experiments, we detected attenuated levels of IFNβ mRNA and increased levels of viral ORF57 mRNA (S1A and S1B Fig). NLRX1 mRNA levels were monitored by qRT-PCR (S1C Fig).
NLRX1 was identified as a negative regulator of MAVS. Therefore, in order to further detail the mechanism by which NLRX1 facilitates KSHV reactivation, we looked at IFNβ regulation by MAVS. MAVS has previously been shown to play a role in limiting KSHV reactivation. Therefore, we hypothesized that NLRX1 blocks MAVS to attenuate type I interferon responses in KSHV-infected cells, and thus facilitates KSHV reactivation. To prove this, we first tested the integrity and functionality of the MAVS signaling pathway. As shown in Fig 4A, we transfected poly I:C into KSHV infected iSLK.219 cells and successfully triggered MAVS dependent IFNβ induction, which is indicative of a functional RLR pathway. Moreover, NLRX1 depletion in these cells resulted in higher induction of IFNβ, suggesting NLRX1 blocks MAVS signaling in KSHV-infected iSLK.219 cells without reactivation. NLRX1 knockdown efficiency was monitored by qRT-PCR (Fig 4B). We also tested if poly I:C, an activator of the MAVS pathway, could mimic the effect of NLRX1 knockdown to inhibit KSHV reactivation. As shown in S5A–S5F Fig, poly I:C induced IFNβ during reactivation, and this correlated with inhibition of KSHV reactivation as determined by RFP fluorescence and viral lytic gene expression.
To further test if NLRX1 blocks MAVS signaling in KSHV-infected cells upon reactivation, we used BX795, an inhibitor of TBK1, which acts directly downstream of MAVS. As shown in Fig 4C, when we treated cells with NLRX1 siRNA and BX795, we observed significant inhibition of IFNβ compared to the NLRX1 siRNA and vehicle only treated group at 24 and 48 hours post reactivation. IFNβ levels in the NLRX1 siRNA and BX795 treated group were similar to that of the NS siRNA treated group. NLRX1 knockdown efficiency was monitored by qRT-PCR (Fig 4D). To eliminate the possibility that NLRX1 directly acts as a negative regulator of IRF3, we tested whether NLRX1 could modulate IFNβ promoter luciferase activation by an IRF3 super activator (SA). As shown in S1D–S1F Fig, NLRX1 overexpression inhibited dRIG-I- or MAVS- dependent activation of the IFNβ promoter, but the IRF3(SA) activated IFNβ promoter activation was not affected. These data suggest that NLRX1 inhibits MAVS in KSHV-infected cells and supports KSHV lytic replication.
We next investigated whether BX795 inhibition of the MAVS-TBK1 node would rescue the effect of NLRX1 depletion on KSHV reactivation. As shown in Fig 5A, while NLRX1 depletion resulted in less RFP positive cells than the control NS siRNA group, BX795 treatment partially rescued the block of lytic replication. RFP intensity quantitation also showed significant inhibition of RFP intensity in NLRX1 siRNA treated samples compared to the control group, and a subsequent increase of RFP intensity when the NLRX1 siRNA group was treated with BX795 (Fig 5C). No significant variations were observed in GFP intensity among groups (Fig 5B). We then monitored KSHV viral particles from the reactivated cells from each group harvested at 24, 48 hours and 72 hours post reactivation. We detected significantly fewer KSHV virions in cells transfected with NLRX1 siRNA compared to control NS siRNA samples, and a partial rescue of KSHV viral genomes upon BX795 treatment (Fig 5D). Similar patterns were observed when we monitored KSHV ORF57 gene transcription levels in reactivated iSLK.219 cells (Fig 5E). NLRX1 depletion led to a significant inhibition of ORF57 gene transcription, but this was partially rescued by BX795 treatment (Fig 5E).
To further corroborate our experimental data obtained with BX795 treatment, we also utilized TBK1 specific siRNA. As shown in Fig 6A, while NLRX1 depletion resulted in fewer RFP positive cells than the control NS siRNA sample, siNLRX1+siTBK1 treatment rescued the block to KSHV lytic reactivation and replication. RFP intensity quantitation also showed significant inhibition of RFP intensity in NLRX1 siRNA treated samples compared to the control samples and a recovery of RFP intensity when the NLRX1 siRNA group was co-transfected with siNLRX1 and siTBK1 (Fig 6C). No significant variations were observed in GFP intensity among groups (Fig 6B). We also monitored multiple KSHV viral gene transcripts in reactivated iSLK.219 cells. As shown in Fig 6D–6G, NLRX1 depletion led to significant inhibition of ORF57, K8.1 and vIRF1 gene transcription, but they were all rescued when the cells were co-transfected with TBK1 siRNA. NLRX1 knockdown efficiency was monitored by qRT-PCR as shown in Fig 6D. We have also tested if TBK1 knockdown alone promoted KSHV replication by examining viral lytic gene transcription by qRT-PCR. As shown in S6A–S6D Fig, TBK1 knockdown resulted in elevated transcription of viral genes, such as orf57, virf1 and k8.1.
We also investigated NLRX1’s role in KSHV infected PEL cells. BCBL-1 is a KSHV-infected B lymphoma cell line. We transfected NS or NLRX1 siRNA into BCBL-1 cells, and then induced lytic replication of KSHV by addition of TPA and sodium butyrate (NaB) as previously described [11]. The cells and supernatant were harvested at 0 hour, 24 hours and 48 hours post reactivation, and KSHV genome copy number was determined by qRT-PCR. As shown in Fig 7A and 7B, NLRX1 depletion resulted in significant inhibition of KSHV viral replication both in the cells and in the supernatant, as determined by the genome copy number. We also tested KSHV gene transcription levels in reactivated BCBL-1 cells transfected with NLRX1 siRNA or NS siRNA. NLRX1 depletion resulted in significant inhibition of lytic gene transcription such as ORF57 (Immediate early), ORF36 (early), and K8.1 (late) than NS siRNA transfected cells (Fig 7D–7F). NLRX1 knockdown efficiency was monitored by qRT-PCR as shown in Fig 7C. Furthermore, we also introduced another siRNA (siNLRX1 #2) and performed qRT-PCR assay to monitor the status of KSHV lytic genes in reactivated BCBL-1 cells. As shown in S7A–S7D Fig, we observed that both NLRX1 siRNAs modulated KSHV reactivation similarly in BCBL-1 cells.
Because NLRX1 is a negative regulator of IFNβ, we tested if NLRX1 blocked type I interferon responses upon KSHV reactivation. As seen in Fig 8A, NLRX1 depletion enhanced ifnb transcriptional activity compared to the NS siRNA group, confirming NLRX1’s role in restricting IFNβ. We then performed microarray analysis of the JAK/STAT pathway in BCBL-1 cells to explore NLRX1’s effect on KSHV reactivation. We compared genes that were activated in cells treated with NLRX1 siRNA or NS siRNA at 0, 24 and 48 hours post reactivation. At 0, 24 and 48 hours, a significant number of genes were induced at least 2 fold higher in NLRX1 siRNA transfected cells compared to NS siRNA transfected cells, indicating a higher potential of JAK/STAT pathway upregulation when NLRX1 is depleted in BCBL-1 cells (Fig 8B–8D). Moreover, at 24 hours post reactivation, 21 genes were upregulated by more than 2 fold in the siNS group while 41 genes in the siNLRX1 group were upregulated by 2 fold. At 48 hours post reactivation, 29 genes were upregulated by more than 2 fold in the siNS group while 49 genes in the siNLRX1 group were upregulated by more than 2 fold. These data suggest that NLRX1 deficient cells exhibited induction of a much wider variety of JAK/STAT pathway related genes at each time point tested compared to the control groups (Fig 8E–8H). Fig 8I and S2 Table summarizes the overall gene expression of all 84 JAK/STAT related genes tested. As shown, most genes exhibit higher expression levels in NLRX1 depleted cells compared to control cells at each time point tested.
KSHV reactivation from latency is a complex process for both virus and host, which is tightly regulated by a variety of signaling pathways. Efficient lytic replication of KSHV requires disruption of restrictive signaling pathways that keep the virus latent. This can be achieved by the action of either viral proteins or host proteins. For example, upon KSHV reactivation, cytosolic RNA and DNA dependent pathways were reported to be activated and type I interferon was produced to suppress viral replication [8, 19]. Previously, we have reported that in order to facilitate viral lytic replication, KSHV encodes multiple proteins to inhibit type I interferon production, such as vIRF1. vIRF1 knockdown in the context of viral reactivation can result in enhanced IFNβ production and insufficient reactivation [8]. While viral inhibitors of type I interferon are important for KSHV lytic replication, in this study, we focused on exploring the role of a host type I interferon inhibitor, NLRX1, during KSHV reactivation. We demonstrated that NLRX1 is required for optimal KSHV reactivation. NLRX1 deficiency in iSLK.219 cells led to enhanced type I interferon production, as well as global suppression of KSHV genome transcription activity, decreased level of lytic proteins, and attenuated virion production. A similar phenotype was observed in BCBL-1 cells as well, suggesting NLRX1 is critical for KSHV reactivation and subsequent replication in multiple cell lines. We noticed that viral genes clustered together based on their expression patterns following NLRX1 depletion. A future goal will be to further explore the differences among these different clusters to better understand the regulation of KSHV reactivation.
NLRX1 was previously reported to suppress the RIG-I-MAVS signaling pathway, which is triggered by cytosolic RNA [12]. It has also been previously reported that KSHV reactivation generates dsRNA intermediates that can trigger RIG-I-MAVS signaling in KSHV-infected cells [19]. As shown in our study, we demonstrate the functionality of the MAVS-dependent pathway in both KSHV-infected iSLK.219 and BCBL-1 cells. More importantly, NLRX1 negatively regulated MAVS-dependent IFNβ transcription before and after KSHV reactivation in both iSLK.219 and BCBL-1 cells, indicating that NLRX1 regulates KSHV reactivation through a MAVS-dependent pathway. Inhibition of type I interferon induction by the TBK1 antagonist, BX795, mitigated the effect of NLRX1 deficiency on KSHV lytic replication.
We have also explored TBK1’s effect on KSHV lytic reactivation. Although this is the first time that TBK1 was reported to be a negative regulator of KSHV lytic reactivation, TBK1 was previously reported as a restriction factor of RNA viruses, such as Newcastle Disease virus (NDV) and Sendai virus (SeV), by acting downstream of MAVS and positively regulating IFN responses [21, 22]. This correlates with our previously published results that MAVS play a negative role in KSHV reactivation [19]. Moreover, TBK1 also inhibits replication of HSV-1, an alpha herpesvirus [23].
Although NLRX1 plays an important role in KSHV reactivation, we did not observe significant upregulation or downregulation of NLRX1 at either the transcript or protein level during KSHV reactivation. Therefore, it is possible that NLRX1 serves as a steady-state negative regulator of type I interferon to facilitate KSHV reactivation. However, it is also plausible that NLRX1 may undergo some type of post translational modifications upon KSHV reactivation, which might further benefit KSHV lytic replication. A third possibility is that during lytic reactivation, a KSHV encoded protein(s) might also bind directly or indirectly to the NLRX1 signaling complex to regulate its function. In sum, we report for the first time that NLRX1 plays a pivotal role in modulating KSHV reactivation from latency.
iSLK.219 (doxycycline-inducible SLK cells harboring latent rKSHV.219) (a kind gift from D. Ganem) were maintained in DMEM (Corning) supplemented with 10% FBS (Sigma), 1% penicillin and streptomycin (Corning), G418 (250 μg/ml) (Sigma), hygromycin (400 μg/ml) (Corning), and puromycin (10 μg/ml) (Corning). BCBL-1 cells (a kind gift from D. Ganem) were maintained in RPMI (Corning) medium supplemented with 20% FBS, 1% penicillin and streptomycin (Corning), 1% L-glutamine (Corning), and 0.05 mM β-mercaptoethanol (Sigma). All cells were maintained at 37°C in a 5% CO2 laboratory incubator subject to routine cleaning and decontamination. poly(I:C) was purchased from Invivogen. Antibodies were obtained from the following sources: mouse anti-NLRX1 (Jenny Ting laboratory), KSHV ORF45 (MA5-14769) (Thermo Scientific), Goat anti β-actin-HRP (1615) (Santa Cruz), KSHV K8.1alpha (SC-57889; Santa Cruz). The NLRX1 antibody was previously reported [12]. The pCIG2-Puro and pCIG2-NLRX1-FLAG plasmids were previously described [17]. The dRIG-I and MAVS expression plasmids were previously described [12]. pRL-CMV renilla vector was obtained from Promega. IFNβ promoter luciferase was a generous gift from Zhijian Chen, University of Texas Southwestern, Dallas. pUNO-IRF3sa was obtained from Invivogen.
iSLK.219 cells were maintained as described above and were transfected using Lipofectamine RNAiMAX (Life Technologies) according to the manufacturer’s instructions. At 48 hours post-transfection, the medium was changed to DMEM containing 1% Pen-Strep, 10% FBS, and 0.2 μg/ml of doxycycline for reactivation [20]. BCBL-1 siRNA transfections were performed using Lonza nucleofector V kit according to the manufacturer’s recommendations (Lonza). BCBL-1 PEL cells were reactivated with 1 mM sodium butyrate (Sigma) and 25 ng/ml 12-O-tetradecanoyl-phorbol 13-acetate (TPA) (Sigma) where indicated. At 0, 24, 48 or 72 hours post-reactivation, cells and supernatant were collected. RNA was harvested from cells via the RNeasy Plus mini kit (Qiagen) for analysis of levels of viral transcripts. DNA was harvested from both cells and supernatant via DNeasy mini kit (Qiagen) for analysis of genome copy numbers. Protein from cells was harvested for WB analysis.
Chemically synthesized siRNA duplexes were obtained from Dharmacon GE. ON-TARGETplus Non-targeting Control siRNAs #1 (D-001810-01); ON-TARGETplus NLRX1 siRNA (J-012926-10) Sequence: UCGUCAACCUGGUGCGCAA; ON-TARGETplus NLRX1 siRNA #2 (J-012926-12) Sequence: GUGCUGGGUUUGCGCAAGA; ON-TARGETplus TBK1 (29110) siRNA SMARTpool (L-003788-00) Sequences: (J-003788-08) AGAAGGCACUCAUCCGAAA; (J-003788-09) GAACGUAGAUUAGCUUAUA; (J-03788-10) UGACAGCUCAUAAGAUUUA; (J-003788-11) GGAUAUCGACAGCAGAUUA
Total RNA was isolated by using RNeasy RNA extraction kit (Qiagen) and cDNA synthesis was performed using iScript cDNA Synthesis Kit (Bio-rad) according to manufacture protocols. Real-time PCR was performed using a ViiA 6 Real-Time PCR System. A SYBR green assay from Bio-rad was used for human ifnb as well as KSHV ORFs detection. Primers used for SYBR green qRT-PCR were:
KSHV orf57 F: 5’-TGGACATTATGAAGGGCATCCTA-3’; R: 5’-CGGGTTCGGACAATTGCT-3’.
KSHV orf36 F: 5’-TGCGTCCTCTTCCAGTGTTA-3’; R: 5’-GTCAGCAGAGTGTAGCCCAA-3’.
KSHV virf1 F: 5’-CGTGTCCTTTGGTGAAACTG-3’; R: 5’-TCGGCATTATTTCGAGTACG-3’.
KSHV k8.1 F: 5’-AAAGCGTCCAGGCCACCACAGA-3’; R: 5’-GGCAGAAAATGGCACACGGTTAC-3’.
Human actin F: 5’-AAGACCTGTACGCCAACACA-3’; R: 5’-AGTACTTGCGCTCAGGAGGA-3’.
Human ifnb F: 5’-AGTAGGGCGACACTGTTCGTG-3’; R: 5’-GAAGCACAACAGGAGAGCAA-3’.
Human nlrx1 F: 5’-CCTCTGCTCTTCAACCTGATC-3’; R: 5’-CCTCTCGAAACATCTCCAGC-3’.
Human tbk1 F: 5’- CCTCCCTAAAGTACATCCACG-3’; R: 5’- CAATCAGCCATCGTATCCCC-3’.
The relative amount of IFNβ, ORF57, ORF36 and K8.1 mRNA was normalized to actin RNA level in each sample and the fold difference between the treated and mock samples was calculated.
Total RNA was isolated from iSLK.219 or BCBL-1 cells by using RNeasy RNA extraction kit (Qiagen) and cDNA synthesis was performed using iScript™ cDNA Synthesis Kit (Bio-rad) according to the manufacturer’s protocols. Human JAK/STAT Signaling Primer Library was purchased from Realtimeprimers.com (Cat #: HJAK-I), which contains 88 primer sets directed against human JAK/STAT related genes and 8 housekeeping gene primer sets. Fold-Change (2^(- Delta Delta Ct)) is the normalized gene expression (2^(- Delta Ct)) in the Test Sample divided the normalized gene expression (2^(- Delta Ct)) in the Control Sample. Data were analyzed by GENE-E software (https://software.broadinstitute.org/GENE-E/).
We used a real-time qPCR array to quantify all KSHV mRNAs. Briefly, 192 primer pairs were included to target multiple regions towards the 3’ end of each annotated ORF. Multiple reference genes for cellular transcripts were included for normalization. The array results in amplification reactions with similar efficiencies and annealing temperatures and thus allows us to directly compare the expression levels among different mRNAs. qPCR was plated in 384-well plates using the Tecan Freedom Evo liquid handling robot and cycled using Roche LightCycler 480, as previously described. A detailed, step-by-step protocol is available at http://www.med.unc.edu/vironomics/protocols.
Statistical significance of differences in cytokine levels, mRNA levels, viral titers, and luciferase intensity in reporter assay were determined using Student’s t-test. * indicates P<0.05. ** indicates P<0.01.
|
10.1371/journal.pntd.0000787 | Arboviral Etiologies of Acute Febrile Illnesses in Western South America, 2000–2007 | Arthropod-borne viruses (arboviruses) are among the most common agents of human febrile illness worldwide and the most important emerging pathogens, causing multiple notable epidemics of human disease over recent decades. Despite the public health relevance, little is know about the geographic distribution, relative impact, and risk factors for arbovirus infection in many regions of the world. Our objectives were to describe the arboviruses associated with acute undifferentiated febrile illness in participating clinics in four countries in South America and to provide detailed epidemiological analysis of arbovirus infection in Iquitos, Peru, where more extensive monitoring was conducted.
A clinic-based syndromic surveillance system was implemented in 13 locations in Ecuador, Peru, Bolivia, and Paraguay. Serum samples and demographic information were collected from febrile participants reporting to local health clinics or hospitals. Acute-phase sera were tested for viral infection by immunofluorescence assay or RT-PCR, while acute- and convalescent-phase sera were tested for pathogen-specific IgM by ELISA. Between May 2000 and December 2007, 20,880 participants were included in the study, with evidence for recent arbovirus infection detected for 6,793 (32.5%). Dengue viruses (Flavivirus) were the most common arbovirus infections, totaling 26.0% of febrile episodes, with DENV-3 as the most common serotype. Alphavirus (Venezuelan equine encephalitis virus [VEEV] and Mayaro virus [MAYV]) and Orthobunyavirus (Oropouche virus [OROV], Group C viruses, and Guaroa virus) infections were both observed in approximately 3% of febrile episodes. In Iquitos, risk factors for VEEV and MAYV infection included being male and reporting to a rural (vs urban) clinic. In contrast, OROV infection was similar between sexes and type of clinic.
Our data provide a better understanding of the geographic range of arboviruses in South America and highlight the diversity of pathogens in circulation. These arboviruses are currently significant causes of human illness in endemic regions but also have potential for further expansion. Our data provide a basis for analyzing changes in their ecology and epidemiology.
| Over recent decades, the variety and quantity of diseases caused by viruses transmitted to humans by mosquitoes and other arthropods (also known as arboviruses) have increased around the world. One difficulty in studying these diseases is the fact that the symptoms are often non-descript, with patients reporting such symptoms as low-grade fever and headache. Our goal in this study was to use laboratory tests to determine the causes of such non-descript illnesses in sites in four countries in South America, focusing on arboviruses. We established a surveillance network in 13 locations in Ecuador, Peru, Bolivia, and Paraguay, where patient samples were collected and then sent to a central laboratory for testing. Between May 2000 and December 2007, blood serum samples were collected from more than 20,000 participants with fever, and recent arbovirus infection was detected for nearly one third of them. The most common viruses were dengue viruses (genera Flavivirus). We also detected infection by viruses from other genera, including Alphavirus and Orthobunyavirus. This data is important for understanding how such viruses might emerge as significant human pathogens.
| Over the past few decades there has been a global resurgence of arthropod-borne viral pathogens (arboviruses) worldwide [1], [2], particularly those transmitted by mosquitoes. Despite the public health relevance, the geographic range, relative impact, and epidemiologic characteristics associated with arbovirus infection are poorly described in many regions of the world. Arboviruses are a heterogeneous group, but those of medical relevance largely belong to a few virus genera, including Flavivirus, Alphavirus, and Orthobunyavirus. Prominent examples of emergent arboviruses include West Nile virus (WNV; Flavivirus) in North America, Japanese encephalitis virus (JEV; Flavivirus) in Asia, chikungunya virus (CHIKV; Alphavirus) in the Indian Ocean region and dengue viruses (DENV; Flavivirus) worldwide. One common feature shared by many emergent arboviruses is the capacity to expand host and geographical range, owing in part to the plasticity of the RNA genome [3]. Some arboviruses have evolved to exploit humans as their primary reservoir, while others rely on birds or peridomestic animals, with human infection resulting from spill-over from zoonotic replication cycles. Human exposure to sylvatic arbovirus cycles is likely to increase as activities continue to encroach on forested areas worldwide. Tropical areas in particular, with year-round hot and humid conditions, are well-suited for maintaining arboviruses with potential to emerge as significant human pathogens [4]. In the neotropics alone, greater than 145 distinct arbovirus species have been identified [4], many of which have already been associated with human illness.
One limitation of conducting surveillance for arboviral diseases is the generic nature of disease presentation. While severe disease can result, such as hemorrhagic manifestations (DENV and yellow fever virus [YFV]) or neurological disease (WNV, JEV, and Venezuelan equine encephalitis virus [VEEV]), arbovirus infection typically results in mild to moderate febrile illness [2], [5], [6]. Particularly early in disease progression, patients commonly present with undifferentiated febrile illness [5], [7] rendering clinical diagnosis unreliable [8]. In DENV-endemic areas, for example, diseases caused by co-circulating pathogens have been found to be often misclassified [8], [9]. In light of the lack of distinct clinical presentation and the diversity of the etiologic agents, laboratory support has become a critical component of effective surveillance programs.
The impact on human health in endemic regions and the potential for broader spread underscore the importance of improving understanding of arbovirus transmission patterns. Currently the epidemiological characteristics and geographic range for many endemic arboviruses in South America are poorly understood. To begin to address this gap, we established a laboratory-supported clinic-based study to identify the etiologic agents associated with undifferentiated febrile illness in sites in Peru, Ecuador, Bolivia, and Paraguay. Herein we describe the geographic distribution of distinct arboviruses and their relative contribution to human febrile illness in these study sites. In addition, we present the temporal trends and epidemiological characteristics associated with arbovirus infection in Iquitos, Peru, a site where more extensive monitoring was conducted.
In 1990 the U.S. Naval Medical Research Center Detachment (NMRCD) initiated a clinic-based surveillance program to determine the etiologies of febrile illness in Iquitos, Peru [10]–[12]. In 2000 NMRCD collaborated with local Ministries of Health to expand the surveillance program into other regions of Peru and South America, including sites in Ecuador, Bolivia, and Paraguay. In addition to Iquitos, in 2000 the study was implemented at regional sites in or near Piura, Cusco, Tumbes, and Yurimaguas, Peru, as well as Santa Cruz, Bolivia (Figure 1; Table 1). Additional sites were later added in Concepción, Magdalena, and Cochabamba (Villa Tunari and Eterazama), Bolivia; Guayaquil, Ecuador; Asunción, Paraguay; and La Merced and Puerto Maldonado, Perú. Participants were recruited when reporting with acute febrile illness to public, private, or military health facilities in and around these regional centers. Details of the study sites are described in Table 1. Study sites were selected based largely on locations in hot and humid climates conducive for arbovirus transmission, typically situated near or in tropical rainforest regions. Notable exceptions include Piura, Tumbes, and Cusco, which are located in coastal desert (Piura and Tumbes) or highlands (Cusco; Figure 1 and Table 1) regions. It should be noted that the study staff in Cusco (Hospital Regional) on occasion attended to participants arriving from surrounding highlands rainforest regions.
Study protocols (NMRCD.2000.0006 [Peru], NMRCD.2001.0002 [Ecuador], NMRCD.2000.0008 [Bolivia], and NMRCD.2005.0008 [Paraguay]) were approved by the Naval Medical Research Center Institutional Review Board (Bethesda, MD) in compliance with all U.S. Federal regulations governing the protection of human subjects. In addition, the study protocols were reviewed and approved by health authorities in Peru (Dirección General de Epidemiología), Bolivia (Servicio Departamental de Salud, Santa Cruz and Colegio Medico de Santa Cruz), Ecuador (Ministerio de Salud Publica, Comando Conjunto de la Fuerzas Armadas, and Escuela de Sanidad in Guayaquil) and Paraguay (Ministerio de Salud y Bienestar Social and Comité de Ética de Asociación de Rayos de Sol). Written consent was obtained from patients 18 years of age and older. For patients younger than 18 years, written consent was obtained from a parent or legal guardian. Additionally, written assent was obtained from patients between 8 and 17 years of age.
Study subjects included patients 5 years of age or older who presented in outpatient clinics or hospitals with acute, undifferentiated, febrile illness (greater than or equal to 38°C for 7 days duration or less) along with one or more of the following symptoms: headache, muscle, ocular and/or joint pain, generalized fatigue, cough, nausea, vomiting, sore throat, rhinorrhea, difficulty breathing, diarrhea, jaundice, dizziness, disorientation, stiff neck, or bleeding manifestations. Children younger than five years of age were included if they presented with hemorrhagic manifestations indicative of dengue hemorrhagic fever (DHF), including epistaxis, pleural effusion, platelets less than 100,000/ml, petechiae, or bloody stool or vomit. Exclusion criteria included fever in excess of seven days or an identifiable focus of infection, such as sinusitis, pneumonia, acute otitis media, or acute urinary tract infection. Demographic data, medical history, and clinical features for each patient were obtained using a standard questionnaire. In malaria-endemic regions if malaria was suspected, capillary blood from febrile patients was screened for Plasmodium spp. by clinic or hospital personnel according to routine diagnostic procedures at each site. Peripheral blood samples were screened by microscopic analysis of stained thick smear slides. In some sites, owing to the possibility of arbovirus co-infection, malaria-positive patients were subsequently invited to participate in the NMRCD study, with malaria results recorded along with symptoms and demographic information.
During the acute phase of illness blood samples were obtained from each patient, and when possible, convalescent samples were obtained 10 days to 4 weeks later for serological studies. For patients older than 10 years of age, up to 15 mL of blood was collected, and for patients younger than 10 years of age, up to 7 mL of blood was collected. Trained phlebotomists collected blood samples via arm venipuncture using standard methods and universal precautions.
Statistical analyses (Chi-square, Fisher's exact test, and logistic regression) were performed in R version 2.8 (The R Foundation for Statistical Computing, Vienna, Austria)[20]. The significance level was set at α = 0.05.
A total of 20,880 participants from study sites in Bolivia, Ecuador, Paraguay, and Peru were enrolled in the study between May 2000 and December 2007 (Table 2). A total of 18,201 participants (87.2%; Table 2 were included from Peru, 2,089 (10.0%) from Bolivia, 350 from Ecuador (1.7%), and 240 from Paraguay (1.1%). More than half (10,739; 51.4%) of participants were recruited at 13 health clinics or hospitals in and around Iquitos, Peru. For participants where demographic data was available, 10,919 were male (52.3%) and 9,915 were female (47.5%). The median age of participants was 24 (range 0–92 years), with 89.5% between the ages of 6 and 49. In addition to fever, the most commonly reported symptoms included malaise (96.7%), headache (92.4%), chills (90.2%), myalgia (81.4%), arthralgia (76.2%), and hyporexia (75%). A thorough breakdown of symptomology by etiologic agent will be reported elsewhere (TJK, unpublished results).
Of the 20,880 cases of febrile illness included, paired acute and convalescent-phase samples were collected from 13,259 participants (63.5%), while acute-phase only (without convalescent samples) were available from 7,621 participants (36.5%; Table 2). Most participants for whom data was available reported to a health center within four days following disease onset (15,911/19,632; 81.0%). A subset of participants (n = 9,800; 46.9%) were screened for malaria by thick smear prior to enrollment. Of these, 584 (6.0%) were found to be positive. Patient samples were screened for recent infection by members of the Flaviviridae, Togaviridae, and Bunyaviridae virus families, using IFA and IgM ELISA. Evidence for arbovirus infection was observed in 32.5% of febrile cases overall (Table 3), varying significantly across locations (p<0.001) from 9.3% (Cusco) to 39.8% (Iquitos). Of these, diagnoses were considered confirmed for 4,423 febrile episodes (21.2%), including 2,862 positive by IFA or RT-PCR (13.7%; Table 4) with or without accompanying elevated IgM, as well as 1,561 IgM seroconversions (7.5%) without accompanying IFA or RT-PCR confirmation. No arbovirus co-infections were observed by IFA. An additional 2,370 cases (11.4%) were classified as presumptive arboviral infections. Viral isolation and RT-PCR identification was most successful with participant specimens collected within the first four days following disease onset. Of the 15,911 participants reporting within four days post-disease onset (where data was available), 2,456 (15.4%) were IFA-positive for an arbovirus in the acute sample. For those reporting five days or more post-onset, only 3.8% (143/3,721) were IFA-positive.
DENV was the most common malaria co-infection, observed for 11.3% (66/584) of participants reported as malaria-positive, including 17 DENV-3 isolates and one DENV-1 isolate. The rate of DENV infection was significantly lower for malaria-positive participants than for malaria-negative participants (11.3% vs 32.9%; p<0.0001). In contrast, VEEV infection was more common among malaria-positive participants (7.4%) as compared to malaria-negative participants (2.7%; p<0.0001). There were no significant differences between malaria thick smear-positive participants and thick smear-negative participants for the other arboviruses studied (data not shown).
DENV serotypes were the predominant arboviruses detected, accounting for 26.0% of febrile episodes analyzed (Table 5), based on virological (2,662 virus isolates or RT-PCR positives, with or without supporting serology) and serological (1,058 IgM seroconversions and 1,700 participants with elevated DENV IgM, without accompanying positive results by virus isolation or RT-PCR) evidence. Considerable YFV cross-reactivity was observed for DENV-positive samples. Based on the 2,662 cases with definitive DENV diagnosis (IFA or RT-PCR confirmation in the acute sample), 847 (32.0%) also had IgM reactive to YFV antigen in the acute or convalescent sample.
DENV-3 was most commonly isolated serotype, accounting for 81.1% (2,159/2,662) of DENV isolates over the course of the study. In our study, DENV-3 was first detected in sites along the northern coast of Peru (Piura and Tumbes) in 2000 (Figure 2) during a large outbreak of dengue fever in the region [21], [22], although DENV-1 and DENV-2 were the most commonly isolated serotypes during this outbreak. DENV-3 quickly became the dominant serotype in the northeastern rainforest (Iquitos and Yurimaguas), with limited DENV-1 co-circulation in the region in subsequent years (Figure 2). Between 2002 and 2006 little DENV-2 transmission was observed until DENV-2 emerged in the study sites in Bolivia and southern Peru (Puerto Maldonado) in 2007 (Figure 2). DENV-4 was rarely detected during the study period, with only four isolates from study participants. However, more recently this situation has changed dramatically with the 2008 emergence of DENV-4 in northern Peru [23].
Overall, DENV infection was more common among female participants than male participants (28.1% vs 25.5%; p<0.0001), with a statistically significant bias towards older participants (28.0% of participants 30 or older were DENV-positive as compared with 25.9% of participants younger than 30; p = 0.001). However, the epidemiology of DENV infection varied by study site, particularly within Peru. The prevalence of DENV infection was higher among older participants in Puerto Maldonado (p = 0.01), La Merced (p<0.0001), Piura (p<0.0001), and Tumbes (p = 0.015), while no elevated DENV prevalence was observed for older participants in Yurimaguas and Iquitos. In addition, DENV infection was more common among female participants in Puerto Maldonado (p<0.0001), La Merced (p<0.0001), and Tumbes (p = 0.028), but no statistically significant difference was observed in Iquitos, Yurimaguas, or Piura.
Other than DENV, the only flavivirus isolated during the course of the study was YFV, which was isolated from four participants (Table 4). In addition to the four isolates, serological evidence of recent YFV infection (without evidence of DENV infection) was detected in an additional 494 participants, including 143 who seroconverted between acute and convalescent samples. Overall, data on prior vaccination was available for 17,816 participants, 10,667 (59.9%) of whom reported having received YF vaccination. YF vaccine coverage varied widely by study site, ranging from less than 10% in non-endemic sites along the northern coast of Peru (Piura and Tumbes) to 59% in Iquitos, 70% in Yurimaguas, and 77% or greater in Cochabamba (Villa Tunari and Eterazama), Concepción, Junin, Magdalena, and Puerto Maldonado. Study participants with evidence of recent YFV infection based on IgM were significantly more likely than the rest of the overall study population to have reported receiving YF vaccination within the previous 6 months (30.0% vs. 6.7%; p<0.0001).
Recent alphavirus infection was detected for 3.1% (n = 645) of febrile patients (Table 5), including 102 VEEV isolates and 40 MAYV isolates. RNA from a subset of VEEV isolates was extracted, reverse transcribed, amplified, and sequenced. All sequenced isolates were determined to belong to enzootic subtypes of the VEE complex, predominately ID Panama/Peru or Peru/Bolivia genotypes [12], [14], [24] although there was one ID Colombia/Venezuela genotype and one IIID subtype virus identified, both in Iquitos [12]. The majority of VEEV isolates and seroconversions (234/250; 93.6%) were from patients in Iquitos, Puerto Maldonado, and Yurimaguas, Peru. In contrast, MAYV isolates were more prevalent in Bolivia and southeastern Peru (Table 4). Of all MAYV isolations, 57.5% (23/40) were from this region, despite representing only 19.8% of all participants in the study.
EEEV was not isolated from any participant samples during the course of the study. A subset of participant samples were screened for EEEV-reactive IgM (n = 3,014), with serological evidence for EEEV infection in 22 cases (0.7%), including two seroconversions. Unlike the flaviviruses, little serologic cross-reaction (or, alternatively, concurrent infection) was observed among alphaviruses. For the 72 participants with the most well-defined VEEV infections (IFA-positive, plus a convalescent sample available for testing), 7 (9.7%) had IgM reactive to MAYV antigen in either the acute or convalescent sample. For the 24 cases where MAYV was isolated and a convalescent sample was available for testing, no cross-reactivity was observed in the VEEV IgM ELISA.
Arboviruses belonging to the Orthobunyavirus genus of the Bunyaviridae family accounted for approximately 2.5% of all febrile cases (Table 5). In total there were 54 orthobunyavirus isolates, including 30 Group C viruses, 18 OROV isolates, and six GROV (Table 4). The Group C virus isolates were not definitively identified; however, based on serological techniques (ELISA and PRNT), ten were antigenically related to CARV and six were antigenically related to MURV, while 14 could not be antigenically distinguished. Nearly all Simbu Group (OROV) and Group C virus isolates were collected from patients reporting to clinics in Iquitos, Madre de Dios, and Yurimaguas, Peru, while three out of six GROV isolates were obtained from patients in La Merced, Peru, in January and February of 2007 (Table 4). As with the alphaviruses, little serologic cross-reaction was observed within the Orthobunyavirus genus. For the 24 participants IFA-positive for a Group C virus and a convalescent sample available, only 2 had IgM reactive to OROV antigen in either the acute or convalescent sample (8.3%); for the 12 OROV IFA-positive participants with a paired convalescent sample, no reactivity with CARV or MURV antigen was observed.
The Iquitos health centers included in this study cover a geographically stratified area of the city and in 2007 represented nearly 20% (10 of 55) of civilian public health centers in the greater urban health network. Based on the populations assigned to each health center by the local ministry of health (Dirección Regional de Salud -Loreto), in 2007 clinics included in this study were designated to serve approximately 43% of the population of the Iquitos region. Using the population numbers assigned to each health center by DIRESA-Loreto, we estimated incidence rates for the most common arboviral infections in Iquitos beginning with the first full year of the study (Table 6). Over the course of the study, there were 855.9 acute undifferentiated febrile episodes per 100,000 people per year, peaking during periods of highest dengue activity (2002 and 2004; Table 6). DENV incidence rates varied greatly, peaking in 2002 with 554.0/100,000 following the introduction of DENV-3 and averaging 274.7/100,000 over the 7-year period. The average symptomatic incidence rates for other predominant arboviruses were 28.1/100,000 for VEEV, 8.5/100,000 for MAYV, 14.3/100,000 for OROV, and 14.2/100,000 for Group C viruses. Peak transmission rates were observed for these four viruses between 2004 and 2006, including a previously-described outbreak of VEEV in 2006 [24].
It should be noted that the incidence rates above only reflect the participants enrolled in the study. Starting in 2006, demographic data was collected for those who reported to Iquitos health centers and fulfilled the inclusion criteria (acute undifferentiated febrile illness of fewer than 7 days in duration) but declined participation in the febrile surveillance study. In 2006 and 2007, 3,385 and 3,283 febrile patients, respectively, fitting the case definition were examined by study personnel, with 43.3% (n = 1,433) and 36.5% (n = 1,197) of patients agreeing to provide venous blood samples for the surveillance program. During these two years, 94.3% of febrile patients were first screened for malaria by thick smear, with 32.4% of those screened classified as positive. Malaria-negative patients were significantly more likely to accept participation in the surveillance study (51.3%; 2,185/4,256) than malaria-positive patients (9.5%; 193/2,036; p<0.001). Children were significantly less likely to participate than adults (25.3% of eligible children vs. 44.0% of eligible adults chose to participate; p<0.001).
To begin to describe the epidemiology associated with these arboviruses in the Iquitos, demographic characteristics of participants with recent infection by the most common pathogens – DENV serotypes, VEEV, MAYV, OROV, and Group C viruses – were compared with the rest of the participating febrile population in Iquitos. YFV infection, as determined by positive IgM ELISA, was significantly associated with self-reported recent YF vaccination (OR 2.30, 95% CI 1.44—3.57); no similar association was observed for other arboviruses. Thus no further analyses were conducted for YFV IgM-positive participants. Overall, male participants were more common than female (51.4% vs. 48.6%), consistent with the population of Loreto Department as a whole (51.2% male; p = 0.77)[25]. The median age of study participants was 23 (average 26.1), with the highest percentage of participants between the ages of 15 and 29. Both MAYV (p = 0.003) and VEEV (p = 0.009) infection were significantly more common among males, and this effect was only observed among the older age groups (15 years or older), suggesting an occupational exposure. A similar trend for higher prevalence of Alphavirus infection among males was observed in Yurimaguas and Puerto Maldonado, although these analyses were limited by small sample size. Group C virus infection was more common in males in these three sites, although the differences were only statistically significant in Puerto Maldonado (p<0.01). In Iquitos no significant differences were observed between sexes for DENV or OROV (Table 7). DENV infection was significantly more common among participants younger than 15 in Iquitos (p = 0.005); however, this effect was only observed during the earlier years of the study (2002 and 2003 in particular). OROV infection in Iquitos was significantly more common among age groups 15 or older (p = 0.007). For the 30–44 year old age group, MAYV infection was significantly more common than for participants younger than 15 (Table 7).
The health centers in the Iquitos area included in this study were predominantly public clinics and hospitals located within the urban area of the city (n = 8), although the study was also conducted in three military clinics located within the urban area and two public clinics located in rural zones between approximately five and ten kilometers outside the city limits. The majority of participants in the Iquitos area were recruited in the urban clinics (79.6%), while 10.6% and 9.8% were recruited at the two rural clinics and three military clinics, respectively. Using the different categories of clinics as a proxy for potential differences in arbovirus exposure, we compared the relative prevalence of arboviruses among those reporting to the urban, rural, or military clinics. DENV infection was far more common in participants reporting to the urban clinics than rural clinics, whereas VEEV (p = 0.018), MAYV (p<0.001), and Group C viruses (p<0.001) were more common among those reporting to the rural clinics. For OROV infection there was no statistically significant differences among the types of health centers (Table 7).
Participants were recruited year-round, with a peak in December that was largely due to DENV transmission (Figure 3A). Transmission of the arboviruses peaked during different months of the year. Over the course of the study, DENV transmission was most common between October and December with lowest levels between June and August (Figure 3A). Alphavirus transmission was highest between February and July (Figure 3B), with reduced transmission during the second half of the year, while the highest percentage of Group C virus cases was observed between December and February (Figure 3C).
Tropical areas, with year-round hot and humid conditions, are particularly well-suited for maintaining arboviruses with both current public health importance as well as the potential to emerge as significant human pathogens [4]. Therefore in this study we focused on arbovirus transmission in tropical regions of four countries in South America. Our data demonstrate that arboviruses are a common cause of human febrile illness in these sites in South America, accounting for greater than 30% of the febrile episodes analyzed. Importantly, arbovirus-associated disease was not restricted to DENV in most of the locations studied. The other arboviruses identified, including VEEV, MAYV, and OROV, in total were associated with approximately 8% of febrile episodes. Our study has provided source material for various phylogenetic analyses [12], [14], [19], [23], [26] and will provide important baseline data for monitoring changes in arbovirus ecology, epidemiology, and genetics.
There were several significant limitations to our study. First, with the exception of Peru, the number of study sites in the other countries was quite limited. Even in Peru, it is unclear whether these results are indicative of arbovirus circulation in other regions of the country. Another shortcoming was the focus on arboviruses. While clearly these are important pathogens in the tropical rainforest regions, along the desert coast (Piura) and in the highlands (Cusco), other types of pathogens will need to be given greater consideration. Another limitation of our study design is the passive surveillance strategy employed. Clinic-based surveillance is likely to significantly underestimate true arbovirus circulation, as those with milder disease manifestations are less likely to visit a health center. In studies of DENV transmission in Iquitos we have observed that incidence rates calculated from community-based active surveillance are several times higher than those calculated based on passive surveillance (TJK, ACM, and BMF, unpublished results). Accordingly the incidence rates presented here should be interpreted carefully and considered a conservative estimate of the true number of febrile episodes caused by each virus. Another shortcoming of clinic-based surveillance is the difficulty of extrapolating the data to the entire population. As we show here in Iquitos, those who present to the health centers and those that are willing to participate in these studies are often not fully representative of the population at-large, which may lead to biases in age-dependent incidence rates. In addition, with the exception of Iquitos, we did not collect sufficient data from non-participants to fully contextualize these results. Our data suggest that malaria may contribute to approximately 30% of acute febrile illnesses in Iquitos, a figure that would not be apparent based solely on those who enrolled in the study. One advantage of clinic-based passive surveillance is expanded geographic coverage and more limited costs relative to other surveillance strategies, which is critically important when studying the relatively obscure arboviruses described here. Only through the large number of participants presented here were we able to detect sufficient cases of VEEV, MAYV, and OROV for further epidemiological analysis.
Over the course of the study, DENV serotypes were by far the most common arboviruses associated with febrile disease, accounting for 26% of febrile participants. DENV serotypes have emerged dramatically in Latin America over the past decades, to the point that nearly a million cases of dengue fever are reported every year in Latin America, along with thousands of cases of more severe disease that may lead to hemorrhagic manifestations and death [27], [28]. Here we demonstrate that DENV-3 (previously identified as subtype III [26]) was the predominant serotype in the region between 2001–2007, although we also observed significant DENV-1 and DENV-2 transmission in certain regions. Not surprisingly DENV circulation was found to be more region-dependent than country-dependent. Specifically, Tumbes and Piura along the coast of northern Peru share common trends with Guayaquil in Ecuador (Figure 2A), while DENV circulation in Puerto Maldonado in southern Peru is more closely tied to trends observed in Bolivia (Figure 2C). More recently we observed that a genetically conserved strain of DENV-4 was identified in Ecuador (2006), then coastal Peru (2007), before spreading to the tropical rainforests of northeastern Peru (Iquitos and Yurimaguas; 2008) [23]. As multiple serotypes have been circulating in the region severe disease resulting from heterologous secondary infection is increasingly likely to occur [27], [29]. In this study we did not distinguish between primary and secondary infection, and thus further analysis will be needed to identify the genotypes [30] and prior DENV immune status associated with more severe disease outcomes in the region. Regardless, the data described here will provide a springboard for future studies of regional DENV maintenance and dispersion patterns [31] as well as analysis of genetic adaptation and selective pressures.
Other than DENV, the only other flavivirus isolated was YFV. One well-documented hindrance to study flavivirus is the cross-reaction observed among even disparate species [32], [33]. Similarly here we observed significant cross-reaction between DENV and YFV antigen in serum from patients with defined DENV infection, thus there is a possibility that some of the cases have been misclassified. For YFV, we only considered those instances where there was no DENV IgM detected. Furthermore, there was a strong correlation between participants reporting recent YF vaccination and having YFV-reactive IgM, suggesting that these results were not due to cross-reactivity with other flaviviruses circulating in South America. Low grade fever and headache are not uncommon outcomes within the two weeks following YF vaccination [34], [35], so it is possible that these cases are due to the vaccination. It should be noted, however, that flavivirus IgM can be long-lived [32], and thus many of these febrile episodes classified as “YFV infection” may not represent the true etiologic agent. In addition, in our study we only rarely observed severe disease associated with YFV-reactive IgM, suggesting that these cases largely do not reflect natural infection and thus should be interpreted with caution.
In addition to DENV and YFV, there are other flaviviruses circulating in the region that need to be considered, including WNV, Rocio virus (ROCV), Ilheus virus (ILHV), and St. Louis encephalitis virus (SLEV). These flaviviruses have been isolated either from mosquitoes [36], birds [37], or mammals [38], including humans [39], [40], in parts of South America. More closely related to our study areas, ILHV has been isolated from a febrile patient in Ecuador [41], and ILHV and SLEV have been isolated from mosquitoes in Iquitos [36], clearly demonstrating that these viruses are circulating near human populations in the region. None of these viruses were isolated from participants in our study, suggesting that human infection is uncommon. However, in a preliminary survey of a subset of our participants we have identified cases where ROCV, ILHV, SLEV, or WNV IgM was detected, with no reactivity with DENV or YFV antigen (data not shown), with confirmation by virus neutralization assay, considered the most specific tool for flavivirus serology [42]. Overall, the cross-reactivity reported here and elsewhere [32] and the longevity of flavivirus IgM underscore the complications of flavivirus serodiagnosis, which represents a great hindrance for epidemiological surveillance.
The most common Alphavirus species identified were VEEV and MAYV. Scant evidence for human infection with EEEV was identified, consistent with previous reports [43], despite evidence of EEEV circulation in mosquitoes near Iquitos [36], [44], for example. In light of the recent emergence of another alphavirus, CHIKV, in the Indian Ocean region, VEEV and MAYV represent interesting cases to consider with regards to potential for urban emergence. In laboratory studies Aedes spp., the primary vectors for DENV, have been shown to be a competent vector for VEEV [45] and MAYV [46]. Even without adapting to human-Aedes-human cycles, epizootic VEEV subtypes have been associated with large outbreaks of human disease across South America [6]. As recently as 1995 a VEEV outbreak was responsible for nearly 100,000 febrile cases in Venezuela and Colombia [47], [48]. While the VEEV strains isolated in our study all belong to enzootic genotypes of the virus complex [12], [14], [24], genetic studies have demonstrated that enzootic and epizootic subtypes are closely related. A modest number of amino acid changes can alter the viral phenotype dramatically, converting an enzootic strain to an epizootic strain [49]–[52]. Similarly, amino acid variants in the CHIKV E1 protein have been associated with increased epidemic potential [53]–[55]. Several other factors further suggest that potential for neotropical alphavirus emergence is high. In the Iquitos area, while we found that VEEV was more commonly associated with rural clinics (Table 7), many of the participants with confirmed VEEV infection lived within the city and did not report leaving the urban area during the month prior to the febrile illness [24](data not shown). This data is corroborated by a previous study of healthy participants, in which we found that nearly 25% of the urban population carries VEEV-neutralizing antibodies [24]. In addition, based on data collected through this program the geographic range of MAYV and VEEV is wider than had been previously demonstrated, extending to southern Peru and Bolivia [14], [19]. Taken together these factors suggest that the potential establishment of the neotropical alphaviruses as urban pathogens should be closely monitored.
In addition to the flaviviruses and alphaviruses, orthobunyaviruses were significant sources of febrile illness in the study, accounting for 2.5% of febrile episodes analyzed. While all orthobunyavirus isolates came from patients in Peruvian rainforest sites during the course of this study, we did find serological evidence for OROV and Group C viruses in Ecuador and Bolivia. More recently (2008) we have definitively identified Group C viruses in Bolivia, isolated from two participants in the Cochabamba region (data not shown). Like VEEV and MAYV, OROV is an interesting case study with regards to potential for broader emergence. OROV has been associated with numerous outbreaks in Brazil, infecting approximately 500,000 people in South America over the past 45 years [56]–[58]. Two distinct transmission cycles have been proposed, a poorly-defined sylvatic cycle and an urban cycle where OROV is transmitted among humans by the biting midge Culicoides paraensis [58], [59]. In Iquitos, we found that unlike the Group C viruses, VEEV, and MAYV, evidence of recent OROV infection (based on both IgM and virus isolation data) showed no significant bias towards rural clinics, suggesting similar transmission levels between urban and rural sites, consistent with results from an earlier survey of healthy participants in the region [10]. This pattern may reflect a peri-urban transmission cycle, as the majority of the OROV isolates were detected in both the rural sites and an urban site located towards the periphery of the city in 2005 and 2006 during a period of markedly increased transmission (Table 6). OROV isolates from previous Iquitos studies (prior to 1998) were found to belong to lineage II, similar to strains associated with Brazilian OROV outbreaks [56], [60]. Future sequence analysis of the more recent isolates described in this current study from Iquitos, Yurimaguas, and Puerto Maldonado, will provide a more complete description of OROV geographic and temporal genetic variability.
Considering the association of arboviral pathogens with human disease and the potential for wider-scale emergence, disease surveillance is an integral component of public health planning, disease control, and analysis of potential intervention strategies. Unfortunately, for the arboviruses described here syndromic surveillance is complicated by the lack of pathogen-specific signs and symptoms [5], particularly early in disease progression. As with other reports [7], [9], our study underscores the need for laboratory-based support of febrile surveillance studies. Even within our study other pathogens clearly need to be considered, as the majority of febrile episodes in this study were not associated with an arboviral etiology. In Iquitos past studies have linked both Leptospira spp. and Rickettsia spp. with a significant percentage of febrile illnesses [61]–[63]. To-date, solid data are lacking for the other study sites included in this study, although our preliminary results suggest that Leptospira spp. and Rickettsia spp. are common human pathogens in these locations as well (TJK, unpublished results). Admittedly our studies provide little in the way of guidance for patient care but do point toward the need for the development of pharmaceutical therapies for the treatment of a variety of viral infections. In addition the development of rapid and inexpensive diagnostic tools should be given high research priority, particularly to distinguish arbovirus infection from other pathogens where effective and inexpensive pharmaceutical treatment is already available, such as for Rickettsia spp. and Leptospira spp.
|
10.1371/journal.ppat.1000043 | Human-Like Receptor Specificity Does Not Affect the Neuraminidase-Inhibitor Susceptibility of H5N1 Influenza Viruses | If highly pathogenic H5N1 influenza viruses acquire affinity for human rather than avian respiratory epithelium, will their susceptibility to neuraminidase (NA) inhibitors (the likely first line of defense against an influenza pandemic) change as well? Adequate pandemic preparedness requires that this question be answered. We generated and tested 31 recombinants of A/Vietnam/1203/04 (H5N1) influenza virus carrying single, double, or triple mutations located within or near the receptor binding site in the hemagglutinin (HA) glycoprotein that alter H5 HA binding affinity or specificity. To gain insight into how combinations of HA and NA mutations can affect the sensitivity of H5N1 virus to NA inhibitors, we also rescued viruses carrying the HA changes together with the H274Y NA substitution, which was reported to confer resistance to the NA inhibitor oseltamivir. Twenty viruses were genetically stable. The triple N158S/Q226L/N248D HA mutation (which eliminates a glycosylation site at position 158) caused a switch from avian to human receptor specificity. In cultures of differentiated human airway epithelial (NHBE) cells, which provide an ex vivo model that recapitulates the receptors in the human respiratory tract, none of the HA-mutant recombinants showed reduced susceptibility to antiviral drugs (oseltamivir or zanamivir). This finding was consistent with the results of NA enzyme inhibition assay, which appears to predict influenza virus susceptibility in vivo. Therefore, acquisition of human-like receptor specificity does not affect susceptibility to NA inhibitors. Sequence analysis of the NA gene alone, rather than analysis of both the NA and HA genes, and phenotypic assays in NHBE cells are likely to adequately identify drug-resistant H5N1 variants isolated from humans during an outbreak.
| If the avian influenza H5N1 viruses adapt to human hosts, the first step is likely to be a switch in the preference of their viral hemagglutinin (HA) glycoprotein to bind to human rather than avian cell receptors. Such a switch may also alter virus susceptibility to neuraminidase (NA) inhibitors, which are anti-influenza drugs that are likely to be the first line of defense against a pandemic. We generated recombinant A/Vietnam/1203/04-like (H5N1) viruses carrying HA mutations previously shown to alter receptor specificity or affinity. We also discovered a previously unknown route (three simultaneous HA amino acid substitutions) by which highly pathogenic H5N1 viruses can adapt to human receptors. We then used a novel cell-culture–based system (differentiated human airway epithelial NHBE cells) to evaluate the recombinant viruses' resistance to NA inhibitors. None of the HA-mutant recombinants showed reduced drug susceptibility. Our results indicate that the tested HA mutations are unlikely to cause resistance to NA inhibitors in vivo. The NHBE system meets the need for an appropriate cell-culture–based system for phenotypic characterization of drug resistance.
| The spread of highly pathogenic avian influenza A (H5N1) viruses from Asia to the Middle East, Europe, and Africa raises serious concern about a potential human pandemic [1],[2]. H5N1 avian influenza virus has been reported in poultry in 63 countries; 359 human cases have been confirmed in 14 countries, with a mortality rate >60% [3]. A poor fit between avian viruses and human cellular receptors is thought to be one of the main barriers to efficient transmission of H5N1 influenza viruses between humans [2], [4]–[6]. The hemagglutinin (HA) glycoproteins of avian influenza viruses bind to avian cell-surface receptors whose saccharides terminate in sialic acid (SA)-α2,3-galactose (SAα2,3Gal), whereas those of human influenza viruses bind to human receptors whose saccharides end in SAα2,6Gal. A change in receptor specificity from SAα2,3Gal to SAα2,6Gal is thought to be necessary before avian influenza viruses can cause a pandemic [4]–[6].
Neuraminidase (NA) inhibitors (oseltamivir and zanamivir) are anti-influenza drugs that are likely to be the first line of defense in the event of an influenza pandemic, before antigenically matched influenza vaccine is available [1], [7]–[10]. Although HA mutations that alter viral receptor affinity/specificity can contribute to NA inhibitor resistance in vitro by allowing efficient virus release from infected cells without the need for significant NA activity [9], [11]–[18], the importance of HA mutations in the clinical management of influenza in humans remains uncertain [11], [19]–[23]. One important problem is the lack of a reliable experimental approach (i.e., an appropriate cell-culture–based system) for screening viral isolates for drug sensitivity [9],[11],[19],[20]. HA mutations can either increase or mask NA inhibitor resistance in the available assay systems, which are therefore susceptible to false-positive [24],[25] and false-negative [21],[22] results. This problem is likely to reflect a mismatch between human virus receptors and those in available cell-culture systems. The human airway epithelial cells targeted by influenza virus express high concentrations of SAα2,6Gal-containing receptors, which are present at low concentrations in the continuous cell lines used to propagate influenza viruses [9],[11],[19],[20],[26].
To test whether altered receptor-binding properties of the viral HA glycoprotein of highly pathogenic A/Vietnam/1203/04 (H5N1) influenza virus can reduce susceptibility to NA inhibitors in vivo, we generated 31 recombinant viruses carrying amino acid changes within or near the receptor binding site that alter binding affinity or specificity [27]. To evaluate the recombinant viruses' resistance to NA inhibitors, we used, for the first time, a cell-culture–based system that morphologically and functionally recapitulates differentiated human airway epithelial cells ex vivo [28],[29]. Based on our analysis, we propose that the HA mutations would not be expected to mediate resistance of H5N1 viruses to antiviral drugs, oseltamivir or zanamivir.
To test the hypothesis that substitutions in the viral HA gene can contribute to NA inhibitor resistance, we generated recombinant H5N1 viruses harboring HA point mutations that alter viral receptor specificity or affinity to SA receptors, using two approaches. Our group and others [11]–[18],[30],[31] had previously identified a number of HA mutations within and near the receptor binding site that could alter receptor specificity or affinity. However, as a first step in this study, we wished to identify additional HA point mutations that could convert the avian H5 HA to human-type receptor specificity. Previous studies had shown that two HA substitutions (Q226L and G228S) are likely to modulate receptor specificity in the H5 serotype [5]. We therefore passaged the wild-type virus (rgVN1203) and two HA mutants (Q226L and G228S) in MDCK-SIAT1 cells (Madin Darby canine kidney cells altered to express predominantly human-type SAα2,6 receptors). Because of the ability of NA inhibitors to select mutants with altered receptor affinity/specificity during in vitro passage, we also cultured these three H5N1 viruses in MDCK-SIAT1 cells in the presence of 1 µM oseltamivir [12]–[18]. Interestingly, infection with the wild-type virus was undetectable by PCR analysis after two passages with 1 µM of the NA inhibitor in two independent experiments (data not shown). Sequence analysis of the entire HA and NA genes revealed no additional mutations in virus with the G228S substitution after five sequential passages in the presence or absence of the drug. However, virus with the Q226L substitution had acquired two additional HA mutations, N158S (which eliminates a glycosylation site at position 158 [32]) and N248D, after five passages with or without compound. The receptor specificity of this triple-mutant (N158S/Q226L/N248D) virus was determined by measuring its binding affinity to sialoglycopolymers possessing either SAα2,3Gal (p3′SL) or SAα2,6Gal (p6′SL) (Table S1). This H5N1 variant exhibited enhanced affinity for human-like SAα2,6-linked receptor and was unable to bind the avian-like SAα2,3-linked receptor (Figure S1); therefore, the N158S/Q226L/N248D triple mutation is sufficient to completely switch the host receptor specificity of A/Vietnam/1203/04 (H5N1) virus from avian to human.
Our second approach was to use reverse genetics [33] to generate recombinant A/Vietnam/1203/04-like (H5N1) viruses carrying HA mutations previously shown to alter receptor specificity or affinity [11]–[18],[30],[31]. This study characterized a total of 15 HA mutants (Table 1) carrying substitutions at a total of 11 positions (Figure 1A). In addition, to gain insight into how combinations of HA and NA mutations can affect the sensitivity of H5N1 virus to NA inhibitors, we rescued viruses carrying the 15 HA changes together with the H274Y NA substitution. This mutation is most frequently associated with the resistance to the NA inhibitor oseltamivir in the N1 NA subtype [11] and was extensively characterized in A/Vietnam/1203/04 (H5N1)-virus background both in vitro and in vivo [34] (Table S2).
The use of the eight-plasmid reverse genetics system allowed us to predict the viability of all 31 recombinant H5N1 viruses in nature. We were able to rescue all of the recombinant viruses from transfected 293T cells as described previously [33]. However, the introduced N158S, T160A, R229S, N158S/N248D, and Q226L/N248D HA mutations could not be stably maintained in A/Vietnam/1203/04 virus after one passage in MDCK cells in two independent experiments because additional HA mutations were observed (Table S2). Interestingly, A/Vietnam/1203/04 virus simultaneously carrying the Q226L HA mutation and the H274Y NA mutation was genetically unstable, since the stock virus contained a mixture of viruses with Q or L at HA residue 226 as well as a K222I HA substitution. Sequence analysis revealed that the remaining 20 recombinant H5N1 viruses were stably maintained in stock cultures; these viruses grew to comparable titers in the different cell lines used (Table S2).
We measured the affinity of the 21 genetically stable recombinant H5N1 variants, including wild-type virus, for a wide range of high-molecular-weight sialic acid substrates, both natural (fetuin) and synthetic (Table S1). Like most avian influenza strains, wild-type rgVN1203 virus showed a binding preference for avian SAα2,3Gal-receptors (Figure 1B). The introduced HA substitutions had various effects on the receptor binding affinity of the H5 hemagglutinin to one or several SAα2,3Gal-substrates (Figure 1B). Surprisingly, H5N1 mutants carrying the triple mutation N158S/Q226L/N248D exhibited very weak SAα2,6Gal binding, whereas virus with the double mutation N158S/Q226L did not bind to any SAα2,3Gal sialosides but showed enhanced binding affinity to SAα2,6Gal-substrate (Figure 1B).
After observing a discrepancy in the receptor affinity of two N158S/Q226L/N248D HA triple mutants that were independently obtained by passaging in MDCK-SIAT1 cells (Figure S1) or by transfection of 293T-MDCK cells (Figure 1B), we investigated whether the host cell type can determine viral binding properties. We prepared virus stocks by transfecting MDCK-SIAT1 cells, rather than MDCK cells, with the two H5N1 viruses carrying the double N158S/Q226L and triple N158S/Q226L/N248D HA substitutions. Direct sequencing of the double-mutant virus revealed the presence of additional mutations; therefore, we did not assay its receptor specificity. A/Vietnam/1203/04 (H5N1) virus carrying the triple N158S/Q226L/N248D HA mutation and grown in MDCK-SIAT1 cells was genetically stable and demonstrated the switch from avian to human receptor specificity (Figure 1B).
We used the fluorometric NA enzyme inhibition assay [35] to test the susceptibility to oseltamivir and zanamivir of the 20 recombinant H5N1 viruses carrying either HA mutations or both HA and NA (H274Y) changes. None of the recombinants carrying only HA mutations differed from the wild-type virus in their sensitivity to either NA inhibitor (mean IC50±SD, 0.2±0.1 and 1.0±0.1 nM, respectively, for oseltamivir and zanamivir). All double-gene–mutant recombinant viruses were resistant to oseltamivir (the mean IC50 was ∼2060 times that of wild-type virus) but remained highly susceptible to zanamivir (IC50, 1.0±0.2 nM) (data not shown).
We next compared the activity of NA inhibitors in four cell-culture systems that differ in the surface distribution of SAα2,3- and SAα2,6-receptors. We performed plaque reduction assays in MDCK and MDCK-SIAT1 cells and virus reduction assays in human alveolar basal epithelial (A549) and normal human bronchial epithelial (NHBE) cells (Figures 2 and 3).
In MDCK cells, which express predominantly SAα2,3 (avian-type) receptors [26], all HA mutants except those with the S159N and N158S/Q226L substitutions were significantly more resistant to oseltamivir and to zanamivir than the wild-type strain (P<0.01) (Figure 2A). Introduction of the NA H274Y mutation markedly reduced the sensitivity of the recombinant H5N1 viruses to oseltamivir (mean EC50, 270 times that of the wild-type strain) but did not alter their susceptibility to zanamivir. The combined HA and NA (H274Y) substitutions therefore reduced oseltamivir sensitivity synergistically in MDCK cells.
In MDCK-SIAT1 cells, which have predominantly SAα2,6 (human-type) receptors [26], most of the mutants showed resistance to both NA inhibitors (Figure 2B), but fewer viruses were resistant in MDCK-SIAT1 than in MDCK cells. Viruses with Q226L, G228S and Q226L/G228S substitutions, which enhanced binding affinity for SAα2,6Gal receptors (Figure 1B), were as sensitive to both NA inhibitors as the wild-type virus in MDCK-SIAT1 cells but not in MDCK cells (Figure 2). Taken together, our results indicated that the drug sensitivity of the recombinant H5N1 viruses detected in MDCK-SIAT1 cells reflected their affinity for SAα2,6Gal rather than for SAα2,3Gal receptors.
In human lung A549 cells, which have predominantly SAα2,6-receptors [36], the overall sensitivity pattern was similar to that observed in MDCK-SIAT1 cells (Figure 3A). However, we were unable to assay the drug susceptibility of the mutants carrying the double N158S/Q226L and triple N158S/Q226L/N248D HA mutations, because their replication was undetectable. In A549 cells, but not in MDCK and MDCK-SIAT1 cells, only double mutants with the Y161H or K222I HA substitution plus the H274Y NA mutation showed resistance significantly greater (EC50 increased by a factor of 10-106) than that of the H274Y virus to both NA inhibitors (Figure 3A).
In differentiated cultures of NHBE cells (primarily human-type SAα2,6-receptors [28]), none of the HA mutations resulted in increased resistance to oseltamivir or zanamivir, and there was no difference in susceptibility between viruses carrying only the H274Y NA mutation and those carrying HA mutations as well (Figure 3B). Therefore, the combined HA–NA mutations had a negligible effect on the NA inhibitor sensitivity of H5N1 viruses in NHBE cultures. All recombinant viruses carrying a single HA substitution were slightly less susceptible to zanamivir than to oseltamivir (the EC50 differed by a factor of ∼5) in NHBE cells. This finding was consistent with the data obtained by NA enzyme inhibition assay (see above).
Our results answer several fundamental questions about the effect of HA mutations on the host receptor affinity and NA inhibitor susceptibility of highly pathogenic influenza H5N1 viruses. Importantly, we found that alteration of receptor specificity or affinity does not alter sensitivity to NA inhibitors. In light of global concern about pandemic preparedness, it is crucial not only to understand what mutations might endow H5N1 viruses with human receptor specificity [4]–[6] but also to anticipate the clinical consequences of such adaptation. With increasing clinical use and stockpiling of NA inhibitors for pandemic preparedness, it is also crucial to elucidate molecular mechanisms that contribute to drug resistance.
Mutations at specific positions (129/134, 182, 192, 227, and 226/228) in the HA gene of H5N1 influenza virus were recently shown to reduce or eliminate binding affinity to avian-type receptors and enhance affinity to human-type receptors [5],[6],[37],[38]. Here, we have identified another possible route of adaptation to human receptors: simultaneous amino acid substitutions at HA positions 158 (that results in the loss of a glycosylation site), 226 and 248. Our study demonstrated the importance not only of residues near or within the receptor binding site, but also those structural elements that are located in its vicinity that can affect host receptor specificity. H5N1 influenza viruses in H5 clade 1, such as the virus used here, would be extremely unlikely to acquire all three mutations. However, most members of the clade 2.2 family now circulating in Europe, the Middle East and Africa already lack a glycosylation site at HA position 158 [1]. This natural feature reduces the required mutations to only two, thus enhancing the probability of such an occurrence. Taken together, we can conclude that different amino acid substitutions in the H5 HA enable to cause a shift from the avian- to human-type specificity and along with the H5N1 evolution other strain-specific mutations cannot be excluded.
HA glycosylation has been reported to affect the specificity or affinity of influenza viruses for cellular receptors [39]–[41]. Our results directly demonstrate that HA glycosylation plays a role in the viability of avian H5N1 influenza viruses in MDCK-SIAT1 cells, which express predominantly human-like SAα2,6Gal receptors. We demonstrated that the loss of an MDCK-SIAT1 cell–synthesized oligosaccharide at position 158 of HA, adjacent to the receptor binding site, increased the capacity of the virus to bind to these cells. One possible explanation is that the removal of an oligosaccharide attachment site from the tip of the HA1 subunit eliminates steric hindrance that limits the accessibility of the receptor pocket by this SA-containing oligosaccharide [41]. Further, our results confirm previous reports [39]–[41] that the effect of oligosaccharide removal can differ with the host cell type (MDCK vs. MDCK-SIAT1) in which the H5N1 virus is grown. Therefore, the receptor affinity of the avian H5N1 influenza virus may also be affected by the host cell–determined composition of other oligosaccharides on the two H5 HA subunits.
Our results are consistent with what is known about the role of the functional balance between the receptor-binding (HA) and receptor-destroying (NA) activities of the surface glycoproteins in efficient influenza virus infection [42],[43]. The genetic and phenotypic instability of 11 of our recombinant H5N1 viruses in vitro may reflect the functional mismatch of their HA and NA glycoproteins. The balance between HA and NA functions could also explain the diverse pattern of influenza virus susceptibility to NA inhibitors observed in different cell-culture systems [11],[19],[20],[42],[43]. The disparate HA–NA balance required to infect MDCK, MDCK-SIAT1, and A549 cells, together with the differences in SA receptors between these cell lines and human respiratory epithelial cells, significantly limit the suitability of these commonly used cell lines for phenotypic characterization of NA inhibitor resistance.
NHBE cells cultured ex vivo [28],[29],[44] offer a new cell-culture–based system that functionally and morphologically recapitulates normal differentiated human airway epithelium; this system allows improved evaluation of the NA inhibitor sensitivity of avian influenza viruses that are potential human pathogens. Taken together, our data demonstrate a parallel between virus susceptibility determined by NA enzyme inhibition assay (which appears to predict in vivo results [11],[19],[20]) and virus susceptibility in NHBE cells (an ex vivo model). NHBE cells [28],[44], which express the sialic acid receptors present in humans, may offer an optimal system for maintaining viral fitness and, as a consequence, for prediction of influenza virus resistance to NA inhibitors in vivo. Our results suggest that the HA mutations that alter the receptor specificity or affinity of highly pathogenic H5N1 viruses are unlikely to mediate concomitant resistance to NA inhibitors in vivo.
However, we cannot exclude the possibility that the HA mutations might contribute to the selection of certain NA mutations that lead to drug resistance simply by altering HA–NA balance. Indeed, recent observation that H5N1 viruses from clade 2 isolated in 2005 demonstrated a 25- to 30-fold decrease in sensitivity to oseltamivir carboxylate compared with clade 1 viruses and none of the mutations known to confer NA inhibitor resistance was observed [45], suggests that the decrease in sensitivities may be due to drift mutations in the NA and HA proteins. Additionally, our finding that A/Vietnam/1203/04 virus carrying the Q226L HA mutation, which is known to switch the receptor specificity in the H3 HA subtype [5], and the H274Y NA mutation was not genetically stable, could provide evidence that some HA mutations have the potential impact on the acquisition of mutations in NA, including those that can lead to decreased drug susceptibility. One, therefore, can speculate that the identification of oseltamivir-resistant viruses as a significant proportion of influenza H1N1 viruses circulating in Europe [46] could be determined by preceding NA and/or HA mutations.
In conclusion, our findings can improve the monitoring of NA inhibitor resistance among viruses with pandemic potential. Further, sequence analysis of the NA gene alone, rather than analysis of both the NA and HA genes, may adequately identify all drug-resistant H5N1 variants. The human airway epithelial cell cultures used in this study could also advance the study of drug resistance mechanisms by serving as a suitable model of the human respiratory cell system for phenotypic characterization of NA inhibitor resistance in clinical testing.
Madin-Darby canine kidney (MDCK), human embryonic kidney (293T) and human alveolar basal epithelial (A549) cells were obtained from the American Type Culture Collection. MDCK cells transfected with cDNA encoding human 2,6-sialyltransferase (MDCK-SIAT1 cells) were kindly provided by Dr. Mikhail N. Matrosovich. Primary normal human bronchial epithelial (NHBE) cells were obtained from Cambrex Bio Science. All cell cultures were maintained as previously described [26],[28],[31],[36],[44].
Eight plasmids were constructed from the DNA sequences of the 8 gene segments of wild-type A/Vietnam/1203/04 (H5N1) virus for the reverse-genetics generation of recombinant wild-type virus (rgVN1203). Recombinant virus was generated by DNA transfection of 293T cells [33], the HA cleavage site was removed, and the point mutations (Tables 1 and S2, Figure 1A) were inserted into the HA and NA genes of rgVN1203 virus by using a Quickchange site-directed mutagenesis kit (Stratagene) [32]. Stock viruses were prepared in MDCK cells at 37°C for 72 h and their entire HA and NA genes were sequenced to verify the presence of the mutations. The recombinant viruses were designated according to their HA and NA mutations (Tables 1 and S2). All experimental work with the H5N1 recombinant viruses was performed in a biosafety level 3+ laboratory approved for use by the U.S. Department of Agriculture and the U.S. Centers for Disease Control and Prevention.
The NA inhibitors oseltamivir carboxylate (oseltamivir) ([3R,4R,5S]-4-acetamido-5-amino-3-[1-ethylpropoxy]-1-cyclohexene-1-carboxylic acid) and zanamivir (4-guanidino-Neu5Ac2en) were provided by Hoffmann-La Roche, Ltd.
For the first passage, MDCK-SIAT1 cells were infected with influenza H5N1 viruses at a multiplicity of infection (MOI) of 0.001 PFU/cell and cultivated for 72 h in infection medium [containing 4% bovine serum albumin, sodium bicarbonate, 100 U/ml of penicillin, 100 µg/ml of streptomycin sulfate, 100 µg/ml of kanamycin sulfate, 1 µg/ml of L-1-(tosyl-amido-2-phenyl)ethyl chloromethyl ketone (TPCK)–treated trypsin (trypsin) (Worthington Diagnostics)] with or without 1 μM oseltamivir [31]. Four additional passages identical to the first one were then performed sequentially.
The genetic stability of recombinant H5N1 viruses was monitored by plaque assay and by sequencing of the HA and NA genes after transfection of 293T cells and after one passage in MDCK/MDCK-SIAT1 cells. Influenza virus was defined as genetically stable if it was able to replicate efficiently in the cell lines used, maintain a homogeneous plaque phenotype, and did not contain additional subpopulations based on the sequence analysis of the HA and NA genes after one passage in MDCK/MDCK-SIAT1 cells. If different subpopulations were identified, those viruses were designated as unstable (Tables 1 and S2).
The yield of H5N1 viruses in MDCK, MDCK-SIAT1 and A549 cells was defined as log10 of the 50% tissue culture infectious dose (TCID50) as described previously [43]. Briefly, confluent monolayers of cell cultures growing in 96-well microplates were inoculated with serial virus dilutions (each dilution was added to five wells) in the presence of trypsin. After 3 days, virus was titrated by HA assay, and virus titers were expressed as log10TCID50/ml by the end-point method of Reed and Muench [47]. NHBE cells were inoculated by exposure of the apical side to recombinant H5N1 viruses at a MOI of 0.1, as determined by TCID50 assay in MDCK cells. After 1 h incubation, the inoculum was removed and the cells were incubated for 24 h. No trypsin was added to the cultures because previous studies in similar cultures demonstrated efficient proteolytic activation of influenza viruses by endogenous proteases [28]. Viruses released into the apical compartment of NHBE were harvested by adding 300 µl of medium to the apical compartment, allowing it to equilibrate for 30 min, and collecting it. Virus titer was determined as log10TCID50/ml in MDCK cells.
The binding of human influenza viruses to fetuin was measured in a direct solid-phase assay using the immobilized virus and horseradish peroxidase-conjugated fetuin, as described previously [48]. The affinity of viruses for synthetic 3′- and 6′-substrates (Table S1) was measured in a competitive assay based on the inhibition of binding of the labeled fetuin [49]. The association constants (Kass) were determined as sialic acid (Neu5Ac) concentration at the point Amax/2 on Scatchard plots.
NA activity was determined as described by Potier et al. [35]. Briefly, H5N1 viruses and various concentrations of oseltamivir or zanamivir were preincubated for 30 min at 37°C before addition of the substrate 2′-(4-methylumbelliferyl)-α-D-N-acetylneuraminic acid (Sigma). After 1 h, the reaction was terminated by adding 14 mM NaOH and fluorescence was quantified in a Perkin-Elmer fluorometer. The IC50 was defined as the concentration of NA inhibitor necessary to reduce NA activity by 50% relative to that in a reaction mixture containing virus but no inhibitor.
The drug susceptibility of recombinant H5N1 viruses was determined by plaque reduction assay in MDCK and MDCK-SIAT1 cells [50] and by virus reduction assay in A549 and NHBE cells [31],[51]. Briefly, MDCK or MDCK-SIAT1 cells were inoculated with virus diluted to yield ∼50 plaques per well and were then overlaid with infection medium containing oseltamivir (0.0001 to 100 µM) or zanamivir (0.0001 to 100 µM) in the presence of trypsin. The results were recorded after 3 days of incubation at 37°C. At least three independent experiments were performed to determine the concentration of compound required to reduce plaque size by 50%, relative to that in untreated wells (EC50).
A549 cells were inoculated with H5N1 viruses at an MOI of 0.001 PFU/cell and after 1 h of adsorption were overlaid with infection medium containing oseltamivir (0.001 to 100 µM) or zanamivir (0.001 to 100 µM) in the presence of trypsin. Virus yield was determined by TCID50 assay of culture supernatants 72 h after inoculation. The drug concentration that caused a 50% decrease in the TCID50 titer in comparison to control wells without drug was defined as the 50% inhibitory concentration (EC50). The results of three independent experiments were averaged.
NHBE cells were inoculated by exposure of the apical side to recombinant H5N1 viruses at an MOI of 0.1 in the presence of oseltamivir (0, 0.1, 1, 10 µM) or zanamivir (0, 0.1, 1, or 10 µM). These concentrations represent typical plasma minimum and maximum concentrations measured in humans after administration of 75 mg of oseltamivir phosphate or 10 mg of zanamivir, the doses recommended for prophylaxis [10],[52]. After 1 h incubation, the inoculum was removed and the cells were incubated for another 24 h. Viruses released into the apical compartment of NHBE cells were harvested by the apical addition and collection of 300 µl of medium allowed to equilibrate for 30 min. The virus titer was determined as log10TCID50/ml in MDCK cells. |
10.1371/journal.ppat.1003189 | Functional Plasticity in the Type IV Secretion System of Helicobacter pylori | Helicobacter pylori causes clinical disease primarily in those individuals infected with a strain that carries the cytotoxin associated gene pathogenicity island (cagPAI). The cagPAI encodes a type IV secretion system (T4SS) that injects the CagA oncoprotein into epithelial cells and is required for induction of the pro-inflammatory cytokine, interleukin-8 (IL-8). CagY is an essential component of the H. pylori T4SS that has an unusual sequence structure, in which an extraordinary number of direct DNA repeats is predicted to cause rearrangements that invariably yield in-frame insertions or deletions. Here we demonstrate in murine and non-human primate models that immune-driven host selection of rearrangements in CagY is sufficient to cause gain or loss of function in the H. pylori T4SS. We propose that CagY functions as a sort of molecular switch or perhaps a rheostat that alters the function of the T4SS and “tunes” the host inflammatory response so as to maximize persistent infection.
| Helicobacter pylori is a bacterium that colonizes the stomach of about half the world's population, most of whom are asymptomatic. However, some strains of H. pylori express a bacterial secretion system, a sort of molecular syringe that injects a bacterial protein inside the gastric cells and causes inflammation that can lead to peptic ulcer disease or gastric cancer. One of the essential components of the H. pylori secretion system is CagY, which is unusual because it contains a series of repetitive amino acid motifs that are encoded by a very large number of direct DNA repeats. Here we have shown that DNA recombination in cagY changes the protein motif structure and alters the function of the secretion system—turning it on or off. Using mouse and non-human primate models, we have demonstrated that CagY is a molecular switch that “tunes” the host inflammatory response, and likely contributes to persistent infection. Determining the mechanism by which CagY functions will enhance our understanding of the effects of H. pylori on human health, and could lead to novel applications for the modulation of host cell function.
| Helicobacter pylori commonly infects the human gastric epithelium and sometimes causes peptic ulcer disease or gastric cancer, which is the second most common cause of cancer death worldwide. The H. pylori virulence locus most strongly associated with clinical disease, rather than asymptomatic infection, is the cag pathogenicity island (cagPAI). The 40-kb cagPAI consists of approximately 27 genes, several of which encode a type IV secretion system (T4SS) that binds β1 integrins [1], [2] and translocates the CagA oncoprotein into gastric epithelial cells [3]. Phosphorylated and nonphosphorylated forms of intracellular CagA cause complex changes in host-cell signaling that lead to epithelial cell elongation [4], disruption of tight junctions [5], and alteration of cell polarity [6], [7]. The T4SS is also required for induction of interleukin-8 (IL-8), a member of the CXC cytokine family, which has long been used as a convenient assay to characterize the inflammatory potential of H. pylori strains [8], [9]. It has been proposed that IL-8 induction is mediated by cagPAI-dependent translocation of peptidoglycan, activation of nucleotide-binding oligomerization domain 1 (NOD1), and stimulation of NF-κB [10]. However, this remains controversial, as some have suggested that IL-8 and other NF-κB-dependent proinflammatory responses are mediated primarily by toll like receptors and MyD88, rather than NOD1 [11]. Very recently, a NOD1- and CagA-independent pathway of IL-8 induction has also been described [12].
The prototypical T4SS is the VirB secretion apparatus of Agrobacterium tumefaciens, which consists of 11 VirB proteins (encoded by virB1-11) and the coupling protein, VirD4 [13]. Although the function of the H. pylori T4SS proteins cannot be easily assigned based on the distantly related A. tumefaciens, functional and structural studies are beginning to emerge. Mutagenesis studies have demonstrated that 15 genes on the cagPAI are required for H. pylori induction of IL-8 [14], [15]. One such gene is cagY, which encodes the H. pylori VirB10 orthologue. CagY is a large protein of approximately 220 kDa that is thought to mediate contact between the inner and outer bacterial membrane [16], similar to what has been described in A. tumefaciens and other Gram-negative bacteria [17]. However, cagY is much larger than virB10 from A. tumefaciens, and it has an unusual sequence structure in which an extraordinary number of direct DNA repeats are found in a small 5′ repeat region (FRR) and a large middle repeat region (MRR) of the gene [18]. Potential DNA rearrangements predicted by these repeats invariably yield in-frame insertions or deletions that result in variant proteins. The observation that variant CagY proteins are found in different H. pylori strains or after passage in mouse models, led to the suggestion that CagY undergoes antigenic variation to evade the host immune response [18] while maintaining T4SS function [19].
Here we demonstrate that experimental infection with H. pylori leads to host immunity-dependent recombination in cagY that is sufficient to eliminate the functionality of the T4SS. Moreover, changes in cagY during experimental infection could also turn on the capacity to induce IL-8 and phosphorylate CagA, suggesting that the function of CagY diversity is not to evade the host immune response but rather to modulate it. We propose that CagY functions as a molecular switch or perhaps a rheostat that “tunes” the host inflammatory response by altering the function of the T4SS so as to maximize persistent infection.
H. pylori strains adapted to colonization of mice frequently lose the capacity to induce IL-8 and translocate CagA into gastric epithelial cells [20], [21], which are measures of a functional T4SS. The cagPAI is retained and the mechanism is unknown [21]. Since mice are not a natural host for H. pylori, we asked whether similar changes occur during infection of rhesus macaques, which are commonly infected with H. pylori that is indistinguishable by comparative genomic hybridization from that which infects humans [22]. Five rhesus monkeys were previously challenged with a single colony of wild type (WT) H. pylori J166 that has a functional cagPAI [23]. Multiple output colonies recovered from each monkey up to 14 months post inoculation (PI) were co-cultured with AGS gastric cells to determine their capacity to induce IL-8, which was normalized to the WT control strain. IL-8 induction resembled WT in bacteria recovered early after challenge, but decreased over time in 4 of 5 monkeys (Figure 1A–D). In one monkey, all but one bacterial colony induced IL-8 at levels≥WT, even after 14 months of colonization (Figure 1E). Selected rhesus output colonies that induced low IL-8 (designated rOut1 and rOut2) or high IL-8 (rOut3) in AGS cells were also tested in KATO III gastric cells. Similar results were obtained (Figure S1A). These results demonstrate that H. pylori infection of rhesus monkeys results in a population of strains that have lost the capacity to induce the pro-inflammatory cytokine, IL-8, though there are individual differences among animals. Since loss of T4SS function occurs in macaques as well as mice, yet differs among individuals, it may represent a physiologic accommodation to the gastric environment.
Systematic mutagenesis experiments have demonstrated that 15 genes on the cagPAI (cagδ, cagγ, virB11, cagY, cagX, cagW, cagV, cagU, cagT, cagM, cagL, cagI, cagH, cagE, cagC) are essential for H. pylori to fully induce IL-8 [14], [15]. In some strains, cagA is required as well [24]. To determine if change in one or more of these genes was responsible for loss of IL-8 induction during colonization of rhesus monkeys, we amplified and sequenced each of these genes from WT J166 and from a rhesus output strain (rOut1) that had lost the capacity to induce IL-8. Each of the 16 genes was identical between WT J166 and rOut1 with the exception of cagY, in which a 321 bp fragment was deleted.
Dot-plot analysis (Figure S2) demonstrated that, like in strains J99 and 26695 [18], [25], cagY in H. pylori J166 has a large number of direct repeats that are organized into a 5′ repeat region (FRR) and a middle repeat region (MRR), in which the 321 bp deletion in rOut1 was located. The large number of direct repeats in cagY could permit deletion or duplication of the intervening region with preservation of an open reading frame, and might alter the functionality of the cagPAI. Since high throughput DNA sequencing of cagY is difficult owing to its large size and repeat structure, we used PCR-RFLP to determine if recombination in cagY occurred during infection of rhesus monkeys, and if it was associated with altered capacity to induce IL-8. Figure S3A shows representative cagY PCR-RFLP patterns from WT J166 and rOut1-3, each of which is unique. Each monkey was colonized by multiple unique cagY variants with the exception of one (31811), in which all but one colony induced IL-8≥WT and had a cagY that was indistinguishable from that in WT J166 (Figure 1F). We next compared the cagY PCR-RFLP from all 81 output colonies with their capacity to induce IL-8, and asked if cagY was the same (solid circles) or different (open circles) from that of WT J166 (Figure 1). Among all monkey output colonies that had normalized IL-8 induction ≥1.0, 96% (23 of 24) had the same cagY PCR-RFLP fingerprint as WT J166, while only 25% (14 of 57) of colonies with IL-8 induction <1.0 showed the same fingerprint (Fisher's exact test, two-tailed, P<0.0001). Output strains in which cagY differed from WT J166 typically showed IL-8 induction similar to the mean (±SEM) of a cagY deletion mutant (0.29±0.04). Loss of IL-8 induction without an apparent change in cagY may sometimes occur due to the inability of PCR-RFLP to detect frameshift mutations that lead to premature stop codons in cagY, or to a change in other cagPAI genes, including cagA, which is essential for full induction of IL-8 in H. pylori J166 (Figure S4). To determine if changes in cagY might be due simply to frequent recombination during in vitro culture, WT H. pylori J166 was passaged daily for 5 weeks, and each week 6 individual colonies were isolated and examined by PCR-RFLP. Of the 30 clones tested, 28 (93%) showed the same RFLP pattern and similar mean (±SEM) IL-8 induction (0.91±.01) as WT J166; the two clones with a different cagY RFLP showed reduced induction of IL-8 (0.32±.00). Since loss of IL-8 induction and change in cagY were common during experimental infection but not during in vitro passage, these results suggest that H. pylori infection of rhesus macaques selects for allelic diversity in cagY that is associated with decreased capacity to induce IL-8.
Recombination in cagY might be associated with changes in IL-8, but not mechanistically linked to the function of the cagPAI. Therefore, we used contraselection [26], [27] to replace the cagY in WT J166 with the cagY gene from rOut1 or rOut2, each of which induced low IL-8 and had a unique cagY RFLP pattern. The cagY gene from streptomycin resistant J166 was deleted by homologous recombination with the cat-rpsL cassette (chloramphenicol resistant, dominant streptomycin susceptible), and then transformed with chromosomal DNA from either WT J166 (restoring the WT cagY allele) or one of the two rhesus output strains. Transformants that were chloramphenicol susceptible and streptomycin resistant (due to loss of the cassette), and had the appropriate cagY gene by PCR-RFLP and confirmed by full-length DNA sequence analysis, were then tested for induction of IL-8 and translocation of CagA. As expected, deletion of cagY in J166 markedly reduced IL-8 induction, and replacement of the WT cagY allele restored expression of CagY and induction of IL-8 (Figure 2). In contrast, replacement with cagY from rOut1 and rOut2, which induced low IL-8, did not restore IL-8 induction, even though the CagY protein was expressed. Although it was uncommon, we also identified a rhesus output strain (rOut3) that induced IL-8 at a level similar to WT J166, but had a unique cagY allele. As expected, replacement of the WT cagY allele with cagY from rOut3 maintained the capacity to induce IL-8. Only those strains that induced IL-8 were also capable of inducing CagA translocation and phosphorylation. These results demonstrate that recombination in cagY is sufficient to alter the functionality of the T4SS encoded by the cagPAI.
Identification of the direct repeat structure of cagY suggested that frequent in-frame recombination events may be a mechanism of antigenic variation to avoid the host adaptive immune response [28]. To test this hypothesis, we inoculated WT H. pylori J166 into WT C57BL/6 and RAG2−/− mice, which do not have functional B or T cells and develop little or no gastric inflammation after H. pylori infection [29]. H. pylori colonization levels were approximately 10-fold higher in RAG2−/− mice compared to WT mice (Figure S5A). Similar to the results in rhesus monkeys, bacteria recovered from WT mice resembled input H. pylori early after challenge (Figure 3A). However, at 12 and 16 weeks PI, bacteria from WT mice showed a significant loss in IL-8 induction (P<0.01) and change in cagY (P<0.001) compared to colonies from RAG2−/− mice, which uniformly resembled WT J166 in IL-8 induction and showed no changes in cagY by RFLP analysis (Figure 3B). We next replaced the cagY allele in WT H. pylori J166 with that from mouse output strains that changed cagY and either lost (mOut1 and mOut2) or maintained (mOut3 and mOut4) IL-8 induction in AGS cells, which was confirmed in KATO III cells (Figure S1B). Similar to the results with rhesus output strains (Figure 2), induction of IL-8 and phosphorylation of CagA in mouse output strains were phenocopied when their cagY allele was used to replace that in WT J166 (Figure 4). Interestingly, the bacterial population within each individual mouse was relatively homogenous, showing either WT levels of IL-8 and cagY indistinguishable from input, or low IL-8 and one or at most two unique cagY variants (Figure 3C). These experiments demonstrate that CagY-mediated change in function of the H. pylori T4SS is dependent on an intact host immune system.
Although H. pylori-induced signaling cascades in host cells are complex and poorly understood, it is clear that T4SS-mediated pro-inflammatory responses are dependent upon activation of the transcription factor, NF-κB [30]. Therefore, we examined NF-κB activation using an AGS cell line stably transfected with a luciferase reporter construct. AGS cells were co-cultured with WT J166 or isogenic J166 strains encoding cagY from monkey or mouse output strains. Phorbol myristate acetate (PMA) and deletions in the entire cagPAI or in cagY were used as positive and negative controls, respectively. Similar to strains with a deletion in cagY or the entire cagPAI, cagY variants that failed to induce IL-8 and translocate CagA (rOut1,2; mOut1,2) also failed to activate NF-κB (Figure 5). In contrast, introduction of cagY alleles from strains that induced IL-8 and translocated CagA (rOut3; mOut3,4) showed significantly increased activation of NF-κB, though rOut3 did not achieve a level similar to WT J166. These results suggest that cagY mediated alterations in T4SS function is mediated largely by NF-κB.
Previous analysis of 14 full-length CagY sequences in the NCBI non-redundant protein data base suggested that the MRR is organized into two α-helical principal motifs, which occur in tandem arrays of one to six 38–39 residue A motifs flanked by single copies of a 31 residue B motif [19]. Both principal motifs are made up of three distinct submotifs, which remain invariant in their order. This annotation suggests that CagY variants that are selected in vivo are likely a result of duplication or deletion of principal motif segments, without compromising the underlying submotif composition. To examine this, we first identified the A and B principal amino acid motifs in the CagY MRR of WT H. pylori J166. Similar to other H. pylori strains previously described [19], the CagY MRR of H. pylori J166 is organized into six tandem arrays of one to four A motifs flanked by B motifs (Figure 6). We next examined the motif structure of CagY from rhesus (rOut1-3) and mouse (mOut1-4) output strains that were previously characterized (Figures 2 and 4). All output strains from monkeys and mice with variant cagY alleles had lost one or more A or B motifs, though there were multiple CagY motif structures associated with the same IL-8 phenotype. One output strain each from monkey (rOut1) and from mouse (mOut1), which had both lost the capacity to induce IL-8, had identical motif structures. Interestingly, loss of a single A motif was sufficient to markedly reduce IL-8 induction (rOut2), while loss of 14 A and B motifs (mOut3), representing a reduction in predicted size from 233 kDa to 175 kDa, was not. Although we were unable to identify a motif pattern associated with the IL-8 phenotype, these results suggest that CagY function is based on a higher order structure and not on any critical motif within the MRR.
Output colonies that have variant cagY alleles and no longer induce IL-8 still express CagY (Figures 2 and 4), but they might not make T4SS pili, or the pili might have altered structural features. To test this possibility, we used field emission scanning electron microscopy (FEG-SEM) to image H. pylori strains co-cultured with AGS cells. As expected, WT J166 but not a cagPAI deletion mutant produced pilus-like structures (Figure 7). This is consistent with previous studies demonstrating that the cagPAI is essential for the formation of a T4SS [2], [15], [31], [32]. Pili of similar dimensions were previously reported to be present in WT strain 26695, but absent in H. pylori 26695 with deletions of cagT, cagE, cagL, and cagI, all of which are required for a functional T4SS [15]. Using this imaging approach, we examined isogenic strains of H. pylori J166 in which the cagY gene had been replaced with alleles from strains that did (rOut3, mOut3) or did not (rOut2, mOut2) induce IL-8 and translocate CagA (Figures 2 and 4). Regardless of cagPAI functionality, all strains made pilus structures (Figure 7). Although the pili were less prominent on some strains that had defects in T4SS function, we were unable to identify a reproducible association between cagPAI function and quantitative measures of pilus number or morphology (Table S1).
Pilus structures were also seen in H. pylori J166 with a deletion of cagY (Figure 7); similar results were obtained with a cagY deletion in H. pylori strain 26695 (Figure 7). To investigate the cellular localization of CagY, we performed immunogold SEM using antibody to the CagY MRR to stain H. pylori co-cultured with AGS cells. Antibody to CagA was used as a positive control. Although CagY label was seen scattered over the bacterial cells in WT H. pylori, no staining was found on or near the pilus structure (Figure 8). In contrast, CagA was identified both on the cell surface and closely approximated to the tips of pili in WT H. pylori, which has been reported previously [2]. CagA was not detected in association with pili in a cagY deletion mutant, in which the T4SS is not functional, and there was markedly reduced CagA labeling on the surface of the cagY mutant bacteria compared to WT (Figure 8). The absence of detectable CagY in association with pili is consistent with the finding that a ΔcagY mutant produces pili that are indistinguishable from those in the WT strain. Together, the EM results suggest that the loss of function that occurs with changes in CagY results from a functional change in the T4SS without any detectable structural defect in the T4SS pilus.
Studies of H. pylori pathogenesis were long hampered by the inability of investigators to successfully colonize mice. Since the difficulty was attributed primarily to H. pylori strain differences, mouse-adapted strain SS1 was derived, which has become the standard for animal experimentation [33]. However, it was later realized that H. pylori SS1 did not induce IL-8 or translocate CagA [20], [21], despite having an intact cagPAI detected by microarray [34]. The reason for this is unknown. It was recently reported that the original human isolate, designated pre-mouse SS1 (PMSS1), does have a functional cagPAI [35]. We therefore hypothesized that SS1 had undergone recombination in cagY during mouse passage that eliminated its capacity to induce IL-8 and translocate CagA. To test this hypothesis, we first inoculated PMSS1 into WT C57BL/6 and RAG1−/− mice, and examined IL-8 induction and cagY RFLP in colonies recovered 8 weeks PI. Similar to the results with strain J166 (Figure 3), colonies from WT but not RAG1−/− mice showed loss of IL-8 induction that was associated with recombination in cagY (Figure 9A). These results are consistent with a previous report demonstrating loss of T4SS function after challenge with PMSS1 in adult but not neonatal mice, which control effector T cell responses by H. pylori-specific regulatory T cells [35].
The cagY allele in SS1 is much larger than that in PMSS1 and has a markedly different PCR-RFLP pattern (Figure 9B). To determine if the increase in size of cagY was responsible for loss of a functional T4SS in SS1, we used contraselection to exchange the cagY genes between PMSS1 and SS1, and tested the strains for induction of IL-8 and translocation of CagA. As expected, H. pylori PMSS1 induced IL-8 and translocated CagA (Figure 9C), while SS1 did not (Figure 9D), although both expressed CagA and CagY. However, when cagY from SS1 was introduced into PMSS1, it could no longer translocate CagA or induce IL-8 (Figure 9C), indicating that the SS1 CagY was not functional. Interestingly, when cagY from PMSS1 was introduced into SS1, CagA translocation and IL-8 induction increased, but not to the level of PMSS1 (Figure 9D), suggesting that alteration in cagY is not the only defect in the T4SS of SS1. Together, these results suggest that H. pylori SS1 underwent recombination in cagY during mouse passage, which eliminated the functionality of the T4SS, reduced its inflammatory capacity, and enhanced its colonization of mice.
Recombination in cagY could be a mechanism by which H. pylori modulates rather than evades the host inflammatory response. If so, in vivo cagY recombination might sometimes confer an increase in the function of the T4SS, and enhance rather than reduce H. pylori inflammatory potential. To address this hypothesis, we undertook experiments to investigate possible alterations in cagY that might occur if animals were challenged with H. pylori mOut2, which had undergone cagY recombination that eliminated function of the T4SS (Figure 4). As a first step, to exclude the possibility that additional mutations could have occurred in mOut2 that conferred loss of T4SS function, we used contraselection to replace the cagY in mOut2 with that from WT J166. The results demonstrated that replacement of cagY in this strain with cagY from WT J166 was sufficient to restore induction of IL-8 in mOut2 (Figure S6). In three of four monkeys infected with mOut2 (36001, 35951, 35930), most colonies recovered two weeks after challenge resembled the input, with low IL-8 induction and the same cagY PCR-RFLP (Figure 10A). However, by eight weeks there was a significant increase in the capacity to induce IL-8 that was accompanied by changes in the cagY RFLP. One of these three monkeys (36001) was sampled repeatedly up to 24 weeks post inoculation; all output colonies recovered 8 weeks or more PI induced IL-8 and expressed a cagY that differed from that in mOut2 (Figure S7). A fourth monkey (36018) was colonized with a mixed population of cagY variants, but nearly all induced low IL-8 similar to that of the challenge strain. We next infected C57BL/6 WT and RAG2−/− mice with mOut2, and analyzed IL-8 induction and cagY RFLP up to 16 weeks PI. Similar to infection with WT J166, colonization density of mOut2 was greater in RAG2−/− mice than in C57BL/6 mice (Figure S5B). In general, strains recovered from both WT and RAG2−/− mice induced low IL-8 similar to the input mOut2, with no change in cagY (Figure 10B). A few colonies from both WT and RAG2−/− mice showed increased IL-8, which was accompanied by a change in cagY. Strains from mice and monkeys that recovered IL-8 induction showed novel cagY RFLP fingerprints that did not revert to WT J166. These results demonstrate that in vivo recombination in cagY can either eliminate or restore the function of the T4SS encoded on the H. pylori cagPAI. Since CagY that confers a non-functional T4SS appears stable in mice, modulation may be driven more by inflammation rather than adaptive immune responses.
The capacity to evade or circumvent host defense is considered a signature of pathogenic bacteria that distinguishes them from closely related commensals [36]. The mechanisms by which this occurs are varied, and they include elaboration of toxins that inhibit the function of immune cells, iron sequestration, antigenic variation of surface structures, intracellular invasion, and inducing host expression of immunosuppressive cytokines, to name just a few. But bacterial pathogens not only avoid host immune responses, they also sometimes exploit them. This is perhaps best understood for infection with Salmonella enterica serotype Typhimurium, where the T3SS-dependent host inflammatory response is required for colonization of mice [37]. Inflammation generates tetrathionate, an electron acceptor that can be used by S. Typhimurium but not by competing microbiota [38]. Inflammation also induces epithelial cells to express lipocalin-2 and calprotectin, which sequester iron and zinc from the gut microbiota but not from S. Typhimurium because it expresses specialized high affinity metal transporters [39], [40]. Thus, from the bacterial point of view, the host inflammatory response can be both deleterious and advantageous.
The hallmark of infection with H. pylori is chronic active gastritis comprised of polymorphonuclear leukocytes together with Th1, Th17, and Treg CD4+ lymphocytes [41]. The cagPAI is central to the inflammatory response because H. pylori strains bearing the cagPAI are more often associated with clinical disease in humans, rather than asymptomatic infection. These epidemiologic observations are supported by studies showing that strains harboring isogenic deletions within the cagPAI cause less gastritis and precancerous pathology in animal models than do strains with an intact cagPAI [35], [42], [43]. Yet from the bacterial perspective, the cagPAI has mixed effects. On the one hand, enhanced inflammation induced by the T4SS partially controls the infectious burden and presumably decreases transmission and therefore fitness. On the other hand, T4SS-mediated injection of CagA enhances the fitness of H. pylori by altering epithelial cell polarity and increasing bacterial iron acquisition, which permits it to grow on the apical epithelial cell surface [7], [44]. Here we demonstrate that H. pylori has evolved a novel solution to this dilemma, in which cagY, an essential component of the T4SS, has highly repetitive DNA elements that undergo rearrangements that can change the functionality of the cagPAI. These rearrangements may occur in vivo, but they are likely also present in the bacterial inoculum, since we could identify cagY variants relatively easily in vitro (2 of 30 clones examined). Although we have not formally identified recombination as the mechanism (e.g., horizontal gene transfer is possible), this seems most likely given the high frequency of repetitive elements within the cagY gene. We propose that cagY is a sort of contingency locus [45] that generates diversity at the population level and enhances bacterial fitness by allowing adaptation to changing conditions that may be found within one host or during transmission to another.
The most obvious pressure that may select for variant cagY alleles is the host adaptive immune response. Earlier studies suggested that the repeat structure of cagY represented a mechanism for antigenic variation to evade adaptive immunity [18], which is consistent with our finding that variant cagY alleles develop during colonization of WT but not RAG−/− mice (Figures 3, 9). However, strains recovered from monkeys and WT mice infected with H. pylori J166 sometimes maintained the cagY of the input strain, even after prolonged colonization when adaptive immunity would be fully developed (Figures 1, 3). Moreover, humans chronically colonized with H. pylori do not mount a serum immune response to CagY [18]. Thus, avoiding adaptive immunity may not be adequate to explain our results. An alternative hypothesis is that CagY variants serve not to evade the host immune response, but rather to “tune” it so as to establish the optimal homeostatic conditions of inflammation under which H. pylori is most fit. This hypothesis is supported by our finding that infection of monkeys and mice can select H. pylori strains with either loss of function (Figures 1,3,9) or gain of function (Figure 10A) in the T4SS, and the observation that the cagY genotype is relatively stable in WT mice when it confers a non-functional T4SS (Figure 10B). Finally, the very fact that many functional and non-functional variants of CagY arise in vivo by recombination, suggests that inflammation must be more advantageous to the bacterium in some situations than in others. Studies in humans have sometimes identified patients with mixed populations of cagPAI+ and cagPAI− strains [46]. Some have suggested that there is in fact a dynamic equilibrium between cagPAI+ and cagPAI− strains, creating a sort of H. pylori quasispecies, where some PAI variants may be better suited for transmission to a new host, and others better adapted for chronic persistence [47]. cagPAI+ strains isolated from an individual patient may also differ markedly in functionality of the T4SS [48], which might be explained by variations in the motif structure of CagY, but could also arise from mutations in other cagPAI genes. However, given the high frequency of cagY recombination, it seems likely that this mechanism is a much more common strategy by which H. pylori modulates its capacity to induce inflammation than is, for example, frameshift mutation, or gain or loss of the entire cagPAI.
There may also be differences in the relative fitness of H. pylori strains with a functional or a non-functional T4SS, depending on the inflammatory response of an individual host. When infected with WT H. pylori J166, most monkeys selected for loss of function in the T4SS, though one did not, even after 14 months of colonization (Figure 1). Similarly, when infected with mOut2, which has a non-functional T4SS, most monkeys selected for strains with a gain of function, but one did not (Figure 10A). Interestingly, in the one monkey available for long-term follow up, all strains recovered up to 24 weeks PI continued to induce IL-8 (Figure S7). Individual differences in strains recovered from outbred macaques may reflect host polymorphisms in the inflammatory response to H. pylori, which are well known to exist in humans and to have important clinical consequences [49]. Differences were also seen in individual WT C57BL/6 mice, which sometimes had persistent colonization with WT J166, even after prolonged infection when most mice selected for non-functional cagY variants (Figure 3). At first glance this is surprising, since inbred C57BL/6 mice are usually thought to be genetically identical. However, infection of mice with Helicobacter can yield both a resistant (low bacterial load, severe pathology, extensive CD4+ T cell infiltration, high IFN-γ) and a tolerant phenotype [50], so inbred mice may in fact be more genetically diverse than is usually thought [51], [52]. If inflammation is critical to the H. pylori lifestyle, yet is variable among hosts, modulation of T4SS function by recombination in cagY may provide a flexible mechanism to colonize and adapt to heterogeneous populations.
Strains expressing variant cagY alleles with loss of T4SS function are indistinguishable from a cagPAI or cagY KO in their IL-8 induction and CagA phosphorylation, which suggests that they are defective in translocation of CagA and peptidoglycan. Structural and functional studies of the VirB10 orthologue in other Gram-negative bacteria provide some basis for speculation on potential mechanisms by which this might occur. Cryo-EM and crystallography studies of the T4SS encoded by the conjugative plasmid pKM101 showed that VirB10 assembles with VirB7 and VirB9 to form the outer surface of a core complex that spans the inner and outer bacterial membranes [17], [53]. The C-terminus portion of CagY that is homologous to VirB10 also forms a complex with the H. pylori VirB9 orthologue (CagX) [16], [54]. Similar to the energy coupling protein TonB, VirB10 in A. tumefaciens undergoes an energy dependent conformational change that is required for complex formation with VirB7 and VirB9, and subsequent delivery of the T-DNA substrate [55]. Recently a mutation has been identified in VirB10 from A. tumefaciens that confers a secretion system defect and regulates substrate passage across the bacterial outer membrane [56]. Hence, one mechanism by which CagY variants might alter function of the H. pylori T4SS is by gating the transfer of CagA, peptidoglycan, or other bacterial effectors across the host cell membrane.
Changes in the CagY MRR might also affect T4SS function by altering the binding to β1 integrins, which is essential for CagA translocation and signaling [1], [2]. A previous study suggested that the CagY MRR decorates the T4SS pilus [31]; another reported that pili are not observed after deletion of cagY, though the data were not shown [2], [32]. Changes in the modular structure of the MRR might affect T4SS function, either directly or by changing the integrin binding of other T4SS components required for pilus assembly [15]. However, we failed to find evidence of the CagY MRR on the surface of the T4SS pili, and no differences in pilus morphology were observed after deletion of cagY. Moreover, yeast two-hybrid studies suggest that β1 integrin binding occurs only with the CagY C-terminus [1], which is the region with homology to the A. tumefaciens VirB10 that spans the inner and outer bacterial membrane, However, extrapolation from studies of A. tumefaciens may be limited, because the predicted molecular mass of H. pylori CagY is 220 kDa, much larger than the predicted 45 kDa VirB10 from A. tumefaciens, which does not contain a region orthologous to the H. pylori MRR. For the moment, these inconsistencies remain unresolved.
In conclusion, we have identified a functional plasticity in the H. pylori T4SS. We propose that immune-driven host selection of rearrangements in CagY modulates the function of the H. pylori T4SS and “tunes” the host inflammatory response so as to maximize persistent infection. Future studies should address the mechanism by which CagY recombination alters T4SS signaling, and identify the immune effectors that select CagY variants.
All animal experiments were performed in accordance with NIH guidelines, the Animal Welfare Act, and U.S. federal law. All experiments were carried out at the University of California, Davis under protocol #15597 approved by U.C Davis Institutional Animal Care and Use Committee (IACUC), which has been accredited by the Association of Assessment and Accreditation of Laboratory Animal Care (AAALAC). All animals were housed under these guidelines in an accredited research animal facility fully staffed with trained personnel.
H. pylori strain J166 has a functional cagPAI and colonizes both mice [27] and rhesus macaques [42]. H. pylori SS1 is a mouse-adapted derivative [33] of strain PMSS1, which is a human clinical isolate that has a functional cagPAI and also colonizes mice [35]. All H. pylori plate cultures were performed on brucella agar (BBL/Becton Dickinson, Sparks, MD) supplemented with 5% heat-inactivated newborn calf serum (Invitrogen, Carlsbad, CA) and either ABPNV (amphotericin B, 20 mg/liter; bacitracin, 200 mg/liter; polymyxin B, 3.3 mg/liter; nalidixic acid, 10.7 mg/liter; vancomycin, 100 mg/liter) or TVPA (trimethoprim, 5 mg/liter; vancomycin, 10 mg/liter; polymyxin B, 2.5 IU/liter, amphotericin B, 2.5 mg/liter) antibiotics (all from Sigma), for mouse and monkey experiments, respectively. H. pylori liquid cultures for mouse and monkey inoculation were grown in brucella broth with 5% NCS and antibiotic supplementation for approximately 24 h (optical density at 600 nm 0.35 to 0.45), pelleted by centrifugation, and suspended in brucella broth. All H. pylori cultures were grown at 37°C under microaerophilic conditions generated either by a 5% CO2 incubator or by a fixed 5% O2 concentration (Anoxomat, Advanced Instruments, Norwood, MA). A complete list of strains and plasmids is shown in Table S2.
Male and female specific pathogen free rhesus macaques aged 3 to 6 years were derived at the California National Primate Research Center from the day of birth using methods previously described to ensure that they had normal gastric histology and were free of H. pylori infection [57]. Animals were housed individually and fed commercial primate chow (Purina) and fruit, with water available ad libitum. Macaques were orogastrically inoculated by endoscopy with 109 CFU of H. pylori suspended in 2 ml of brucella broth. Endoscopy with gastric biopsy was performed with ketamine anesthesia (10 mg/kg given intramuscularly) after an overnight fast at defined time points PI. Specific-pathogen (Helicobacter)-free, female C57BL/6 and RAG2−/− mice (Taconic, Germantown, NY), or C57BL/6 and RAG1−/− mice (Jackson Laboratories) were housed in microisolator cages and provided with irradiated food and autoclaved water ad libitum. At 10 to 12 weeks of age mice were fasted for 3 to 4 hr and then challenged with 2.5×109 CFU of H. pylori suspended in 0.25 ml of brucella broth administered by oral gavage with a ball-end feeding needle. All mice were euthanized between 2 and 16 weeks post inoculation (PI) with an overdose of pentobarbital sodium injection (50 mg/ml IP). Stomachs were cut longitudinally, and half was placed in brucella broth, weighed, ground with a sterile glass rod until the mucosal cells were homogenized, and then plated quantitatively by serial dilution on brucella agar supplemented with 5% NCS and ABPNV. Multiple single colony isolates recovered from mice and monkeys were characterized for their capacity to induce IL-8 and translocate CagA. All animals were housed under protocols approved by ALAAC and the U.C. Davis Institutional Animal Care and Use Committee.
IL-8 was measured essentially as described previously [58]. Approximately 2.5×105 human AGS gastric adenocarcinoma cells (ATCC, Manassas, VA) were seeded in six well plates, washed two times with 1× PBS, and overlaid with 1.8 ml RPMI/10% fetal bovine serum and bacteria diluted in 200 µl brucella broth to give an MOI of 100∶1. Brucella broth with no bacteria served as a baseline control. Supernatants were harvested after 22 hours of culture (37°C, 5% CO2), stored at −80°C, and then diluted 1∶4 prior to IL-8 assay by ELISA (Invitrogen, Camarillo, CA) performed according to the manufacturer's protocol. WT H. pylori J166 and its isogenic cagY deletion were included on every plate as positive and negative controls, respectively. Results in AGS cells were confirmed selectively using KATO III gastric adenocarcinoma cells (ATCC, Manassas, VA) grown in RPMI 1640 (Gibco BRL, Grand Island, NY) with 20% fetal bovine serum. To account for variability in the assay, IL-8 values were normalized to WT H. pylori determined concurrently.
AGS cells stably transfected with an NF-κB luciferase reporter (Promega E849A, Madison, WI) were plated without antibiotics in a 48-well plate at a density of 3×104 cells per well for 24 hr prior to co-culture. H. pylori strains were grown overnight in liquid culture, diluted 10-fold in fresh media, and re-incubated for 4 hr to achieve log phase growth. Bacterial cells were washed once in sterile PBS and co-cultured with the AGS cells at an MOI of 10∶1 for 4 hr. Phorbol myristate acetate (PMA, 0.5 µg/mL) was used as a positive control. After 4 hr of co-culture, the media was removed, 100 µL/well of lysis buffer (Promega E4030) was added and mixed on an orbital shaker at 500 rpm for 10 min. To measure the luciferase activity, 100 µL of substrate (Promega E4030) and 20 µL of cell lysate were mixed and immediately read in a luminometer.
Expression of CagA, phosphorylated CagA, and CagY were detected by immunoblot. For detection of CagA translocation, AGS cells were washed twice with 2 ml RPMI 1640 (Invitrogen) containing 1 mM sodium orthovanadate, and pelleted by centrifugation (14,000 g, 30 sec). Pellets were lysed in 100 µl of NENT (1% NP40, 5 mM EDTA, 250 mM NaCl, 25 mM Tris, 1 mM sodium orthovanadate, 1 mM phenylmethylsulfonyl fluoride), centrifuged (14,000 g, 3 min), and electrophoresed in a 7.5% polyacrylamide gel (BioRad, Hercules, CA). Proteins were transferred to a PVDF membrane (Millipore, Billerica, MA), blocked overnight in 3% BSA in TTBS (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.05% Tween 20, 3% bovine serum albumin), and incubated for 1 hr with mouse anti-phosphotyrosine IgG (Santa Cruz Biotechnology, Santa Cruz, CA) diluted 1∶5,000. Blots were washed three times for 5 min each in TTBS and incubated for 1 hr with horseradish peroxidase (HRP)-conjugated anti-mouse IgG (GE Healthcare, Buckinghamshire, UK) diluted 1∶10,000. Bound antibody was detected with chemiluminescence using ECL reagents (GE Healthcare, Bukinghamshire, UK). The blot was then incubated in stripping buffer (0.1 M β-mercaptoethanol, 10% SDS and 0.5 M Tris, pH 6.8) for 30 min at 50°C, washed and blocked as before, and immunoblotted for 1 hr with rabbit IgG antibody (1∶5,000) to CagA (Austral Biological, San Ramon, CA). Blots were washed in TTBS, incubated for 1 hr with anti-rabbit HRP-conjugated IgG (GE Healthcare, Buckinghamshire, UK) at 1∶20,000 dilution, and visualized by chemiluminescence. CagY expression was detected by electrophoresis of sonicated bacterial proteins on a 7.5% polyacrylamide gel, incubating with rabbit antiserum (1∶10,000) to CagY [18] as primary antibody and HRP-conjugated anti-rabbit IgG (1∶20,000) as secondary antibody, followed by chemiluminescent detection.
cagY genotyping was performed by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP). A fragment containing the cagY gene was PCR amplified using the Expand Long Template PCR System (Roche Diagnostics, Indianapolis, IN). Reactions were performed in a total volume of 50-µl containing 100 ng of genomic DNA, 0.3 µM of each primer (sense 5′-CCGTTCATGTTCCATACATCTTTG-3′; anti-sense 5′-CTATGGTGAATTGGAGCGTGTG -3′), 0.35 mM of each dNTP, 3.75 U of Expand Taq DNA polymerase, and 1× buffer containing 1.75 mM MgCl2. PCR products were purified (QIAquick PCR Purification Kit, QIAGEN Sciences, Maryland, MD), adjusted to a concentration of 120 µg/ml, and digested overnight at 37°C separately with DdeI, BfucI, and HinfI (New England BioLabs, Ipswich, MA). Digested DNA was separated by 3% (HinfI) or 5% (DdeI, BfucI) agarose gel electrophoresis and then stained with ethidium bromide. Gels were examined and cagY from each output colony was determined to be the same as that of the J166 input strain if RFLP patterns were identical for all three restriction enzymes. Oligonucleotide primers for amplification, sequencing, and PCR-RFLP analysis of cagY are shown in Table S3.
Alleles of cagY were exchanged between H. pylori strains using contraselectable streptomycin susceptibility [26] modified essentially as described previously [27]. The 1,420 bp cat-rpsL cassette encoding chloramphenicol resistance and dominant streptomycin susceptibility was amplified with primers (RpsLF, C2CamR) that contained SacI and BamHI restriction sites, ligated between fragments of DNA upstream (1,348 bp, primers cagXF, cagYR) and downstream (1,122 bp, primers cagYF and virB11R) of cagY that contained complementary restriction sites, and cloned into pBluescript (Stratagene, La Jolla, CA). H. pylori was made streptomycin resistant by transformation with genomic DNA from a mutant of strain 26695, which contained an A-to-G change at codon 43 of rpsL, and selection on streptomycin (10 µg/ml). Transformation of streptomycin-resistant H. pylori with plasmid containing the cat-rpsL cassette and flanking cagY sequences, with selection on chloramphenicol (5 µg/ml), resulted in the replacement of bp 13 to 6,135 of cagY. The cagY gene of interest was then reinserted by transformation of the cagY knockout with genomic DNA from the donor strain and selection on streptomycin. Streptomycin-resistant, chloramphenicol-sensitive colonies were fully sequenced at the cagY locus to confirm that they had undergone the desired genetic exchange.
H. pylori was imaged by FEG-SEM using methods previously described [15]. In brief, H. pylori and AGS human gastric cells were co-cultured at an MOI of 100∶1 on tissue culture-treated coverslips (BD Biosciences) for 4 h at 37°C in the presence of 5% CO2. Cells were fixed with 2.0% paraformaldehyde, 2.5% glutaraldehyde in 0.05 M sodium cacodylate buffer for 1 hr at 37°C. Coverslips were washed with sodium cacodylate buffer and secondary fixation was performed with 1% osmium tetroxide at room temperature for 30 min. Coverslips were washed with sodium cacodylate buffer and dehydrated with sequential washes of increasing concentrations of ethanol. Samples were then dried at the critical point, mounted onto sample stubs, grounded with a thin strip of silver paint at the sample edge, and sputter-coated with palladium-gold before viewing with an FEI Q250 FEG scanning electron microscope. Image analysis was performed using Image J software.
Bacteria were co-cultured with AGS cells and fixed as for FEG-SEM. Cells were then washed three times in 0.05 M sodium cacodylate buffer before blocking in 0.1% cold fish skin gelatin in 0.05 M sodium cacodylate buffer for 1 hr. Primary polyclonal rabbit antibodies to CagA and the CagY MRR [18] were applied overnight followed by three buffer washes and application of secondary goat anti-rabbit antibody conjugated to 20 nm gold particle (Electron Microscopy Sciences, Hatfield, PA) for 4 hr. After three buffer washes, cells were fixed again (2.0% paraformaldehyde, 2.5% glutaraldehyde in 0.05 M sodium cacodylate) for 1 hr to stabilize the antibody interactions, washed, and then treated with 0.1% osmium tetroxide for 15 min followed by three additional buffer washes and sequential ethanol dehydration. Cells were dried at the critical point and carbon-coated before imaging with an FEI Quanta 250 FEG-SEM. Gold particles were confirmed with backscatter imaging analysis. As negative controls, uninfected AGS cells were processed in parallel, or application of the primary antibody was omitted.
cagPAI genes known to be involved in IL-8 induction were amplified and sequenced using primers shown in Table S3. cagY genes were amplified with primers in flanking genes using Expand Long Template PCR system (Roche, Indianapolis, IN). Purified PCR products were cloned into pDrive (Qiagen, Valencia, CA) and plasmids were sequenced with dye terminator chemistry. PCR products were sometimes sequenced directly for verification. To confirm the number of 390 bp repeats in the FRR, the cagY PCR products were run on 0.4% agarose gels at 18 volts for 16 hr. The size of the PCR product minus 477 bp gave an estimate of total cagY size. All DNA sequences of cagY have been deposited in GenBank under accession numbers JQ685133–JQ685155.
Data were analyzed using a 2-tailed Student's t test (Prism 5.0) unless otherwise indicated. A P value<0.05 was considered statistically significant.
|
10.1371/journal.pgen.1006799 | Evolutionary forces affecting synonymous variations in plant genomes | Base composition is highly variable among and within plant genomes, especially at third codon positions, ranging from GC-poor and homogeneous species to GC-rich and highly heterogeneous ones (particularly Monocots). Consequently, synonymous codon usage is biased in most species, even when base composition is relatively homogeneous. The causes of these variations are still under debate, with three main forces being possibly involved: mutational bias, selection and GC-biased gene conversion (gBGC). So far, both selection and gBGC have been detected in some species but how their relative strength varies among and within species remains unclear. Population genetics approaches allow to jointly estimating the intensity of selection, gBGC and mutational bias. We extended a recently developed method and applied it to a large population genomic dataset based on transcriptome sequencing of 11 angiosperm species spread across the phylogeny. We found that at synonymous positions, base composition is far from mutation-drift equilibrium in most genomes and that gBGC is a widespread and stronger process than selection. gBGC could strongly contribute to base composition variation among plant species, implying that it should be taken into account in plant genome analyses, especially for GC-rich ones.
| In protein coding genes, base composition strongly varies within and among plant genomes, especially at positions where changes do not alter the coded protein (synonymous variations). Some species, such as the model plant Arabidopsis thaliana, are relatively GC-poor and homogeneous while others, such as grasses, are highly heterogeneous and GC-rich. The causes of these variations are still debated: are they mainly due to selective or neutral processes? Answering to this question is important to correctly infer whether variations in base composition may have functional roles or not. We extended a population genetics method to jointly estimate the different forces that may affect synonymous variations and applied it to genomic datasets in 11 flowering plant species. We found that GC-biased gene conversion, a neutral process associated with recombination that mimics selection by favouring G and C bases, is a widespread and stronger process than selection and that it could explain the large variation in base composition observed in plant genomes. Our results bear implications for analysing plant genomes and for correctly interpreting what could be functional or not.
| Base composition strongly varies across and within plant genomes [1]. This is especially striking at the coding sequence level for synonymous sites where highly contrasted patterns are observed. Most Gymnosperms, basal Angiosperms and Eudicots have relatively GC-poor and homogeneous genomes. In contrast, Monocot species present a much wider range of variation from GC-poor species to GC-rich and highly heterogeneous ones, some with bimodal GC content distribution among genes, these differences being mainly driven by GC content at third codon position (GC3) [1]. Commelinids (a group containing palm trees, banana and grasses, among others) have particularly GC-rich and heterogeneous genomes but GC-richness and bimodality have been showed to be ancestral to Monocots, suggesting erosion of GC content in some lineages and maintenance in others [2]. As a consequence, in most species, synonymous codons are not used in equal frequency with some codons more frequently used than others, a feature that is called the codon usage bias [reviewed in 3]. This is true even in relatively homogeneous genomes such as in Arabidopsis thaliana [e.g. 4].
Which forces drive the evolution of genome base composition and codon usage is still under debate. Mutational processes can contribute to observed variations between species and within genomes [e.g. 5]. However, mutation can hardly explain a strong bias towards G and C bases, as it is biased towards A and T in most organisms studied so far [Chapter 6 in 6]. Selection on codon usage (SCU) has thus appeared as one of the key forces shaping codon usage as it has been demonstrated in many organisms both in prokaryotes and eukaryotes [reviewed in 3]. Codon bias can thus result from the balance between mutation, natural selection and genetic drift [7]. The main cause for SCU is likely that preferred codons increase the accuracy and/or the efficiency of translation but other mechanisms involving mRNA stability, protein folding, splicing regulation and robustness to translational errors could also play a role [3,8,9]. In some species, SCU appears to be very weak or inexistent, typically when effective sizes are small [10], as typically assumed for mammals [but see 8]. However, mammalian genomes exhibit strong variations in base composition, the so-called isochore structure [11], which are mainly driven by GC-biased gene conversion (gBGC) [12]. gBGC is a neutral recombination-associated process favouring the fixation of G and C (hereafter S for strong) over A and T (hereafter W for weak) alleles because of biased mismatch repair following heteroduplex formation during meiosis [13]. Although gBGC is a neutral process–i.e. the fate of S vs. W alleles is not driven by their effect on fitness—gBGC induces a transmission dynamic during reproduction identical to natural selection for population genetics [14]. Therefore, we here refer to it as a “selective-like” process as opposed to mutation and drift. gBGC has been experimentally demonstrated in yeast [15,16], humans [17,18], birds [19] and rice [20]. Many indirect genomic evidences also supported gBGC in eukaryotes [21,22] and even recently in some prokaryotes [23], although it seems to be weak or absent in some species as Drosophila [24] where selection on codon usage predominates [25,26,27,28].
In plants, both SCU [4,29,30] and gBGC [21,31,32] have been documented, but how their magnitudes and relative strength vary among species remains unclear. Recently, it has been proposed that the wide variations in genic GC content distribution observed in Angiosperms could be explained by the interaction between gene structure, recombination pattern and gBGC [33]. Increasing evidence suggests that in various organisms, including plants, recombination occurs preferentially in promoter regions of genes, or near transcription initiation sites [34,35,36]. This generates a 5’-3’ recombination gradient, and consequently a gBGC gradient, which could explain the 5’-3’ GC content gradient observed in GC-rich species, such as Commelinids [1,2]. A mechanistic consequence is that short genes, especially with no or few introns, are on average GC-richer [37]. A stronger gBGC gradient and/or a higher proportion of short genes would increase the average GC content and simple changes in the gBGC gradient can explain a wide range of GC content distribution from unimodal to bimodal ones [33].
So far, the magnitude of gBGC and SCU has been quantified only in a handful of plant species [29,30,32,38]. As in other species studied, weak SCU and gBGC intensities were estimated. The population-scale coefficients, 4Nes or 4Neb, are usually of the order of 1, where Ne is the effective population size and s and b the intensity of SCU and gBGC respectively [26,29,30,32,38,39]. However, high gBGC values (4Neb > 10) have been estimated in the close vicinity of recombination hotspots in mammals [38,40] and across the entire honeybee genome [41]. Differences in population-scale intensities can be due to variation in Ne and/or in s or b. For gBGC, b is the product of the recombination rate r and the basal conversion rate per recombination event, b0. Within a genome, variations in gBGC intensities are mainly due to variation in recombination rate [e.g. 38]. Among species, b0 can also vary. For instance, b was estimated to be 2.5 times lower in honeybees than in humans but recombination rate is more than 18 times higher [41], suggesting that b0 could be 45 times lower in honeybees than in humans. The very intense population-scale gBGC in honeybees is thus explained by the combination of a large Ne and extremely high recombination rates [41].
Several methods have been developed to estimate the intensity of SCU and gBGC, either from polymorphism data alone, or from the combination of polymorphism and divergence data [e.g. 26,27,38]. These methods rely on the fact that preferred codons (for SCU) or GC alleles (for gBGC) are expected to segregate with higher frequency than neutral and un-preferred or AT alleles, fitting a population genetics model with selection or gBGC to the different site frequency spectra (SFS). As demography affects SFS, it must be taken into account in the model. Moreover, mutations must be polarized, i.e. the ancestral or derived state of mutations must be determined using one or several outgroup species. Otherwise, selection or gBGC can be estimated from the shape of the folded SFS by assuming equilibrium base composition [42] or allowing only recent change in base composition [e.g. 25,26,27], which is not the case in mammals [43] and some Monocots [2], for example. As errors in the polarization of mutations can lead to spurious signatures of selection or gBGC [44], this issue must also be taken into account.
We specifically address the following questions: (i) do neutral or selective forces mainly affect base composition? (ii) if active, what are the intensities of gBGC and SCU and how do they vary across species? (iii) are the average gBGC and the 5’-3’ gBGC gradient stronger in GC-rich genomes? To do so we used and extended the recent method developed by Glémin et al. [38] that controls for both demography and polarization errors. We applied it to a large population genomic dataset of 11 species spread across the Angiosperm phylogeny to detect and quantify the forces affecting synonymous positions. Our results show that base composition is far from mutation-drift equilibrium in most studied genomes, that gBGC is a widespread process being the major force acting on synonymous sites, overwhelming the effect of SCU and contributing to explain the difference between GC-rich (Commelinids, here) and GC-poor genomes (Eudicots and yam, here).
We focused our analyses on 11 plant species spread across the Angiosperm phylogeny with contrasted base composition and mating systems (Fig 1 and Table 1). To survey the wide variation observed in Monocots, and in line with the sampling of a previous study [2], we sampled one basal Monocots (Dioscorea abyssinica, yam), two non-grass Commelinids (Musa acuminata, banana and Elaeis guineensis, palm tree) and three grasses with contrasted mating system (Pennisetum glaucum, pearl millet, Sorghum bicolor, sorghum and Triticum monococcum, einkorn wheat). In Eudicots, both Rosids (Theobroma cacao, cacao and Vitis vinifera, grapevine) and Asterids (Coffea canephora, coffee tree, Olea europaea, olive tree and Solanum pimpinellifolium, tomato) are represented. For practical reasons cultivated species have been chosen but we only sampled wild individuals over the species range, except for palm tree for which cultivated individuals were sampled (See S1 Table for sampling details). In this species cultivation is very recent without real domestication process (19th century [45]). For each species, we used RNA-seq techniques to sequence the transcriptome of about ten individuals plus two individuals from two outgroup species, giving a total of 130 individual transcriptomes. Using transcriptomes has been shown to be a useful approach for comparative population genomics with no or minor bias for genome wide comparison [46,47]. When a well-annotated reference genome was available (see Material and methods), we used it as a reference for read mapping. Otherwise we used a de novo transcriptome assembly already obtained for these species (focal + outgroups) [48] (Table 1 and S2 Table). After quality trimming and mapping of the raw reads, we kept contigs with at least one read mapped for every individual, giving between more than 24,000 (P. glaucum) and 45,000 (in O. europaea) contigs per species (Table 1). This initial dataset was used for gene expression analyses (see below). Genotype calling and filtering of paralogous sequences were performed using the read2snp software [47] for each species separately, and coding sequence regions were extracted (see Material and methods). The resulting datasets were used to compute nucleotide diversity statistics that did not require any outgroup information. The number of identified SNPs varies from 4,409 in T. monococcum (which suffered from the lowest depth of sequencing) to 115,483 in C. canephora. Variations in the numbers of SNPs also revealed the large variation in polymorphism levels with πS ranging from 0.17% in E. guineensis to 1.22% in M. acuminata. The level of constraints on proteins, as measured by the πN/πS ratio, varies between 0.122 in T. monococcum and 0.261 in E. guineensis (Table 2). For the analyses requiring polarized SNPs, we also added orthologous sequences from two outgroups to each sequence alignment of the focal species individuals (see Material and methods). The number of polarized SNPs ranged from 3,253 in S. pimpinellifolium to 89,793 in M. acuminata. Other details about the datasets are given in Table 2. Overall, although the dataset does not represent the full transcriptome of each species it allows large-scale comparative analyses.
We first looked at base composition: GC3 varies from 0.38 to 0.44 in Eudicots and from 0.46 to 0.56 in Monocots (Table 2). As observed in previous studies [2,43], these values tend to be lower than genome wide averages (when available) but the relative differences in base composition among species were conserved, notably the GC-poorness of Eudicots compared to Monocots. Grass species exhibited a bimodal GC3 distribution except T. monococcum where bimodality was not apparent (S1 Fig). This is likely because the sequencing depth was lower for this species so that GC-rich genes (most likely short ones [37]) have been under sampled. We also characterized codon usage in each species by computing the Relative Synonymous Codon Usage (RSCU) for every codon as the frequency of a particular codon normalised by the frequency of the amino acid it codes for (S3 Table, S2 Fig). Patterns of RSCU were relatively consistent between species but reflected differences of GC content between them, notably a higher usage of G or C-ending codons in GC-rich species.
In order to evaluate the possible effect of selection on codon usage, we defined the sets of preferred (P) and un-preferred (U) codons for each species. The fitness consequences of using optimal or suboptimal codons should be higher in highly expressed genes, causing the usage of optimal codons to increase with gene expression (and that of non-optimal ones to decrease). Thus, we defined preferred (or un-preferred) codons as codons for which RSCU increases (or decreases) with gene expression as in [49] (see Materials & methods for more details). S3 Table shows detailed results for each species. In contrast with genome-wide codon usage, nearly all species showed a bias towards preferred codons ending in G or C (Table 2, Fig 2 and S3 Table), only P. glaucum and S. bicolor showing a more balanced AT/GC sharing of codon preference. Preferences for two-fold degenerated codons were highly conserved across species, with only GC-ending preferred codon except for aspartic acid and tyrosine in P. glaucum (Fig 2, S3 Table). Preferences for other amino acids were slightly more labile but there were always one preferred GC-ending and one un-preferred AT-ending codon common to all species. Frequency of optimal codons of a gene (Fop, i.e. the frequency of preferred codons [50]), increased with expression as expected but the difference in Fop between the most highly and most lowly expressed genes was weak to moderate (from ~5% in C. canephora to 15% in T. monococcum and M. acuminata) and tended to be higher in Commelinid species (Fig 3). Because most preferred codons ended with G or C, GC3 and expression were also positively correlated in all species.
To determine which forces affect variation in base composition and codon usage among species, we first evaluated whether base composition at synonymous sites was at mutation-drift equilibrium. Glémin et al. [38] showed that the asymmetry of the distribution of non-polarized GC allele frequencies (measured by the skewness coefficient of the distribution) was a robust test of this equilibrium. This statistic is not affected by possible polarization errors (see later for more on polarization errors). A skewness coefficient equal to 0 is expected under equilibrium whereas negative (or positive) values mean higher (or lower) GC content than expected under mutation-drift equilibrium. The same rationale can be applied to codon frequencies. We found that GC content and the frequency of preferred codons were significantly higher than predicted by mutational effects in all species, with the exception of coffee, which interestingly showed a lower GC content than expected under mutation-drift balance (Table 3).
As base composition equilibrates slowly under mutation pressure [33], non-equilibrium conditions could be due to long-term changes in mutational patterns. To test further whether selective-like forces can explain the excess of GC and preferred codons, we developed a modified MacDonald Kreitman test [51] comparing W→S (or U→P) to S→W (or P→U) polymorphic and divergent sites (Material & Methods and S1 Text). SNPs and fixed mutations (substitutions) were polarized by parsimony using two outgroup taxa for each focal species. We built contingency tables by counting the number of polymorphic or divergent sites for each of the two mutational categories. From these contingency tables, we computed neutrality, NI, [52] and direction of selection, DoS, [53] indices. In the case of selective-like forces favouring the fixation of W→S or U→P mutation, NI values are expected to be lower than 1 and DoS values to be positive. P-values were computed from a Chi-squared test on the contingency tables. NI was lower than 1 and DoS positive in all species except S. pimpinellifolium (Table 3), indicating that selective-like forces drove the fixation of GC and preferred codon alleles. In P. glaucum, although significant, the departure from the neutral expectation for GC content is minute, which can be explained by very weak gBGC but also by a recent increase in its intensity (see Results below and S1 Text). Overall, this analysis showed that in most species selective-like forces tended to drive base and codon composition away from their mutational equilibrium. Selection and gBGC are the two known alternatives whose effects have to be distinguished.
Although they may have different mechanistic causes and biological consequences, selection and gBGC leave similar evolutionary footprints and are not easy to disentangle, especially in species where most preferred codons end in G or C (Table 2). We first applied correlative approaches to try to disentangle both processes. Then we tried to quantify their respective intensities.
Under the SCU hypothesis, departure from neutrality should be stronger for highly expressed genes and/or genes with strongly biased codon composition. Under the gBGC hypothesis, departure from neutrality should increase with recombination rates. However, recombination data was not available in our datasets. As gBGC leads to an increase in GC content, departure from neutrality should thus also increases with GC content. We split synonymous SNPs and substitutions into eight groups of same size according to their GC3 or their gene expression level (measured by the mean RPKM values across all individuals of a given population), and computed the NI and DoS indices for each category based on W/S or U/P changes. For all species except D. abyssinica and S. bicolor, we found a strong positive (or negative) correlation between GC3 and DoS (or NI), indicating a stronger bias in favour of S alleles in GC-rich genes (Fig 4). In contrast, the relationship between expression level and DoS or NI measured on codon usage was weaker, with more variable and on average lower correlation coefficients (Fig 4). These results tend to point out gBGC as a stronger force than SCU affecting synonymous variations in our datasets.
We then split our datasets into four independent categories based on two GC3 groups crossed by two expression level groups to test which factor has the strongest effect on the bias towards S or P alleles. The rationale is that SCU should make the bias towards P alleles increase with gene expression independently of GC3. On the other hand, gBGC should increase the bias towards S alleles with GC3 independently of gene expression. We found that DoS clearly increased with GC3 in all species for both lowly and highly expressed genes, with the exception of D. abyssinica and S. bicolor where it decreased for lowly expressed genes, and S. pimpinellifolium where there was little change for lowly expressed genes. On the other hand, the effect of expression on DoS was inconsistent or only weak in most species (Fig 5). These results confirm that the effect of gBGC appears stronger than the effect of SCU.
To evaluate further the forces affecting base composition we estimated the intensity of selection (S = 4Nes) and gBGC (B = 4Neb) from site frequency spectra (SFS). SFS for all species are represented in S3 Fig. We used the method recently developed by Glémin et al. [38] that takes SNP polarization errors into account, which avoids observing spurious signature of selection or gBGC. As mentioned above, the observed pattern in P. glaucum (excess of GC content but almost no departure from neutrality according to the NI and DoS indices, see Table 3) suggests a recent change in the intensity of selection and/or gBGC. Also, transition to selfing, which usually can be very recent in plants [54], could have effectively shut down gBGC in the recent past due to a deficit in heterozygous positions. To capture these possible changes of fixation bias through time, we extended the model of [38] by combining frequency spectra and divergence estimates as summarized on Fig 6 (and see S2 Text for full details). Divergence is determined by both mutation and selection/gBGC so it is not possible to disentangle these two factors from the divergence data alone. However, if we assume constant and identical mutation bias at the polymorphism and the divergence level, this leave enough degrees of freedom to fit an additional S or B parameter. Thus, we assumed a single mutation bias but two different selection/gBGC intensities, one fitted on polymorphism data and the other on divergence. We evaluated the statistical significance of the shift in intensity by a likelihood ratio test with the model where the two intensities were equal (i.e. no change over time). Simulations showed that not taking polarization errors into account can bias selection/gBGC estimates as already shown in [38] and also leads to spurious detection of changes in selection/gBGC intensities (S2 Text). Simulations also showed that the estimated differences between the two intensities were often underestimated. This is expected as B values estimated in the model correspond to averages over the conditions that mutations have experienced during their lifetime (drift and gBGC/selection intensities), so it depends on when changes occurred. However, the method accurately retrieved the appropriate weighted averages for B0 and B1 and efficiently accommodates for demographic variations (see S2 Text). Overall, the test of heterogeneity of selection/gBGC is a conservative approach. If we relax the assumption of constant mutational bias, changes in both bias and selection/gBGC are no more identifiable. Recent S/B estimates are not affected but ancestral estimates are underestimated (resp. overestimated) when mutation bias decreases (resp. increases). However, the method is still powerful to detect departure from a constant regime of selection/mutation/drift equilibrium (S2 Text).
We applied the method to the total frequency spectra, either for W/S or U/P polymorphisms and substitutions. In all species, significant (at the 5% level) gBGC or SCU were detected but at low intensity (B or S < 1, Table 4). In four species (P. glaucum, E. guineensis, D. abyssinica and V. vinifera) we found significant differences between ancestral and recent intensities for gBGC and/or SCU. In particular, the recent significant increase in gBGC in P. glaucum (from 0.224 to 0.524, Table 4) can explain why NI is very close to one (or DoS close to zero) (see above and S1 Text). On average, Monocots, especially Commelinids species tended to exhibit stronger gBGC than Eudicots and B tended to increase with mean GC3, but no relationship is significant with only 11 species when either B0 or B1 are used. However, using the constant B estimates (S4 Table), weakly significant relationships were found for the difference between Commelinids and other species (Wilcoxon test: p-value = 0.0519) and the correlation between B and GC3 (ρSpearman = 0.691, p-value = 0.023). No significant relationship was found for SCU. No significant relationship between B or S and πS was found either.
As the two processes are entangled, it is difficult to properly and separately estimate their respective intensities. To do so, we developed a second extension of the method of [38]. Combining the two processes, nine kinds of mutations can occur (see S2 Text). By assuming that selection and gBGC act additively, it is in theory possible to estimate separately the two effects. We fit a general model to the nine SFS and the nine substitution counts, with a constant mutation bias, two B and two S values. The details of the model are reported in S2 Text. Simulations showed that the method could efficiently estimate both gBGC and SCU but tended to slightly underestimate recent gBGC and overestimate recent SCU (S2 Text). When the distributions of SNPs and substitutions are highly unbalanced (typically S/P and W/U states are confounded and there are very few WS-PU and SW-UP mutations), it is more difficult to detect both effects with a significant level (S2 Text). Finally, if assignation of codon preference is not perfect, typically for four-fold and six-fold degenerated codons, this could also underestimate SCU and reduce the power to detect it, especially for highly unbalanced dataset for which it is anyway inherently difficult to distinguish gBGC and SCU (see S2 Text). For both selection and gBGC and both ancestral and recent periods, we either fixed the value to 0 or let it be freely estimated, leading to 16 different models. For each species, the best model according to AIC criteria (see Methods) is given in Table 5 while all results are given in S5 Table. In six species the model with only gBGC was the best one, this could also include M. acuminata where it was not possible to disentangle between gBGC and SCU. For three species, the best model included both gBGC and SCU and only S. pimpinellifolium appeared to be affected by SCU but not gBGC. If codon preferences were perfectly determined, this result is expected to be robust and conservative because simulations suggest that SCU is slightly more easily detected than gBGC. If there were some errors in codon preference identification, this can partly explain that SCU was less often detected. However, the species for which SCU was not detected did not present the most unbalanced codon preference (see Table 2) and identification error rate should have been rather high (>20% see S2 Text) to strongly bias results. Overall, this confirms that synonymous sites are widely affected by gBGC in the studied plant species and that SCU either only plays a minor role or is partly masked by the effect of gBGC.
This method also allowed us to estimate mutation bias. As already observed in most species, mutation was biased towards AT alleles, with a bias slightly ranging from 1.6 to 2.2 (Table 4), which is of the same order as what was found in humans [38,55]. Interestingly, C. canephora was again an exception with almost no mutational bias (λ = 1.05).
So far, we considered either global effects at the transcriptome scale or variations among genes belonging to different categories. However, most plant species exhibit a more or less pronounced gradient in base composition from 5’ to 3’ [1], which is strongly linked to exon-intron structure [37]. In particular, in some species the first exon is much GC-richer than other exons. Moreover, it has been proposed that this gradient could be due to a gBGC gradient associated with a recombination gradient [33]. To quantitatively test this hypothesis, we separated SNPs and fixed derived mutations as a function of their position along genes. The best choice would have been to split them according to exon ranking [37]. However, as exon annotation was lacking (or imprecise) for most species in our datasets, we split contigs into two sets: the first 252 base pairs, corresponding to the median length of the first exon in Arabidopsis, banana and rice (Gramene database [56]), used as a proxy for the first exon, and the rest of the contig. We then estimated B on these two sets of contigs. Some imprecision in the “first exon” definition and variation in transcript length among species reduced the power of this analysis and results should be interpreted with caution. However, we did not expect that it could create artifactual B gradient as the use of a stringent criterion reinforced the observed patterns despite reducing datasets (see below).
For all species except D. abyssinica and S. pimpinellifolium, the ancestral B was higher in the first part than in the rest of contigs. The signature was less clear for recent B as far less values were significant. Ancestral and recent B were not significantly different in most species (S6 Table). To illustrate the global pattern, Fig 7 shows average gBGC gradients for all species, i.e. assuming the same ancestral and recent B values. Interestingly, while there was no clear taxonomic effect on global gBGC estimates (Table 4), there was a sharp difference between Commelinid species and the others for the first part of contigs (Wilcoxon test p-value = 0.030, Fig 7C), in agreement with the strong 5’– 3’ GC gradient observed in these species [1,2]. B values and GC3 tended to be positively correlated on the first part of contigs (ρSpearman = 0.591, p-value = 0.061) but not significantly in the rest (ρSpearman = 0.382, p-value = 0.248). These analyses were performed on all contigs but some of them do not start by a start codon. We restricted the analyses to the subset of contigs starting by a start codon and we found very similar results with stronger statistical support: in the first exon, B was significantly higher in Commelinids than in other species (Wilcoxon test p-value = 0.0043) and B values and GC3 were significantly and positively correlated both on the first part of contigs (ρSpearman = 0.80, p-value = 0.0052) and in the rest of contigs (ρSpearman = 0.70, p-value = 0.0208) (S6 Table and S4 Fig). In line with previous results showing that first exons contribute to most of the variation in GC content among species [2,33,37], these results show that species also mostly differ in their gBGC intensities in the first part of genes.
It has already been shown that base composition in grass genomes is not at mutation-drift equilibrium with both gBGC and selection increasing GC content despite mutational bias toward A/T [31]. Our results demonstrate that even in GC-poor genomes base composition is not at mutation-drift equilibrium, implying that selective-like forces are widespread in all the 11 plant species we studied. In all species, either the skewness and/or the DoS/NI statistics show evidence of departure from equilibrium and purely neutral evolution (Table 3). All species except C. canephora have higher GC content than predicted by mutational effect alone, which could be explained by a mutation/gBGC (or selection)/drift balance.
The case of C. canephora remains intriguing. Mutation seems not to be biased towards AT as observed in all mutation accumulation experiments [reviewed in 57] and through indirect methods [58]. So far, GC biased mutation has only been observed in the bacteria Burkholderia cenocepacia [57]. However, despite no apparent or very weak AT mutational bias and evidence of both recent and ancestral gBGC (Table 4), GC content is rather low (GC3 = 0.42, Table 2) and lower than expected under mutation pressure alone (1/(1+λ) = 0.49) as revealed by the positive skewness statistics (Table 3). Preferred codons mostly end in G or C (Table 2) so that SCU is not a possible explanation for this low GC content. Rather, a recent change in mutation bias is a more probable explanation. Using B0 = 0.154 or B1 = 0.243 (Table 4), a mutational bias of 1.61 or 1.76 would be necessary to reach the observed GC3 (= 0.42). Such values are in the same range as observed for the other species. D. abyssinica is another intriguing case where DoS decreases with GC content, contrary to other species (Fig 4). We currently have no clear hypothesis to explain this pattern and it should be viewed with caution because DoS is estimated with few substitutions in this species but it would be compatible with an increase in AT mutation bias with GC content. Further investigation of mutational patterns in these species would be useful to understand better these two intriguing cases.
Beyond departure from equilibrium, comparison of ancestral and recent gBGC or selection also reveals the dynamic nature of forces affecting base composition. At least four species (P. glaucum, E. guineensis, D. abyssinica and V. vinifera) show evidence of significant change in gBGC and/or SCU intensity over time (Table 4). If we consider the first part of genes only, changes also occurred in M. acuminata and T. cacao (S6 Table). Moreover, our method is conservative (see S2 Text) so we may have missed variations in other species. Changes occurred in both directions. In the three selfing or mixed mating species (S. pimpinellifolium, T. monococcum, and S. bicolor) the ancestral gBGC or SCU intensity is significantly positive but the recent one is null. This is supported by the rather recent evolution of selfing in these species, which nullifies the effect of gBGC through the increase in homozygosity levels and reduces the efficacy of selection [59]. In other species, gBGC or SCU have increased (e.g. P. glaucum) or decreased (e.g. V. vinifera). Recalling that B = 4Nerb0 (see Introduction), this could be explained by changes in effective population size (Ne) recombination rate (r), gBGC intensity per recombination event (b0) and also conversion tract length, which might also vary among species [60]. To date, we do not know anything about the stability of b0 across generations and how fast it can evolve. In some species, such as mammals, recombination can evolve very rapidly, at least at the hotspot scale [61] but it can be more stable in other species like in birds [62], yeast [63] or maize [64]. Moreover, we average gBGC over the whole transcriptome so recent genome-scale changes in recombination should be necessary to explain changes in B. Although recent changes in r and b0 are possible, changes in effective population size over time appears to be the most likely explanation.
Selective-like evolution and non-equilibrium conditions can have practical impacts on several genomic analyses. First, gBGC can lead to spurious signatures of positive selection [65], significantly increasing the rate of false positive in genome scan approaches in mammals [66]. This problem should also be taken into account in plant genomes, even in GC-poor ones. Second, SCU/gBGC and non-stationary evolution, due for instance to changes in population size, can strongly affect the estimation of the rate of adaptive evolution through McDonald-Kreitman approaches, especially at high GC content [67]. In species far from equilibrium such as Commelinids, it should be an issue to consider.
We detected gBGC in all but one species but its intensity is rather weak (Tables 4 and 5 and S4 and S5 Tables), of the same order to what was estimated in humans [38] but lower than in other mammals [39], maize [72], and particularly honey bee [41]. Low values can be explained by the fact that we only estimated average B values. In many plants studied so far, recombination was found to be heterogeneous along chromosomes [e.g. 36] and locally occurring in hotspots [e.g. 34,35,64], so that gBGC can be locally much higher than average estimates. However, we did not apply the hotspot model proposed by [38] because it behaves poorly when not constrained by additional information on hotspot structure, which we lack in the species studied here. In addition, recombination hotspots are preferentially located outside genes, especially in 5’ upstream regions (and 3’ downstream regions to a lesser extent) [34,35,36]. As we estimated gBGC intensities within coding regions, this can also explain why we only estimated rather weak B values.
A consequence of this specific recombination hotspot location is the induction of a 5’– 3’ recombination gradient along genes (or more generally an exterior to interior gradient if also considering downstream location) [34,35]. Recently, it has been proposed that this recombination gradient could explain the 5’– 3’ gradient observed in grasses and more generally in many plant species [33]. We tested this model by looking at signatures of gBGC along contigs in our datasets. In agreement with this model, we found stronger gBGC signatures at the 5’ end of contigs compared to the rest of contigs in most of our species (Fig 7). The fact that we observed this gBGC gradient in both Eudicots and Monocots suggests that all these species share the same meiotic recombination structure with preferential location of recombination in upstream regions of gene, which was hypothesized to be the ancestral mode of recombination location in Eukaryotes [73].
Glémin et al. [33] also proposed that changes in the steepness of the recombination/gBGC gradient could explain variation in GC content distributions among species, from unimodal GC-poor to bimodal GC-rich distributions. Alternatively, if gradients are stable among species, changes in gene structure, especially the number of short mono-exonic genes and the distribution of length of first introns, could also generate variations in GC content distribution [33,37]. Here we found that, in the first part of genes, gBGC is the highest in Commelinid species, which exhibit the richest and most heterogeneous GC content distributions (Fig 7). This result parallels the sharp difference in GC content in first exons between rice and Arabidopsis whereas the centres of genes have a very similar base composition [37]. Our results support the hypothesis that genic base composition in GC-rich and heterogeneous genomes has been driven by high gBGC/recombination gradients. As GC content bimodality is likely ancestral to monocot species and has been lost several times later [2], our results suggest that an increase in gBGC and or recombination rates occurred at the origin of the Monocot lineage.
Overall, we show that selection on codon usage only plays a minor role in shaping base composition evolution at synonymous sites in plant genomes and that gBGC is the main driving force. Our study comes along an increasing number of results showing that gBGC is at work in many organisms. Plants are no exception. If, as we suggest, gBGC is the main contributor to base composition variation among plant species, it shifts the question towards understanding why gBGC may vary between species and more generally why gBGC evolved. Our results also imply that gBGC should be taken into account when analysing plant genomes, especially GC-rich ones. Typically, claims of adaptive significance of variation in GC content should be viewed with caution and properly tested against the “extended null hypothesis” of molecular evolution including the possible effect of gBGC [65].
We focused our study of synonymous variations in 11 species spread across the Angiosperm phylogeny with contrasted base composition and mating systems, Coffea canephora, Olea europaea, Solanum pimpinellifolium, Theobroma cacao, Vitis vinifera, Dioscorea abyssinica, Elaeis guineensis, Musa acuminata, Pennisetum glaucum, Sorghum bicolor and Triticum monococcum. A phylogeny of these species is shown in Fig 1. For practical reasons, we chose diploid cultivated species but focused our analysis on wild populations except in Elaeis guineensis where domestication is very recent and limited (19th century [45]). Using the same methodology as [48], we sequenced for each species the transcriptome of ten individuals (12 in the case of C. canephora and V. vinifera, nine in the case of S. bicolor and five in the case of D. abyssinica) plus two individuals coming from two outgroup species, using RNA-seq (see S3 Text for details). After quality cleaning, reads were either mapped on the transcriptome extracted from the reference genome (when available, see Table 1) or on the de novo transcriptome of each species (including outgroups) obtained from [48]. For C. canephora and its outgroups, no transcriptome was available. We thus applied the same methodology and pipeline as in [48] to assemble and annotate contigs. For banana, M. acuminata, Robusta coffee tree, C. canephora, and for the outgroup Phoenix dactylifera, genome sequences were available but the quality of mapping was not optimal because of problems of definition of exon/intron boundaries. We thus preferred assembling a new transcriptome from our data using the same protocol. Details of the assemblies for all species are given in S2 Table. Details of data processing are provided in S4 Text. Only contigs with at least one mapped read for each individual was kept for further analysis. Expression levels for each individual in each contig were computed as RPKM values (i.e. the number of Reads per Kilobase per Millions mapped reads). We called genotypes and filtered out paralogs for each species individually using the read2snp software [47] (see S4 Text for details). Genotypes were called when the coverage was at least 10x and the posterior probability of the genotype higher than 0.95. Otherwise, the genotype of the individual was considered as missing data. Orthology between focal and outgroup individuals was determined by best reciprocal blast hit. Finally, we aligned orthologous contigs (focal and outgroup individuals) sequences using MACSE [74].
We scanned contig alignments in each focal species for polymorphic sites. We only considered gapless sites for which all focal individuals were genotyped. Only bi-allelic SNPs were considered. In the highly selfing T. monococcum, the deficit in heterozygous sites can lead to abnormal site frequency spectra that are difficult to analyse. We thus used an allele sampling procedure that effectively divides the number of chromosomes by two by merging together homologous chromosomes in each individual. For heterozygous sites, one allele was randomly chosen. For the mixed mating S. bicolor and S. pimpinellifolium, we used the full SFSs.
SNPs were polarized using parsimony by comparing alleles in focal individuals to orthologous positions in outgroups. For each polymorphic site, the ancestral allele was inferred to be the one identical to both outgroup species, while the other allele was inferred to be derived. All polarized SNPs are marked ancestral → derived for the remainder of the paper. A and T bases were grouped together as W (for weak) while G and C bases were grouped together as S (for strong). We thus classified mutations as W→S, S→W or neutral with respect to gBGC (S←→S or W←→W).
In each species, preferred (P) and un-preferred (U) codons were defined using the ΔRSCU method [49]. In each contig, we computed for each codon its RSCU value, or relative frequency (i.e. its frequency in a contig normalized by the frequency of its amino-acid in the same contig). Contigs were divided into eight groups of identical size based on their expression levels (RPKM values averaged over all individuals). For each codon, we compared its RSCU in the first (least expressed) and last (most expressed) class using a Mann-Whitney U test. Codons were annotated as preferred (or un-preferred) if their RSCU increased (or decreased) significantly with gene expression levels. All other codons were marked as non-significant. All synonymous SNPs for which an ancestral allele is unambiguously identified were annotated with regards to codon preference: mutations increasing codon preference (from un-preferred to either non-significant or preferred, or from non-significant to preferred) were annotated U→P while mutations decreasing codon preference (from preferred to either un-preferred or non-significant, or from non-significant to un-preferred) were annotated P→U. Mutations not affecting preference were considered as neutral with respect to SCU.
Using the three species alignments (Focal + two outgroups), we also counted and polarized substitutions specific to the focal species lineage. Divergent sites were determined as sites that were fixed in the focal population and different from both outgroups. Only sites identical in both outgroups were considered. As described above for SNPs, substitutions were classified as W→S, S→W or neutral, and U→P, P→U and neutral.
We performed a modified McDonald-Kreitman (MK) test [51], comparing W→S to S→W polymorphic and divergent sites on one hand (gBGC set) and U→P to P→U polymorphic and divergent sites on the other (SCU set). The underlying theory is detailed in S1 Text. For each category, the total number of synonymous polymorphic and divergent sites was computed following the criteria detailed above. We performed a Chi-squared test for each set. Significant tests indicate that sequences do not evolve only under mutation pressure: selection and/or gBGC must be at work. Furthermore, we computed for each set a neutrality [52] and a direction of selection [53] indices as follows:
NI=PWS/PSWDWS/DSW
DoS=DWSDWS+DSW−PWSPWS+PSW
Where PWS and PSW are the number of W→S and S→W SNPs and DWS and DSW the number of W→S and S→W substitutions respectively. Assuming constant mutational bias, NI values lower than 1 or positive DoS values indicate SCU and/or gBGC of similar or stronger intensity at the divergence than at the polymorphism level. Respectively, NI values higher than 1, or negative DoS values indicate stronger selection and/or gBGC at the polymorphism than at the divergence level (see S1 Text).
Because these statistics rely on polarized polymorphisms and substitutions, they are potentially sensitive to polarization errors, which could lead to spurious signature of selection/gBGC [38,44]. Importantly, we showed in S1 Text that the sign of both statistics is insensitive to polarization errors (as far as they are lower than 50%) and that polarization errors decrease the magnitude of the statistics, which makes our tests conservative to polarization errors.
To estimate gBGC and SCU we extended the method of Glémin et al. [38] as detailed in S2 Text. The rationale of the approach is to fit population genetic models to the three derived SFS including fixed mutations (W→S or U→P, S→W or P→U, and neutral). Parameters estimated are ancestral (B0 or S0) and recent (B1 or S1) gBGC or selection, mutational bias (λ), as well as other parameters (see S2 Text for details). We ran a series of nested models where B0 and B1 (or S0 and S1) are either fixed to zero or freely estimated, plus one model where they are set to be equal. Models were compared by the appropriate likelihood ratio tests (LRT). To jointly estimate gBGC and selection, we also extended the model by fitting nine SFS corresponding to the combination of the three basic SFS (e.g. W→S and P→U see S2.1 Table in S2 Text for the complete list). We tested all combinations of models where each parameter can be either null or freely estimated, so from the null neutral model, B0 = B1 = S0 = S1 = 0, to the model with the four parameters being freely estimated. As all models are not nested, we then chose the best model using the Akaike Information Criterion (AIC). When AICs were very close we chose the model with the lowest number of free parameters.
|
10.1371/journal.pgen.1007027 | MiR-125a Is a critical modulator for neutrophil development | MicroRNAs are universal post-transcriptional regulators in genomes. They have the ability of buffering gene expressional programs, contributing to robustness of biological systems and playing important roles in development, physiology and diseases. Here, we identified a microRNA, miR-125a, as a positive regulator of granulopoiesis. MiR125a knockout mice show reduced infiltration of neutrophils in the lung and alleviated tissue destruction after endotoxin challenge as a consequence of decreased neutrophil numbers. Furthermore, we demonstrated that this significant reduction of neutrophils was due to impaired development of granulocyte precursors to mature neutrophils in an intrinsic manner. We showed that Socs3, a critical repressor for granulopoiesis, was a target of miR-125a. Overall, our study revealed a new microRNA regulating granulocyte development and supported a model in which miR-125a acted as a fine-tuner of granulopoiesis.
| MicroRNAs are critical epigenetic modulators in development, physiology and disease processes. Many miRNAs are involved in immune cell development and function, like miR-150 for B cells, miR-181a for T cells. However, studies of miRNAs involvement in granulocyte development and function and related diseases are still limited. In this study, we developed engineering MiR125a knockout mice to study the function of miR-125a in vivo. We identified MiR125a knockout mice had decreased neutrophil numbers and reduced infiltration of neutrophils in the lung in LPS shock model. We deduced that this significant reduction of neutrophils was due to impaired development of granulocyte precursors to mature neutrophils in an intrinsic manner. Furthermore, we demonstrated that Socs3, a major repressor that negatively regulates granulocyte development, was a target of miR-125a. This finding not only reveals a new microRNA involving granulocyte development, but also provides insights into the new mechanism of miR-125a during action in endotoxemia.
| Neutrophils, also known as polymorphonuclear leukocytes (PMNs), are the most abundant granulocytes which play a crucial role in immune defense and inflammatory reaction. Given that the post-mitotic nature of mature neutrophils, they have short lives about only a few days [1] and need to be regenerated constantly through granulopoiesis, a part of hematopoiesis occurring in the bone marrow of adult mammals. During granulopoiesis, hematopoietic stem cells, at the top of the hematopoietic hierarchy, produce multilineage progenitors and precursors-common myeloid progenitors (CMP) and subsequently granulocyte-monocyte progenitors (GMP) which differentiate into mature granulocytes including eosinophils, basophils and neutrophils [2]. In general, granulopoiesis is in a basal physiological condition. However, emergency granulopoiesis can be rapidly induced to produce large number of neutrophils if severe systemic infection occurs [3].
Hematopoiesis is regulated by a group of cytokines. G-CSF is one of the major cytokine that regulates cell proliferation, differentiation and survival during the neutrophil lineage commitment [4, 5]. The receptor of G-CSF is mainly expressed in granulocytic progenitor cells and mature neutrophils [6].The binding of G-CSF to its receptor triggers receptor dimerization and tyrosine phosphorylation of JAK1, JAK2 and TYK2, which belong to the Janus family of protein tyrosine kinases (JAKs) [7]. These then phosphorylate residues in the cytosolic part of the G-CSF receptor and subsequently activate mitogen-activated protein (MAP) kinase like ERK pathway [8] and the signal transducers and activators of transcription (STATs) including STAT1 and STAT3 (4, 10). SOCS3, as the major repressor of G-CSF signaling, belongs to the suppressor of cytokine signaling (SOCS) family of proteins [9], which can be recruited to phosphorylated cytokine receptors and inhibit JAK catalytic activity and subsequently inhibit activation of ERK and STATs. Moreover, mice with Socs3 conditionally knocked out in hematopoietic cells [10, 11] develop neutrophilia and inflammatory pathologies.
MicroRNAs (miRs or miRNAs) are universal post-transcriptional regulators in animals and plants. Primary miRNAs are first transcribed by RNA polymerase II or III and are then excised to mature miRNAs (~22 nucleotide) that bind to 3’ untranslated regions (UTR) of their target mRNAs to silence gene expression [12]. More than 1000 miRNA genes have been identified in mammalian genomes [13]. And over 60% of protein-coding genes could be targeted by miRNAs according to computational prediction [14]. Due to their specific features, miRNAs have the ability of buffering gene expression programs and contributing to the robustness of biological systems [15]. Thus they play important regulatory roles in different biological processes. Decades of researches have shown that miRNAs involve in mammalian blood cell development and function [16]. For instance, miR-181a was found to modulate T cell selection [17] and miR-150 was identified as a controller of B cell development [18–20] as well as megakaryocytic versus erythrocytic lineage commitment [21]. In addition, miR-223, which was found highly expressed in neutrophils, played a role in regulating the proliferation of granulocyte progenitors and also mediated the inflammatory function of neutrophils [22, 23].
MiR-125a and miR-125b belong to the miR-125 family, which play a crucial role in many different cellular processes including cell differentiation, proliferation and apoptosis [24]. In order to systematically study the function of miR-125a in vivo, we developed miR-125a knockout mice. We examined the hematopoiesis of these mice and found fewer neutrophils in both bone marrow and peripheral blood in the absence of miR-125a. As a consequence of decreased number of neutrophils, MiR125a knockout mice were demonstrated with reduced infiltration of neutrophils in the lung and alleviated tissue destruction in an endotoxin challenge model. Furthermore, we found out that the reduction of neutrophils was due to impaired proliferation of immature granulocyte to mature neutrophils in an intrinsic manner. We showed that Socs3, a critical repressor for granulopoiesis, was a target of miR-125a. Together, these results suggest that miR-125a is an important regulator of basal granulopoiesis.
To fully understand the physiological role of miR-125a in vivo, we generated the MiR125a knockout mice as previously described [25]. These mice are fertile, born at the expected mendelian ratio, and not shown any abnormalities during their growth. However, we found that the white blood cell differential count revealed decreased numbers of neutrophils in MiR125a-/- mice (1.4 ± 0.3 x 106cells/mL versus 2.2 ± 0.4 x 106 cells/mL) (p<0.0001) while other mature hematopoietic lineage cells including other granulocytes (eosinophils and basophils) were normal (Table 1). Flow cytometry analyses of neutrophils in the bone marrow and peripheral blood confirm these results (Fig 1A). Next we did a bone marrow transfer assay to find out whether reduced granulopoiesis in MiR125a-/- mice are due to impaired cell-autonomous development or altered cytokine production from the bone marrow stromal cells. We found that decreased number of neutrophils reconstituted with MiR125a-/- bone marrow cells was both in MiR125a+/+ and MiR125a-/- recipients (Fig 1B). These results demonstrate that miR-125a contributes to reduced granulopoiesis in a cell-autonomous way. In addition, morphological analysis shows that neutrophils in MiR125a-/- mice are as mature as those in wild-type mice (Fig 1C). We then examined the expression of miR-125a in different stages of myeloid development and found that miR-125a was highly expressed in hematopoietic stem cells and decreased during maturation of myeloid progenitor cells, indicating that miR-125a may be involved in regulating granulocyte development (Fig 1D).
In order to examine whether miR-125a also plays a role in regulating neutrophil function, we tested the ability of activation, migration and killing pathogens between wild-type and MiR125a-/- neutrophils. Gene expression profiling data of bone marrow neutrophils stimulated with gram-negative bacterial lipopolysaccharide (LPS) showed that most of inflammatory factors and chemokines were induced equally either from MiR125a-/- or MiR125a+/+ mice (S1A Fig). Then in vitro transwell assay showed MiR125a-/- neutrophils had no detectable abnormality in fMLP or CXCL1 or CXCL2-dependent chemotaxis and migration (S1B Fig). We then used phorbol myristate acetate (PMA) or LPS to stimulate neutrophils and measured the production of reactive oxygen metabolites, which were important for neutrophils to kill pathogens. FACS analysis revealed no difference in the release of reactive oxygen species between wild-type and knock-out neutrophils (S1C Fig). Furthermore in vitro killing assay also demonstrated MiR125a-/- neutrophils had normal ability to clear bacteria and fungi (S1D Fig).
Neutrophils are known to be recruited at inflammatory tissue sites and play a critical role in sepsis and tissue damage [26]. We therefore performed experimental endotoxaemia by injecting a sub-lethal intraperitoneal dose of LPS to MiR125a-/- mice for 24 hours and measured neutrophil infiltration in the lungs by flow cytometry. Lungs of MiR125a-/- mice accumulated fewer neutrophils than those of MiR125a+/+ mice (Fig 2A). In addition, we checked the lung sections of MiR125a-/- and wild-type mice. Consistently with the FACS analysis, lungs of MiR125a-/- mice show less severe histopathological change, including congestion (hyperplasia of alveolar walls and alveolar collapse), edema (pulmonary interstitial edema), inflammation (neutrophil infiltration) and hemorrhage (engorgement of the capillaries) (Fig 2B). We also found MiR125a-/- mice had significantly reduced serum amounts of aspartate aminotransferase (ALT), blood urea nitrogen (BUN), creatine kinase (CK) and creatinine (CREA), which were indicators for organ damages (Fig 2C).
We next challenged both MiR125a-/- and wild-type mice with a lethal dose of LPS. We observed that MiR125a-/- mice were more resistant to lethal septic shock (Fig 2D). However, serum concentrations of inflammatory cytokine IL-6 and TNF-α during sepsis were similar (Fig 2E). In addition, normal Il6 and Tnfa mRNAs were expressed in peritoneal macrophages and bone marrow-derived macrophages after stimulation with LPS (S2 Fig). To further study whether there is any macrophage involvement, we depleted endogenous macrophages by using clodronate liposomes in wild-type mice and transplanted with MiR125a+/+ or MiR125a-/- bone marrow-derived macrophages. Then we administrated these mice with the lethal dose of LPS. Results did not show any difference in mortality (Fig 2F). These results implied that cytokine production induced by Toll-like receptors on macrophages did not contribute to resistance to LPS in MiR125a-/- mice. Thus resistance to a lethal dose of LPS and decreased neutrophils in the lungs with endotoxaemia in MiR125a-/- mice are likely caused by reduced granulopoiesis.
To study the mechanism of decreased neutrophil numbers in MiR125a-/- mice, we performed flow cytometry analysis on bone marrow cells in both wild type (WT) and knockout (KO) mice to examine whether the frequency of progenitor cells was disturbed. We found that the numbers of myeloid progenitors did not change (Fig 3A). We then performed colony forming assays on methylcellulose and analyzed them for myeloid precursors in complete medium. There is no significant difference in the frequency of myeloid precursors and numbers of granulocyte colonies (Fig 3B). For greater precision, we performed colony assays in the medium only containing variant concentrations of G-CSF and found that there was also no change in colony numbers (Fig 3C). However, we did notice that colonies from mutant mice were smaller and the cell number in one colony was less than those found in control mice (S3 Fig). Thus we sorted Lin-Sca1-c-Kit+CD34hiCD16/32hi GMPs by FACS and estimated their developmental capacity in a CFU assay. We also found the colony number did not change (Fig 3D) but the colony size and the cell number per colony from MiR125a-deficient GMPs decreased in the presence of G-CSF (Fig 3E–3G). Thus, it suggested that the development of granulocyte progenitors might be impaired in MiR125a-/- mice.
Since the number of granulocyte progenitors remained unchanged, it would appear that reduction of neutrophils only was due to increased cell death or impaired proliferation from granulocyte progenitors to mature neutrophils. To test the first possibility, we examined cell death rate of Ly6Ghi cells from bone marrow by staining them with Annexin V and propidium iodide. We found no difference in the rate of cell death between MiR125a-/- and wild-type mice (Fig 4A). We then performed in vivo BrdU-pulsing assays to analyze neutrophils generation in bone marrow (Fig 4B) and spleen (Fig 4C). Flow cytometry results showed that neutrophils from MiR125a-/- mice incorporated less BrdU than wild-type mice, indicating that cell proliferation had decreased during the differentiation of granulocyte progenitors into neutrophils.
It has been reported that CD11b+ Gr-1+ neutrophils in bone marrow are composed of three populations, including CD11bhi Gr-1hi cells (mature Neu), CD11blowGr-1hi cells (immature Neu) and CD11bintGr-1int cells (promyelocytes/myelocytes) [27–29]. According to this, we found the percentage of immature neutrophils was significantly lower in the bone marrow of MiR125a-deficient mice while the percentages of promyelocytes/myelocytes and mature neutrophils had no change (Fig 5A). In addition, we found BrdU-incorporating cells in the population of immature and mature neutrophils were significantly lower in MiR125a KO mice compared with WT controls while the population of promyelocytes/myelocytes had no change (Fig 5B). Because of post-mitotic nature of mature neutrophils, these BrdU-incorporating mature neutrophils mostly came from BrdU-incorporating immature neutrophils during their last division. Thus we deduced that the neutropenia of MiR125a-deficient mice could be due to reduced cell proliferation of CD11blowGr-1hi immature neutrophils.
As G-CSF is the major cytokine during granulocyte differentiation, we purified neutrophils from bone marrow cells and stimulated them with variant concentrations of G-CSF and counted the cell number after 24 hours. We found that the survival number of bone marrow neutrophils from wild-type mice increased substantially with increased G-CSF concentration while bone marrow neutrophils from MiR125a-/- mice did not increase in number (Fig 6A). We then analyzed apoptosis percentage and BrdU-incorporated cell ratios in response to G-CSF. In accordance with the observation in vivo, the amount of BrdU-incorporation was less in the absence of miR-125a (Fig 6B) while the apoptosis percentage has no change (Fig 6C). In addition, we found the mRNA levels of Gcsfr and several essential trancriptional factors for granulopoiesis like Pu.1, Gata-1, Cebpa, Cebpb and Cebpe did not change (S4 Fig). These results suggest that decreased cell proliferation in MiR125a-deficient mice might be due to impaired G-CSF signaling.
To investigate the molecular mechanism that contributes to impaired G-CSF-dependent proliferation, we examined activation of STAT1, ERK and STAT3 under the G-CSF signaling pathway (Fig 6D). In repeated experiments, we found that the ratio of phosphorylated STAT1, ERK and STAT3 vs. total STAT1, ERK and STAT3 was markedly weaker and less prolonged in different level in MiR125a-/- neutrophils in response to G-CSF (Fig 6E). This result indicates that the upstream in G-CSF signaling is impaired. However, we noticed that phospho-STAT3 was moderately enhanced while total STAT3 was much higher in MiR125a-/- bone marrow neutrophils. To determine whether the moderately enhanced p-STAT3 involves in mediating the decreased cell proliferation during maturation of MiR125a-/- GMPs, we cultured MiR125a-/- GMPs with G-CSF in the presence of STAT3 inhibitor S3I-201 or DMSO in CFU assays. Results show that inhibiting STAT3 cannot rescue the decelerated cell proliferation of MiR125a-/- GMP (S5A–S5C Fig). Therefore, according to these data, STAT3 is unlikely to mediate decreased granulocyte differentiation in MiR125a-/- mice.
Due to impaired G-CSF signaling pathway in MiR125a-deficient mice, we deduced that miR-125a might target a repressor in this signaling. SOCS3 is the major suppressor of G-CSF signaling and neutrophils differentiation [10, 30, 31]. Furthermore, we indeed detected higher SOCS3 protein expression levels in purified neutrophils lacking miR-125a compared to wild-type (Fig 7A). Thus we tested whether miR-125a directly targeted Socs3. We firstly predicted possible target sites in 3’UTR of Socs3 by using RNAhybrid and RNA22, and we found miR-125a has a potential binding site in the 3’UTR of Socs3 (Fig 7B). Then to confirm whether Socs3 is targeted by miR-125a, we cloned the full length of the 3’UTR of Socs3 onto a construct fused to the renilla reporter gene and mutated the predicted seed sequences. We co-transfected these plasmids with synthetic miR-125a oligonucleotide or negative control oligonucleotide in 293T cells respectively. The results indicated that miR-125a suppressed renilla luciferase activity but the mutants completely inhibited the suppression of the renilla luciferase activity (Fig 7C). These results demonstrate that miR-125a directly targets Socs3. But there remains a question whether Socs3 is a true target of miR-125a to regualte granulopoiesis. To address this issue, we did rescue experiments as follows. Firstly, we used shRNA to knock down Socs3 expression in MiR125a-deficient bone marrow cells and then did CFU assays in the presence of G-CSF. Results are shown that knockdown of Socs3 decrease Socs3 mRNA expression (Fig 7D). And the colony size (Fig 7E) and the cell number per colony (Fig 7G) both increase after Socs3 knockdown. However, the colony number does not change (Fig 7F). Next, we did a in vivo rescue assay by isolating short-term hematopoietic stem cells (ST-HSCs) from the bone marrow of MiR125a knockout mice, and we transduced these ST-HSCs with concentrated lentivirus of a Socs3 shRNA or a Ctrl shRNA, both of which contain GFP reporters. Then the transduced cells were collected and injected into the irradiated recipient wild-type mice. Six weeks later, the number of granulocytic progenitors and mature neutrophils was measured by FACS. Consistently with the results of in vitro CFU assay, we found that mice transduced with Socs3 shRNAs had significantly more GFP+ bone marrow neutrophils than those transduced with Ctrl shRNAs (Fig 8A). However, the number of GFP+ granulocytic progenitor CMPs and GMPs was not affected after Socs3 inhibition (Fig 8B). Furthermore in vivo BrdU-pulsing assays showed that BrdU incorporation of GFP+ CMPs and GFP+ GMPs did not change after Socs3 knockdown (Fig 8C). Taken together, both in vitro and in vivo experiments successfully rescue the decelerated neutrophil development caused by miR-125a deficiency and further confirm that Socs3 is the main factor of regulating neutrophil development from GMPs to mature neutrophils rather than earlier progenitors in MiR125a deficient hematopoiesis.
Previous studies have demonstrated that ectopic expression of miR-125a contributes to expansion of hematopoietic stem cell pool [32, 33]. However, we found an unexpected observation that the numbers of other mature hematopoietic lineage cells were not affected besides neutrophils in MiR125a knockout mice (Table 1). These inconsistent results might be explained by the reason that over-expression experiments may lead to gain-of-function phenotypes which cannot be found in knockout mice. Therefore, our results show miR-125a has an indispensable role in regulating neutrophil production.
Neutrophils as well as monocytes-macrophages are the first line of defense in response to systemic inflammation caused by pathogen infection or injury. Under endotoxin challenge, monocytes-macrophages release inflammatory factors such as TNF-α recruiting neutrophils in several organs to mediate tissue destruction [26]. Depletion of neutrophils protects the liver against injury from endotoxin [34]. Thus, like monocyte-macrophages, neutrophils also play a crucial role in endotoxemia. Our study reveals that MiR125a-/- mice have decreased numbers of neutrophils compared to wild-type mice. In addition, in our LPS shock model, we observed resistance to a lethal dose of LPS in MiR125a-/- mice but the concentration of TNF-α and IL-6 in the serum remained unchanged compared to control mice. Furthermore, macrophage reconstitution experiments indicated that macrophages did not contribute to resistance to LPS shock in MiR125a-/- mice. Therefore, we eliminated the possibility that MiR125a-/- macrophages exhibited less cytokine production in response to stimulation of Toll-like receptors. Importantly, we found less neutrophil infiltration in the lungs and alleviated multiple organ damage in MiR125a-/- mice after LPS challenge. As we also detected MiR125a-/- neutrophils were as mature and functional as those in wild-type mice. Therefore, we deduced that resistance to a lethal dose of LPS in MiR125a-/- mice was mainly due to reduced neutrophil numbers in granulopoiesis.
Granulopoiesis is part of hematopoiesis that maintains the peripheral neutrophil pool steady. In our study, we found MiR125a knockout mice showed neutropenia. We considered the main reason for the neutropenia was probably due to decreased cell proliferation from granulocyte progenitors to mature neutrophils in MiR125a-/- mice. The following are the main evidences demonstrated in this paper. Firstly, numbers of myeloid progenitors including CMPs and GMPs do not change according to FACS and CFU analyses, suggesting miR-125a may not regulate GMPs or even earlier progenitors. Secondly, the colony size is smaller and the cell number per colony is decreased from MiR125a-deficient GMPs, implying miR-125a involves in the respectively late stage of granulocyte development. Thirdly, immature and mature neutrophils are incorporated less BrdU in MiR125a KO mice while BrdU-incorporating promyelocytes/myelocytes have no change, meaning that miR-125a mediates cell proliferation during the differentiation from immature neutrophils to mature neutrophils. In addition, there is no difference in the rate of cell death between MiR125a-/- and wild-type mice by staining with Annexin V and propidium iodide, excluding the possibility that miR-125a-mediated cell death of neutrophils. Furthermore, other granulocytes (eosinophils and basophils) are not affected in MiR125a knockout mice (Table 1) also indicating that miR-125a is specific for regulating immature neutrophils rather than affecting earlier common granulocyte precursors.
To investigate the molecular mechanism of miR-125a in regulating neutrophil development, we checked the activation of G-CSF signaling pathway in wild-type and MiR125a deficient neutrophils. G-CSF is the major growth factor during each developmental stage of granulopoiesis [35]. STAT3, STAT1 and ERK are downstream transcription factors in G-CSF signaling [36]. From western blot analysis, we found MiR125a deficiency mainly caused impaired G-CSF signaling pathway through weakening the phosphorylation ratio of downstream transcription factors. But it made us a little bit confused. Although the phosphorylation ratio of STAT3 was reduced, phospho-STAT3 was moderately enhanced while total STAT3 was much higher in MiR125a-/- neutrophils. In order to solve this problem, we used STAT3 inhibitor S3I-201 in GMP CFU assays. Results demonstrated that inhibiting STAT3 cannot rescue the decelerated differentiation from MiR125a-/- GMP. Thus we deduce that the phenomenon of the enhanced total STAT3 might be through other unknown mechanisms and it is unlikely to mediate decreased granulocyte differentiation in MiR125a-/- mice. Owing to the weak G-CSF signaling in MiR125a-deficient mice, we deduce that miR-125a might target a repressor in this pathway. SOCS3 is the principal suppressor of G-CSF signaling. It can bind to pY729 of the G-CSF receptor and directly inhibit receptor binding to JAKs, thus repressing downstream signaling [30, 31, 37]. Particularly the mice in which Socs3 is conditionally knocked-out in bone marrow have increased neutrophil number and enhanced cellular responses to G-CSF including an increase in proliferative capacity [10, 11]. In our study, we actually identified Socs3 as a direct target of miR-125a. And the expression of Socs3 was indeed enhanced in MiR125a-/- neutrophils, weakening G-CSF signaling and eventually reducing neutrophils differentiation (S6 Fig). Furthermore, both in vivo and in vitro rescue experiments demonstrated that Socs3 indeed was the main target of miR-125a to regulate late stage development of neutrophils rather than earlier progenitors. Nevertheless, we deduce that miR-125a promotes granulopoiesis mainly by targeting suppressor Socs3.
MiRNAs are abundant regulators of transcriptional programs. They serve as fine-tuners of biological systems by giving signaling pathways a threshold to protect from unwanted or wrong signals and making signal output more precise and appropriate [38]. In many signaling pathways, the expression of miRNAs can be induced or repressed in response to outside stimuli and form feed-forward or feedback mechanisms with other signaling components [13]. However, basal expression of miRNAs is important for cell-type-specific gene expression through acting as switches like transcriptional factors during cell lineage determination [39]. Hematopoietic lineage differentiation is also switched by miRNAs. For example, miR-150 for B cell [18–20], megakaryocytic and erythrocytic lineage commitment [21], and miR-223 for granulocytic differentiation [22, 23]. In this paper, we proposed a model that miR-125a served as a positive regulator of physiological granulopoiesis by amplifying G-CSF signal strength and duration. In order to get a view of the regulation of miR-125a, we examined whether the expression of miR-125a was also affected by G-CSF signaling. However, we did not detect a significant change of the expression of miR-125a in granulocytes after G-CSF stimuli. As we found that miR-125a was decreased during maturation of granulocytes, we detected the expression of its target Socs3 which was also down-regulated and the expression of miR-125a and its target Socs3 exhibited a positive correlation in granulocyte development (S7 Fig). Although this kind of correlation between miRNA and its targets is against the repressive nature of miRNA-mediated gene regulation, bioinformative analysis shows that it is prevalent [40]. Because miRNAs often repress target genes through translational inhibition and have minor effects on target mRNA levels, so miRNAs and their targets levels are mainly controlled by upstream transcription factors [40]. According to this model, both Socs3 and miR-125a are down-regulated during granulopoiesis and down-regulated miR-125a leads to up-regulated Socs3 as a feed-forward signal. Thus this circuit can tune upstream signal fluctuation and eventually maintain SOCS3 protein homeostasis. As Socs3 is a critical negative regulator of granulopoiesis, its level in progenitors of granulocytes can affect the neutrophils differentiation and any significant change may lead to pathological consequences, namely neutrophilia and neutropenia. From this view, miR-125a modulation eventually provides a steady device to maintain differentiation and homeostasis of neutrophils rather than to simply repress the expression of Socs3.
In conclusion, we showed that miR-125a can positively regulate granulopoiesis. We demonstrated that miR-125a positively regulated G-CSF-dependent proliferation during the development of granulocytes by targeting Socs3. Our findings reveal a new microRNA involving granulocyte development and provide insights into the function of miR-125a during hematopoiesis. Future genetic studies will focus on how miR-125a is regulated during hematopoietic development.
MiR125a knockout mice were generated as previously described [25] and maintained under specific pathogen–free conditions at Institute of Health Sciences, Chinese Academy of Sciences animal breeding facility, according to institute guidelines. 8 to 12-week-old MiR125a knockout mice and their littermate controls were used for experiments. All experiments involving mice were in accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals of 1988, issued by the State Scientific and Technological Commission for China. And these experiments were approved by the Biomedical Research Ethics Committee of the Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences.
To analyze neutrophils, single cell suspensions of bone marrow or peripheral blood or spleen were stained with CD11b PerCP-Cyanine5.5 (eBioscience 45-0112-82) and Ly-6G-APC (eBioscience 17-5931-82). To measure neutrophil infiltration in the lung, lung tissues were cut into very small fragments and digested by collagenase and DNase I for 20 minutes at 37°C. Single cell suspensions were then stained with CD45-FITC (BD pharmingen, 553080), Ly-6G-APC and CD11b PerCP-Cy5.5. To detect the myeloid progenitor cells, bone marrow cells were pre-stained with biotin-conjugated mouse lineage panel (BD pharmingen, 559971), and then stained with streptavidin-V450 (BD horizon, 560797), Sca-1-PE-Cy7 (BD pharmingen, 558162), c-Kit-PE (BD pharmingen, 553355), CD34-FITC (BD pharmingen, 560238) and CD16/32-APC (eBioscience, 17-0161-82). Flow cytometry was conducted on a FACS Aria (BD Biosciences).
The recipient mice were fed with acidic (pH 2.6), antibiotic water for one week before irradiation and then were given 8.0 Gy irradiation by using a 137Cesium Gammacell source. 4 hours later, the mice were injected with 2x107 bone marrow cells from the donor mice via tail vein and then were kept on giving acidic antibiotic water for the rest of their lives.
To sort hematopoietic stem cells and progenitor cells, bone marrow cells were pre-enriched by depleting lineage positive cells (Stemcell, 19756). Hematopoietic stem cells were then sorted by Sca1+c-Kit+Lin-. CMPs were sorted by Sca1-c-kit+Lin-CD34+CD16/32-. GMPs were sorted by Sca1-c-kit+Lin-CD34+CD16/32+ and MEPs were sorted by Sca1-c-kit+Lin-CD34-CD16/32-. The purity of each cell population reached 95%. Neutrophils were isolated from bone marrow or peritoneal cavity by using the Neutrophil Isolation Kit (Miltenyi Biotec, 130-097-658). The purity of the isolated neutrophils was about 90%, as determined by flow cytometry.
Total RNA was isolated using TRIzol reagent (Life technologies). RNA quality was assessed with an Agilent 2100 Bioanalyzer (Agilent), and only samples with an RNA integrity number > 8 were used. Global mRNA expression in bone marrow neutrophils with or without LPS stimulation samples from and MiR125a+/+ and MiR125a-/- mice were assayed with the Affymetrix GeneChip Mouse Genome 430 2.0 Array. Data were deposited in GEO (GSE63739, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE63739) and analyzed with R and the associated BioConductor packages.
Isolated bone marrow neutrophils were resuspended in 0.1% BSA 1X Hanks balanced salt solution containing calcium and magnesium (Gibco) and plated in 3 μm Transwells (1X105 cells per Transwell, Corning) in the absence or presence of the indicated chemokine in the lower chamber (0.1 mM fMLP, Sigma; 250 ng/mL MIP-2, Peprotech; 1μg/mL KC, Peprotech). After incubation at 37°C for 3 hours, numbers of cell that migrated through transwell were counted.
Isolated bone marrow neutrophils were incubated in the presence of 1 μM dihydrorhodamine (Sigma) during stimulation with different concentrations of PMA (Sigma) for 15 minutes or LPS for 4 hours (Sigma). The oxidative burst of neutrophils was then analyzed by flow cytometry.
2x105 Candida albicans strain SC5314 or 1x107 Citrobacter rodentium were incubated with or without 5x105 bone marrow neutrophils in flat-bottom 96-well plates for 4 hours. Then all wells were treated with 0.02% triton-X 100 in PBS for 5 minutes. Surviving bacteria or fungi were incubated with 10μl MTT (5mg/mL) for 4 hours at 37°C then formazans were dissolved in DMSO and fluorescence was measured at 570 nm absorption wavelength.
Total RNA was isolated with TRIzol reagent (Life Technology). Expression of microRNAs in sorted cell populations was determined by quantitative PCR using the TaqMan MicroRNA Assay (Applied Biosystems). MicroRNA expression was normalized to snoRNA202. Socs3, Il6, Tnfa, Gcsfr, Pu.1, Gata-1, Cebpa, Cebpb and Cebpe mRNA expression levels were quantified by using SYBR PrimeScript reverse-transcription–PCR kit (Takara). Expression levels were normalized to endogenous expression of Gapdh.
Wild-type mice were first depleted of endogenous macrophages by pre-treatment with 100 μl clodronate liposome Clophosome-A (FormuMax Scientific) on Day1 and Day2. On Day3, these mice were transplanted with 1x107 MiR125a +/+ or MiR125a -/- bone marrow-derived macrophages. Macrophage depletion was detected by flow cytometry on Day3 and Day6 and the spleen and bone marrow macrophages were depleted >90%.
To count the number of GMPs, 5x104 bone marrow cells were cultured in methylcellulose (Mouse Methylcellulose Base Medium, R&D Technologies) added to various concentrations of recombinant murine G-CSF (R&D Technologies). After 10 days, colony numbers were counted. To quantify multi-potential progenitors and lineage-restricted progenitors, 2x104 bone marrow cells were plated in complete methylcellulose medium (Stemcell, 03434). After 12 days, colonies were counted and analyzed morphologically.
Bone marrow cells were cultured in 10% FBS RPMI 1640 medium (Life Technologies) for 48 hours, washed and stained for Ly-6G-APC and Annexin V FITC and PI (BD Biosciences) and analyzed by flow cytometry.
For the in vivo BrdU-incorporation experiment, mice aged 8–10 weeks were intraperitoneally injected with 200 μl of a 10mg/mL BrdU solution. After 3 days, mice were sacrificed and the spleen and bone marrow cells were harvested to detect BrdU-positive neutrophils. For in vitro BrdU-labeling of cells, bone marrow neutrophils were isolated and stimulated with 100 ng/ml G-CSF for 24 hours followed by incubating cells with 10 μM BrdU for 1 hour. BrdU-positive neutrophils were detected by using the BrdU flow kit from Pharmingen (BD Biosciences, 559619) with a FITC-labeled anti-BrdU antibody. Neutrophils were stained with CD11b-Percp Cy5.5 and Ly-6G-APC before fixation and permeabilization of the cells.
TNF-α and IL-6 in mice serum were detected by R&D Technologies duo set ELISA kit. For immunoblotting experiments, bone marrow cells or neutrophils were lysed with RIPA buffer and blotted with indicated antibodies. P-STAT3, STAT3, p-STAT1, STAT1, p-ERK, ERK and SOCS3 were all purchased from Cell Signaling Technology. GAPDH antibodies were obtained from Abcam.
To test whether miR-125a directly target the Socs3 3′ UTR, 293T cells were plated in 96-well plates and transfected with 10 ng wild-type or mutant Socs3 3′ UTR and the synthetic miR-125a oligonucleotide or negative control oligonucleotide by using Lipofectamine 2000 reagent (Invitrogen). Firefly and renilla luciferase activities were determined after 24 hours using the Dual-Luciferase Reporter Assay System (Promega). The values were normalized to firefly luciferase.
To generate a retrovirus construct, MSCV-LTR miR30-PIG (LMP) plasmids were cloned into Socs3-specific hairpin RNA. The target sequence is as follows: CGC GAG TAC CAG CTG GTG GTG A. Plate-E cells were transfected with 30ug LMP shRNA for a dish and retroviruses were harvested from culture supernatant after 48 hours. Mice bone marrow cells were depleted lineage positive cells by magnetic beads, stimulated with G-CSF overnight, then infected with recombinant retrovirus. 48 hours later, green fluorescent protein expressing GMPs were sorted for CFU assays.
To generate a lentivirus construct, pLVX-shRNA2 plasmids were cloned into Socs3-specific hairpin RNA. The target sequence is the same as above. 293T cells were transfected with 15ug pLVX-shRNA2 together with 8ug pMD2.G and 15ug psPAX2 plasmids for one dish. Lentivirus were harvested and concentrated from culture supernatant after 72 hours. Bone marrow cells of MiR125a ko mice were depleted lineage positive cells by magnetic beads, and short-term hematopoietic stem cells (ST-HSC) were sorted by sca-1+c-kit+CD135-CD34+ and resuspended at 1 x 104 in 75 uL StemSpan (StemCell Technologies), supplemented with 50 ng/ml SCF (Peprotech) in a round-bottomed well of a 96-well plate. 2.5 x 107 units of lentivirus were added into each well after 2 hours, predetermined to give about 20% transduction efficiency by measuring of GFP positive cells in pilot experiments. Then plates were spun at 900g for 90 min, and cultured at 37°C with 5% CO2-in-air. Cells were collected and washed after 4.5 hours and per 1.5 x 10^4 ST-HSCs were resuspended in 250 ul PBS which was then injected into each irradiated recipient wild-type mouse.
|
10.1371/journal.pbio.1000184 | Glia and Muscle Sculpt Neuromuscular Arbors by Engulfing Destabilized Synaptic Boutons and Shed Presynaptic Debris | Synapse remodeling is an extremely dynamic process, often regulated by neural activity. Here we show during activity-dependent synaptic growth at the Drosophila NMJ many immature synaptic boutons fail to form stable postsynaptic contacts, are selectively shed from the parent arbor, and degenerate or disappear from the neuromuscular junction (NMJ). Surprisingly, we also observe the widespread appearance of presynaptically derived “debris” during normal synaptic growth. The shedding of both immature boutons and presynaptic debris is enhanced by high-frequency stimulation of motorneurons, indicating that their formation is modulated by neural activity. Interestingly, we find that glia dynamically invade the NMJ and, working together with muscle cells, phagocytose shed presynaptic material. Suppressing engulfment activity in glia or muscle by disrupting the Draper/Ced-6 pathway results in a dramatic accumulation of presynaptic debris, and synaptic growth in turn is severely compromised. Thus actively growing NMJ arbors appear to constitutively generate an excessive number of immature boutons, eliminate those that are not stabilized through a shedding process, and normal synaptic expansion requires the continuous clearance of this material by both glia and muscle cells.
| The synapse is the fundamental unit of communication between neurons and their target cells. As the nervous system matures, synapses often need to be added, removed, or otherwise remodeled to accommodate the changing needs of the circuit. Such changes are often regulated by the activity of the circuit and are thought to entail the extension or retraction of cellular processes to form or break synaptic connections. We have explored the precise nature of new synapse formation during development of the Drosophila larval neuromuscular junction (NMJ). We find that growing synapses are actually quite wasteful and shed significant amounts of presynaptic membranes and a subset of immature (nonfunctional) synapses. The shedding of this presynaptic material is enhanced by stimulating the activity of the neuron, suggesting that its formation is dependent upon NMJ activity. Surprisingly, we find presynaptic membranes are efficiently removed from the NMJ by two surrounding cell types: glia cells (a neuronal ‘support cell’), which invade the NMJ, and the postsynaptic muscle cell itself. Blocking the ability of these cells to ingest shed presynaptic membranes dramatically reduces new synapse growth, suggesting that the shed presynaptic material is inhibitory to new synapse addition. Therefore, our data demonstrate that actively growing synapses constantly shed membrane material, that glia and muscles work to rapidly clear this from the NMJ, and that the combined efforts of glia and muscles are critical for the proper addition of new synapses to neural circuits.
| The wiring of the nervous system, from initial axon sprouting to the formation of specific synaptic connections, represents one of the most dramatic and precise examples of directed cellular outgrowth. Developing axons navigate sometimes tortuous routes as they seek out the appropriate target cells. Once in their target area, interactions between axons and their potential targets are extremely dynamic, attempts are made to identify appropriate postsynaptic partners, and initial synaptic contacts are established [1],[2],[and reviewed in 3]. A next critical step in the formation of functional neural circuits is the remodeling of initial patterns of connectivity. To facilitate the elaboration and refinement of developing neural circuits synaptic partners often remain highly responsive to their environment and add or eliminate synaptic connections [4],[5], frequently in an activity-dependent fashion, presumably to fine-tune connectivity to specific activity patterns.
After the axons have found their partners, two distinct mechanisms can drive the developmental reorganization of synaptic connectivity: intercellular competition between cells for common targets (reviewed in [4],[5]), and the addition/elimination of synapses within a single arbor in response to the physiological demands of the signaling unit [6]–[8]. The former mechanism dictates the circuit “wiring diagram” by defining precisely which subsets of cells will communicate through synaptic contacts; while the latter, in contrast, modulates the strength of connectivity between specific pre- and postsynaptic cells after circuits are assembled.
Early in nervous system development an excessive number of axonal projections and synaptic connections are initially established. What then ensues is cell–cell competition between neurons innervating the same target for limiting target-derived cues or sites of innervation during synaptogenesis. Appropriate synaptic contacts are then strengthened and exuberant processes are destabilized and eliminated through activity-dependent mechanisms [5],[9]. For example, at the mammalian neuromuscular junction (NMJ) muscles are initially innervated by more than one motor input. However, through a process of intercellular competition for motor endplates, all but one motor input are eliminated, with the “losers” retracting wholesale from the motor endplate [2]. Likewise, at the retinotectal projection in frogs, retinal axons initially establish a rough topographic map with substantial overlap between branches. However, these local synaptic terminals ultimately compete for target space and through activity-dependent modulation of synapse stabilization the spatial map of synaptic inputs is ultimately refined to a highly selective subset of inputs [10].
In the intercellular competition model the elimination of exuberant inputs (the “losers”) can entail large-scale elimination of axon branches, and perhaps smaller scale pruning of individual synaptic contacts. During axon and synaptic pruning in mammals and Drosophila entire axon branches are destabilized, degenerate, and are then cleared from the central nervous system by engulfing cell types (reviewed in [5]). Similarly, recent work has shown that excessive motorneuron inputs at the mammalian NMJ also become destabilized, detach from the motor endplate, and undergo axosome shedding. In this process local Schwann cells processively engulf motorneuron terminals in a distal to proximal direction and constitute the force that drives retraction bulbs toward the parent arbor during input elimination [11]. Ultimately, this mechanism results in a reduction of the total number of cells supplying synaptic input to the target cell.
In the second and mechanistically distinct mode of synapse remodeling, individual synaptic contacts are added or removed from a single arbor to strengthen or weaken synaptic input to the target cell. Such changes are generally elicited by changes in the target size or neural activity. For example, Drosophila motorneurons have established synaptic contacts with specific embryonic muscle cells by the end of embryogenesis [12]. At subsequent larval stages individual arbors, along with the target muscle itself, grow in size ∼100-fold [7],[8]. This coordinate increase in muscle size and synaptic contacts at motorneuron terminals serves to increase synaptic input from the motorneuron as needed to drive activation of the expanding muscle fiber. Similar mechanisms appear in place to modulate the balance of neural input versus target cell size in mammals: at the mammalian adult bulbocavernous muscle, testosterone manipulation lead to increases or decreases in muscle size, and these changes were accompanied by respective expansion or shrinkage of the postsynaptic region of the NMJ, respectively [6].
Here we explore the in vivo dynamics of synaptic expansion in motorneuron arbors at the Drosophila NMJ. We show in live preparations that the addition of new synapses during normal synaptic growth entails a large amount of shedding of presynaptic membranes in the form of small debris and a subpopulation of undifferentiated synaptic boutons (ghost boutons) that failed to mature. This process is distinct from intercellular competition, as none of the motorneuron terminals are eliminated. Rather, this mechanism appears to regulate the final size of the terminal arbor. We find that the formation of presynaptic debris (this report) and ghost boutons [13] are modulated by neural activity, as acute stimulation of motor inputs leads to increased appearance of these structures. Intriguingly, presynaptic debris and the subpopulation of ghost boutons that become detached from the parent arbor appear to be actively cleared from the NMJ as they disappear over developmental time. We show that glia dynamically invade the NMJ and phagocytose presynaptically shed debris, and that ghost boutons are engulfed or degraded primarily by muscle cells. Loss of phagocytic function in glia or muscle cells through manipulating the Draper signaling pathway (a key engulfment signaling pathway) results in an accumulation of presynaptic debris or ghost boutons at the NMJ and a severe reduction in NMJ expansion, indicating that continuous clearance of shed presynaptic debris and/or ghost boutons is essential for normal synaptic growth. Thus glia and muscles work together to sculpt connectivity at developing NMJ arbors, clearing multiple types of shed presynaptic structures that are inhibitory to the formation of new synaptic boutons.
In insects, α-HRP antibodies cross-react with neuron-specific membrane antigens [14] likely by binding to carbohydrate moieties present in a number of neuronal membrane proteins, including the cell adhesion molecules Fasciclin (Fas) I and II [15]. Consistently, at the Drosophila larval NMJ α-HRP antibodies labeled the entire presynaptic arbor (Figure 1Ai). However, we also noticed the presence of HRP-immunoreactive puncta at the postsynaptic junctional region, beyond the presynaptic membrane (Figures 1Ai, 1Aii, arrows). These puncta also labeled with antibodies to FasII and did not appear to be connected to the presynaptic arbor (Figures 1Aiii, Aiv). We wondered whether the HRP and FasII-positive postsynaptic staining might correspond to postsynaptic muscle structures, or whether the puncta might be derived from the presynaptic arbor. To distinguish between these possibilities, we expressed a membrane tethered green fluorescent protein (GFP; UAS-mCD8-GFP) in motorneurons using the motoneuron-specific Gal4 driver OK6-Gal4 [16]. We found that the postsynaptic HRP puncta were exactly colocalized with the presynaptically derived GFP signal (Figure 1D, arrow), suggesting that the HRP puncta are likely membrane fragments derived from presynaptic boutons. The presynaptically derived mCD8-GFP puncta were also observed by imaging through the cuticle of intact (undissected) larvae using a spinning disk confocal microscope, indicating that they are naturally occurring and not an artifact of the dissection or sample preparation (Figure 1E, arrows).
The nature of the presynaptically derived puncta was examined using a number of synaptic markers. Cysteine string protein (CSP) and Synapsin (Syn) are presynaptic vesicle proteins that associate with the readily releasable and the reserve pool of synaptic vesicles respectively [17],[18]. We found that the postsynaptic HRP puncta colocalized with CSP (Figure 1B, arrows and inset), but not with Syn immunoreactivity (Figure 1C). The presence of CSP in the HRP puncta further validates the idea that these puncta are presynaptic in origin. Labeling with antibodies against the active zone marker Bruchpilot (Brp) did not reveal immunoreactivity at the postsynaptic HRP-positive puncta (unpublished data). Together these results suggest that during NMJ development the motorneuron sheds membrane fragments (here referred to as presynaptic debris). Based on the presence of CSP but not Syn, the absence of Brp and the presence of FasII, we propose that presynaptic debris might arise from the perisynaptic bouton region.
Studies in many systems have suggested that the state of a mature synapse is the result of a dynamic equilibrium between growth and retraction [19]. Therefore, to determine what conditions lead to the shedding of presynaptic debris, we attempted to perturb this equilibrium by inducing activity-dependent synaptic growth [13]. Previous studies at the larval NMJ show that an acute increase in activity induces a de novo formation of new synaptic boutons. In particular, spaced cycles of stimulation, consisting of either K+-induced depolarization, high frequency nerve stimulation, or light gating of neuronally expressed channelrhodopsin-2 (ChR2), induce rapid structural changes at the NMJ. These changes include an increase in the number and length of dynamic presynaptic filopodia (synaptopods) and the number of undifferentiated boutons (ghost boutons) containing synaptic vesicles but lacking active zones and postsynaptic proteins [13]. Imaging of intact larvae also showed that synaptopods and ghost boutons were naturally occurring structures observed even in unstimulated preparations albeit at low frequency [13].
In our experiments we expressed ChR2 in motorneurons using OK6-Gal4 and stimulated the motorneurons of intact larvae with 5 cycles of spaced light stimulation as previously described [13]. Body wall muscles were then dissected either 30 min or 18 h after the stimulation was complete and labeled with anti-HRP. As a control, we used unstimulated larvae expressing ChR2 in motorneurons but not subjected to the light pulses. Notably, we found that the total area occupied by particles of presynaptic debris around the NMJ was significantly increased 30 min after the end of spaced stimulation (Figure 1F–1I), indicating that acute stimulation of neural activity resulted in an increase in presynaptic debris at the NMJ. Allowing NMJs to recover for 18 h after stimulation resulted in debris returning to wild-type levels (Figure 1I), suggesting the presence of an active mechanism to eliminate presynaptic debris from the NMJ. We conclude that presynaptic debris are normally present at the NMJ and conditions that lead to synaptic growth result in a transient increase in the amount of presynaptic debris, thus shedding of debris is associated with NMJ growth.
We also conducted time-lapse imaging of identified NMJs from live intact larvae expressing ChR2 in motorneurons using C380-Gal4 [20]. These larvae also contained fluorescent markers that allowed us to simultaneously image the pre- and the postsynaptic compartment. In particular, these larvae expressed UAS-mRFP in motorneurons to visualize the presynaptic NMJ arbor and mCD8-GFP::Sh in muscles using the myosin heavy chain (MHC) promoter [21] to visualize the postsynaptic NMJ region. In the MHC-mCD8-GFP::Sh transgene, the GFP is fused to the last ∼150 C-terminal amino acids of the Shaker K+ channel isoform containing a Discs-Large (DLG) PDZ binding site, and thus it is targeted to the postsynaptic region allowing its visualization in vivo [21]. These larvae were subjected to spaced stimulation with light as above, and the same NMJ imaged for 5–15 min at different intervals. Between imaging intervals larvae were returned to the food. As previously reported [13], we found that ghost boutons were present and some of these became stabilized and recruited postsynaptic label. However, we also observed that many of these ghost boutons did not recruit postsynaptic label and disappeared over time (Figure 2A, arrow and inset in right panel).
The presence of presynaptic debris in normal animals, the enhancement of presynaptic debris deposition upon spaced stimulation, and the elimination of some of the newly generated ghost boutons after spaced stimulation suggest that NMJ development involves the continuous shedding of certain presynaptic membrane compartments. Furthermore, the lack of accumulation of these components over developmental time, suggest that they may be actively removed from the NMJ.
To determine if presynaptic debris might originate from the breakdown of ghost boutons that failed to become stabilized and disappeared, we followed the fate of ghost boutons that became detached from the presynaptic arbor and presynaptic debris. In these experiments, identified NMJs from larvae expressing ChR2 and mCD8-GFP in motorneurons were repeatedly imaged through the cuticle as above following spaced stimulation. We found that on several occasions, as ghost boutons detached, debris appeared in the position of the ghost bouton stalk and around the ghost bouton, suggesting that ghost boutons can degenerate directly into presynaptic debris (e.g., Figure 2B and 2C; ghost boutons are marked by white arrows and debris by black arrowheads). In some samples we were able to directly image the disintegration of ghost boutons into smaller fragments (Video S1). However, in other cases, stalks simply disappeared without leaving debris, and detached ghost boutons became smaller and vanished from the NMJ without leaving any obvious debris (Figure 2D and 2E, white arrows). Interestingly, not all presynaptic debris appeared to derive from ghost boutons and their stalks—we also observed the appearance and disappearance of presynaptic debris at NMJ regions devoid of ghost boutons (Figure 2E, black and pink arrowheads), suggesting that presynaptic debris can be generated independently from ghost boutons. In summary, presynaptic debris can apparently arise directly from the breakdown of ghost boutons, or, alternatively may be derived directly from the presynaptic arbor without participation of ghost boutons.
The very low levels of presynaptic debris and ghost boutons observed here in unstimulated larvae and the removal of the extra debris formed upon stimulation, suggested that as NMJs develop, presynaptic membrane debris and disconnected ghost boutons are actively cleared from the NMJ. Signal transduction mechanisms mediating the engulfment of neuronal debris are beginning to be elucidated [22]. Most prominent, the engulfment receptor Draper (Drpr; Ced-1 in Caenorhabditis elegans) is involved in the engulfment of neuronal cell corpses during programmed cell death, the pruning of mushroom body neuron arbors during fly metamorphosis, and in the phagocytosis of injured axons in the fly olfactory system [23]–[26]. We therefore used draper mutants as a tool to block the activity of local engulfing cell types and assayed the effects of loss of Draper function on clearance of shed presynaptic debris and disconnected ghost boutons from the larval NMJ. Strikingly, we found that draper mutant NMJs were highly abnormal, with the presence of unusually large and irregularly shaped boutons and with a marked reduction in the number of glutamatergic type Ib boutons (Figure 3A, 3B, and 3F). Close examination of the NMJs in these mutants revealed that there was also a dramatic increase in the amount of presynaptic debris (Figure 3C–3E, arrows, 3H) and number of ghost boutons (Figure 3E, arrowheads, 3G). Interestingly, we also found that third instar draper mutant larvae had reduced larval motility in behavioral assays (Figure S1), suggesting that the accumulation of presynaptically shed material may adversely affect neuromuscular function. Thus, in the absence of Draper function NMJs develop abnormally and presynaptic debris and ghost boutons accumulate at high levels. These observations suggest that an engulfing cell type might invade, or be a resident component of, the NMJ, and phagocytose shed presynaptic material.
In the fly nervous system Draper is expressed in glia where it has crucial roles in engulfment activity [23]–[26]. To determine if Draper was also present in glial cells at the NMJ, we used α-Draper antibodies [24]. Surprisingly, in addition to its localization in peripheral glia that wrap around motor nerves (Figure 4A), we found that Draper immunoreactivity was present at the postsynaptic region of every synaptic bouton in colocalization with the Drosophila PSD-95 homolog DLG (Figure 4C). This immunoreactivity was specific to Draper, as it was virtually eliminated in draper null mutants (Figure 4B and 4D).
The above observation was surprising, since in contrast to vertebrate NMJs, where terminal Schwann cells completely cover the NMJ [27], at the glutamatergic Drosophila larval NMJ terminal glia have not been reported to cap the synaptic arbor [28],[29]. Instead, NMJ arbors are buried within the muscle surface, which wraps around the boutons forming a layered system of membranes, the subsynaptic reticulum (SSR) [30],[31]. Previous studies have suggested that at the larval NMJ peripheral glia ensheath the segmental nerve, but for the most part, their membranes terminate at the axon branch point or at the first synaptic bouton closest to the branch point [29]. The presence of Draper surrounding the entire NMJ led us to reexamine the organization of glial cell membranes at the NMJ and their relationship to synaptic boutons. For these experiments we expressed a membrane tethered GFP (mCD8-GFP) in peripheral glia, using Gliotactin-Gal4 (Gli-Gal4), and HRP-labeled NMJs from abdominal segments 3 and 4 were systematically examined in fixed preparations. We found that in the majority of cases glial membranes deeply invaded the NMJ (Figure 5), presumably invading the space between the presynaptic motorneuron terminal and the SSR. Some NMJs (2%–40% on average depending on the specific NMJ), particularly those innervating dorsal muscles, appeared completely covered by glial membranes (Figure 5A and 5E; covered NMJs). A majority (80%–100%) of NMJs were associated with lamellipodia-like glial extensions that contacted several boutons (Figure 5A–5C, and 5E). Glia also extended thin filopodia-like processes that contacted synaptic boutons at the same NMJ branch or that exited the branch and interacted with synaptic boutons from a different NMJ branch (Figure 5Av and 5Bv). Glial membrane processes were also observed in association with muscle regions around the NMJ that were completely devoid of synaptic boutons (Figure 5Aiv and 5Civ–v). A small percentage (∼7%) of glial extensions had an elliptical appearance and terminated in bulbous structures of variable size (Figure 5Div–v and 5E). These bulbous structures sometimes surrounded a synaptic bouton (Figure 5Dv, arrowhead). In some NMJs (11%–33%) glial membranes did not invade the NMJ and muscle, and terminated at the nerve branch-point before synaptic boutons (Figure 5Bi–iii; blunt ended).
Interestingly, the pattern of glial extensions was not stereotypic and showed a high degree of variability among segments and identified muscles from different individuals. This observation suggests that the glial processes are likely to extend and retract in a dynamic fashion. This possibility was examined by live imaging preparations expressing mCD8-GFP in peripheral glia with Gliotactin-Gal4. We found that glial processes were indeed at the NMJ, and extended or retracted within a period of minutes (Video S2). These observations indicate that glial cells at the larval NMJ have previously unappreciated dynamics, and that they establish multiple transient associations with the NMJ. However, our studies of Draper localization at the NMJ demonstrated that Draper is present at every NMJ and surrounding each synaptic bouton (Figure 4C). Thus, the extension of glial membranes is unlikely to account for Draper localization at the entire NMJ, raising the possibility that muscles might also contribute to NMJ Draper localization.
In draper mutants, there were some changes in the distribution and frequency of glial extensions. Glial extensions that covered the entire NMJ (covered NMJs) were absent or drastically reduced in frequency, and there were also changes in the distribution and frequency of gliobulbs (Figure S2). In contrast, there was a strong increase in the frequency of blunted projections (i.e., those that end close to the nerve branch point and do not interact with synaptic boutons), and a normal level of lamellipodia-like extensions). These observations suggest that in the absence of Draper function some glial membranes do not extend properly into the NMJ. Thus positive signaling through Draper, perhaps in response to cues released by presynaptic debris, may directly regulate a subset of glial membrane movements at the NMJ.
To address the possibility that Draper might function both in glia and muscle to sculpt the NMJ we selectively expressed a Draper-RNAi designed to knockdown all Draper isoforms in glia or muscles using cell-specific Gal4 strains. RNAi knockdown of Draper in either muscle or glia resulted in a reduction in the number of synaptic boutons, which was not significantly different from the draper null mutant (Figure 6E). This indicates that the removal of Draper from either cell type is sufficient to interfere with NMJ growth. Remarkably, however, downregulating Draper in muscle versus glia had a different consequence for the deposition of presynaptic debris and the appearance of detached ghost boutons. RNAi knockdown of Draper in glia resulted in an increase in presynaptic debris to an extent similar to the draper null mutant (Figure 6C and 6G). However, no significant increase in the number of detached ghost boutons was observed (Figure 6F). If glial extensions are primarily involved in engulfing presynaptic debris, we predicted that we should find HRP positive debris within the glial extensions. We found that this was indeed the case. We found several instances in which glial terminals formed bulb-like structures that contained anti-HRP immunoreactive puncta within (Figure 6D, arrowheads).
In contrast, downregulating Draper in muscle resulted in an increase in the number of ghost boutons (Figure 6B and 6F), but the level of presynaptic debris was similar to wild type (Figure 6B and 6G). Expressing Draper RNAi in motorneurons did not affect the number of boutons, ghost boutons, or the levels of presynaptic debris (Figure 6E–6G). These results support the idea that Draper functions both in muscle and glia, and that the function of Draper in each cell has some degree of specialization. While glial Draper appears to function in removing presynaptic debris, muscle Draper appears to remove ghost boutons fated for elimination. Importantly, these observations also provide the first evidence that muscle cells fulfill a phagocytic function at the NMJ.
Previous studies have shown that the PTB-domain protein dCed-6 functions downstream of Draper [23]. Therefore, we used RNAi knockdown of dCed-6 in muscle or glia as a second approach to blocking glial and muscle engulfment activity. As in draper mutants, downregulating dCed-6 in either muscle or peripheral glia resulted in significant decrease in the number of synaptic boutons (Figure 7A–7D). In contrast, no effect was observed when dCed-6-RNAi was expressed in motorneurons (Figure 7D). Similar to Draper RNAi knockdown, expressing dCed-6-RNAi in muscles or glia had differential consequences for the appearance of presynaptic debris versus ghost boutons. Decreased levels of dCed-6 in muscles led to an increase in the number of ghost boutons, but had no influence in the deposition of presynaptic debris (Figure 7B, 7E, and 7F). Downregulating dCed-6 in glia, on the other hand, led to a significant increase in presynaptic debris deposition, but the number of ghost boutons remained unaltered (Figure 7C, 7E, and 7F). These results are consistent with the notion that dCed-6 functions downstream of Draper during the development of the NMJ. Further, they support the model that both muscle and glia contribute differentially to the clearance of debris versus ghost boutons at the NMJ.
The draper gene gives rise to three different Draper isoforms, each with a unique combination of intracellular and extracellular domains (Figure 8A). Draper-I bears 15 extracellular EGF repeats, whereas Draper-II and -III only contain five [24]. In their intracellular domains, all isoforms contain a potential dCed-6 binding site (NPXY), but the Shark binding site is only present in Draper-I and -II. To determine which of the isoforms might be involved in NMJ development, we first carried out reverse-transcription PCR (RT-PCR) of body wall muscles. Interestingly, we found that Draper-I and III, but not Draper-II were expressed at the neuromuscular system (Figure 8A and 8B). Therefore, we carried out rescue experiments by expressing Draper-I or -III in muscles or glia in a draper null mutant background.
None of the Draper isoforms completely rescued the decrease in bouton number observed in the drpr null (Figure 8C). This is consistent with the observations with cell-specific Draper-RNAi expression, showing that Draper functions both in muscle and glia, and that downregulating Draper in either cell is sufficient to decrease bouton number to an extent similar to the draper null mutant alone. In the case of ghost boutons, expressing Draper-I in glia or Draper-III in muscle completely rescued the mutant phenotype (Figure 8D). However, expressing Draper III in glia or Draper I in muscle also resulted in substantial but incomplete rescue. For the deposition of presynaptic debris, only expressing Drpr-I in glia completely rescued the phenotype, but partial rescue was also observed when Drpr-III was expressed in muscle (Figure 8E). These data provide conclusive evidence that the phenotypes we observe in draper null mutant NMJs indeed map to the draper gene, and that the phenotypes we observe in draper mutants can be significantly rescued by resupplying Draper in glia or muscle cells (Figure 8F). The incomplete rescue of some of the phenotypes by specific isoforms might represent redundant functions by these isoforms, a requirement for multiple isoforms for complete rescue, or simply result from increased Draper expression in transgenic animals.
Here we have studied the in vivo dynamics of synaptic connectivity between single motor inputs and their target muscle cells. We describe a novel event that occurs during the remodeling of single synaptic arbors during development or activity-induced plasticity: the shedding of presynaptic debris and aborted synaptic boutons that failed to stabilize. This process differs from developmental pruning or intercellular competition during synapse elimination, as in those cases entire nerve terminals are eliminated, thereby changing the wiring diagram of a circuit. Rather, we show that the expansion of an already established synaptic input involves significant production of presynaptic membrane debris and the detachment of undifferentiated synaptic boutons destined for elimination from the main arbor. Both glial and muscle cells act in concert to clear the developing NMJ of this shed presynaptic material, and the suppression of engulfing activity in glial or muscle cells leads to highly disrupted NMJ growth. We propose that this novel mechanism might serve to rapidly adapt the size of a growing synaptic terminal to the changing demands of the target cell by shifting the equilibrium between synapse stabilization and synapse destabilization.
During larval development, the NMJ is continuously increasing the size and number of synaptic boutons. This expansion serves as a compensatory mechanism to preserve synaptic strength, despite the massive growth of muscle cells [32]. Our studies provide evidence that normal NMJ growth includes the constitutive shedding of presynaptic membranes. The presynaptic origin of HRP-positive debris was demonstrated by labeling motorneuron membranes with genetically encoded mCD8-GFP, which consistently labeled the debris, by the observation that in some cases ghost boutons that detached from the main arbor disintegrated into debris, and by the finding that the debris also contained presynaptic proteins, such as CSP. Thus, synaptic debris might contain synaptic vesicles or vesicle membrane remnants that failed to be recycled. Interestingly, Brp, an active zone marker [33], was absent from the debris. This absence might reflect its degradation, or alternatively, the derivation of presynaptic debris from periactive regions of the NMJ. Indeed, FasII, which is localized at periactive zones [34] was also present in presynaptic debris.
Acute spaced stimulation of the larval NMJ leads to the formation of dynamically extending and retracting synaptopods, and to the appearance of ghost boutons [13]. While some ghost boutons differentiate by acquiring active zones and postsynaptic proteins [13], here we found that others lost their connection with the presynaptic arbor and were specifically removed. What happens to ghost boutons that detach from the main arbor? In most cases we found that detached ghost boutons rapidly disappeared from the NMJ. On the basis of our finding that suppressing engulfing action in muscle leads to the accumulation of ghost boutons, we propose that these are engulfed directly by muscle cells (Figure 8F).
In other cases we found that ghost boutons, along with the stalk by which they were initially attached to the main arbor, would degenerate into smaller fragments resembling presynaptic debris. Thus at some level, ghost boutons also appear to be able to disintegrate into presynaptic debris. That presynaptic debris and ghost boutons are unique cellular remnants is also argued by the fact that they are differentially engulfed by glia and muscle cells, respectively (Figure 8F). Nevertheless, the detachment and elimination of ghost boutons we describe represents a simple and newly defined mechanism for the removal of excessive synapses formed by individual innervating motorneurons. This process might also serve as a mechanism for rapid stabilization of new synaptic boutons during, for example, periods of increased synaptic or locomotor activity (see below) [13],[35],[36].
The functional significance of shedding presynaptic debris remains unclear. Manipulations that promote rapid synaptic growth, such as acute spaced stimulation, lead to an increase in presynaptic debris suggesting that its production is associated with synaptic growth. While some presynaptic debris appears to be derived from the breakdown of disconnected ghost boutons, we also observed the de novo formation of presynaptic debris in the absence of any ghost boutons. Thus, presynaptic debris is likely directly shed by motorneuron endings. Presynaptically shed debris might derive from dynamically extending synaptopods, whose formation is dramatically enhanced by increasing neural activity [13]. However, in live preparations demonstrating robust synaptopod growth we have yet to directly observe the formation of debris following synaptopod expansion or retraction (Gorczyca M, Ashley J, Fuentes-Medel Y, unpublished data).
The presence of presynaptic debris might highlight the extremely dynamic nature of synapse addition in vivo. Two important mechanisms appear to operate during NMJ expansion. First, the NMJ is shaped by a homeostatic mechanism that maintains synaptic efficacy despite larval muscle growth [32]. Second, the NMJ has the ability to respond to acute changes in activity and sensory experience with rapid modifications in synaptic structure and function. Well-fed larvae placed in a substrate devoid of food show an increase in synaptic strength within 30 min [35], and spaced stimulation induces robust synaptic growth within 2 h [13]. It is tempting to speculate that presynaptic shedding is the byproduct of a mechanism designed to ensure rapid and efficient changes in synaptic performance. For example, the initiation of synaptic bouton formation might be a continuous process. This pool of synaptic boutons might reach an immature stage and if not subsequently stabilized by activity or other signals they might be shed and removed. Such a mechanism would provide a continuous supply of immature boutons ready to stabilize if rapid growth becomes essential.
Glial cells have a key role in the removal of axonal debris and neuronal cell corpses from the central nervous system [22],[37], but mounting evidence also implicates glial cells in the elimination of synaptic inputs. In mammals microglia rapidly spread along neurites of injured motorneurons and displace synaptic inputs through synaptic stripping [38]. At the mammalian NMJ, terminal Schwann cells are also active participants in the activity-dependent elimination of exuberant motorneuron inputs by apparently pinching off fragments of retracting terminals [11].
Here we describe a novel mechanism by which glia, through their phagocytic clearance of shed synaptic debris, can sculpt synaptic connectivity within a single arbor and ultimately modulate the growth of nerve terminals. The formation of shed presynaptic material appears to be autonomous and not require the engulfing action of glial cells since presynaptic debris and ghost boutons accumulate at high levels in draper mutants. Notably, muscle cells collaborated with glia in the removal of shed presynaptic membranes and thus also helped to sculpt the growing NMJ. These observations provide a new view on the role of muscle cells in regulating synaptic growth: muscle cells are not simply postsynaptic target cells that give and receive synaptogenic signals; they are also phagocytes at the NMJ and through engulfing shed presynaptic material can help shape synaptic connectivity.
Why has such presynaptic material not been previously described at the well-studied Drosophila NMJ? This is likely due to the fact that we have assayed NMJ morphology for the first time in engulfment mutants. Even in wild type a very low level of presynaptic debris (this report) and a small number of ghost boutons [13] is observed. However in draper mutants or knockdown animals we observe their dramatic accumulation, which is reminiscent of the process of cell corpse engulfment after apoptotic cell death. Cell corpses are rapidly engulfed during development and thus very few are observed in wild-type animals. In contrast, they accumulate at significant levels in animals with reduced cell corpse engulfment activity, such as C. elegans ced-1 or ced-6 mutants [39].
We found that glial cells extended membrane processes that deeply invaded the NMJ. These cellular interactions were highly dynamic, as demonstrated by our time-lapse imaging, and by the high variability in the extent and type of glial membrane projections we found at the NMJ. Some projections were in the form of thin gliopods that associated with boutons within a branch or that extended across branches. Others resembled flat lamellipodia that associated with synaptic boutons or with the muscle. Given the requirement for glial Draper in the removal of synaptic debris, it is tempting to speculate that glial membranes are continuously and dynamically surveying the NMJ for the presence of synaptic debris, which is then engulfed. Consistent with this notion, we found several examples of glial membranes extending away from the arbor and overlapping with presynaptic debris. We also found that in some cases, HRP positive fragments were found associated with bulbous structures formed by the glial projections, suggesting that glia can engulf presynaptic debris. We also observed glial membrane projections that had the form of boutons, sometimes draping over an entire bouton, or extending well beyond the terminal bouton. While the function of these structures remains unclear we envisage at least two potential roles. First, these might represent glial extensions actively engulfing ghost boutons, although this would be predicted to be a rare event since our cell-type specific analyses argue that muscle cells are primarily responsible for clearance of ghost boutons. Second, these extensions, along with the additional types described above that extend beyond axonal arbors into the muscle, could be physically opening up space in the muscle cell for new bouton formation or process extension.
Interestingly, we found that in draper mutants both disconnected ghost boutons and presynaptic debris accumulated, and this accumulation had a negative effect on NMJ expansion and bouton morphology. Moreover, synaptic growth appeared to be highly sensitive to both types of shed presynaptic material since the accumulation of either ghost boutons or presynaptic debris (when engulfment activity was blocked in muscles or glia, respectively) led to reductions in bouton growth similar to that seen in draper null mutants. As mentioned above, shed material might contain important signaling factors that potently stimulate or inhibit new synapse formation. If, for example, presynaptic debris contains molecules that inhibit synaptogenesis, the accumulation of such material would be expected to negatively regulate synaptic growth. Perhaps a similar type of inappropriate modulation of synaptogenesis by the membrane fragments of pruned terminals also accounts for their rapid clearance from the central nervous system after degeneration.
Drosophila glial cells also engulf neuronal cell corpses and pruned or degenerating axons. Each of these targets is generated by a unique degenerative molecular cascade: cell corpses are produced by canonical apoptotic cell death pathways [40], pruned axons undergo degeneration through a ubiquitin proteasome-dependent mechanism [41], and severed axons undergo Wallerian degeneration via Wlds-modulated mechanisms [26]. Despite their unique pathways of production, each is engulfed by glia through Draper-dependent mechanisms, implying that these engulfment targets autonomously tag themselves with molecularly similar “eat me” cues. Our observations that mutations in draper led to accumulation of presynaptic debris and detached ghost boutons suggests that these new glial/muscle engulfment targets also produce similar cues for phagocytic cells to promote their destruction. If so, these data argue that all the necessary machinery essential for tagging membrane fragments for engulfment are present in a ghost bouton or fragment of presynaptic membrane. Importantly, while a lack of glial-mediated clearance of several targets has been observed in vivo—cell corpses, pruned axons or dendrites, and axons undergoing Wallerian degeneration—almost nothing is known about phenotypic consequences of a lack of glial engulfment function in the nervous system. Here we demonstrate that failure of glia and muscle to clear presynaptically derived material negatively regulates synaptic growth.
In conclusion our studies demonstrate that the process of synaptic growth includes a significant degree of membrane/synaptic instability, and that growing terminals are constantly sloughing off undifferentiated boutons and fragments of membrane. Our observations demonstrate that growing NMJs generate an excess number of undifferentiated synaptic boutons and that only a fraction becomes stabilized and drive the assembly of the postsynaptic apparatus. Exuberant synapses that have failed to form successful postsynaptic contacts are shed, and cleared from the NMJ by glia and muscle cells. The presence of such a pool ensures a continuous supply of nascent synapses available for use to rapidly increase input into the muscle if dictated by dynamic changes in signaling at the NMJ.
The following fly strains were used for this study: draperΔ5 and UAS-Draper-RNAi [26], UAS-dCed-6-RNAi [23]; Repo-Gal4 (a gift from B. Jones), Gli-Gal4 [42], OK6-Gal4 [16], C57-Gal4 and C380-Gal4 [20], UAS-mCD8-GFP [43] UAS-myrRFP (Bloomington Stock Center), MHC-mCD8GFP-Sh [21], and UAS-ChR2 [44]. UAS-Draper-I and UAS-Draper-III were generated by M.A. Logan and will be described in detail elsewhere (MAL and MRF, unpublished data). For larval motility assays, larvae were cultured at 25°C, wandering third instar larvae were collected, briefly washed in distilled water, transferred to the center of a square agar plate, and covered with a transparent lid. After 30 s, total larval movement was followed for 1 min under red light conditions, 60% humidity, at 25°C.
Third instar Drosophila larvae were dissected in calcium free saline [45] and fixed for 10 min with nonalcoholic Bouin's solution unless otherwise noted. Primary antibodies were used at the following dilutions: α-Draper, 1∶5,000 [24]; rabbit α-DLG, 1∶20,000 [46]; mouse α-DLG, 1∶500 (clone 4F3, Developmental Studies Hybridoma Bank, DSHB); α-CSP, 1∶100 [47]; α-Synapsin, 1∶10 (a gift from E. Buchner; [48]; α-Fas II, 1∶3000 [46]; α-GFP, 1∶200 (Molecular Probes); nc82 (α-Brp), 1∶100 (DSHB); FITC or Texas red-conjugated α-HRP 1∶200 (Jackson Immunoresearch). Secondary antibodies conjugated to FITC, Texas Red, or Cy5 (Jackson Immunoresearch) were used at a concentration of 1∶200. Samples were imaged using a Zeiss Pascal confocal microscope and analyzed using the Zeiss LSM software package and ImageJ.
To study the organization of glial membranes at the NMJ we fixed larval body wall muscle preparations of controls and draper mutants expressing mCD8-GFP in glia using the Gli-Gal4 strain for 15 min in 4% paraformaldehyde fix, and double stained the preparations with Texas Red conjugated α-HRP 1∶200 (Jackson Immunoresearch) and α-GFP (Molecular Probes). Glial membrane extensions at identified body wall muscle NMJs from abdominal segments A3 and A4 were scored individually as “blunt ended” (glial membranes terminated at the branch point), “covered” (glial membranes completely ensheathed the NMJ), “gliobulbs” (glial extensions terminated in a bulbous structure), “gliopods” (small finger-like glial membrane projections), and lamellipodia (glial membranes formed flat extensions that partially covered the NMJ). The percentage of NMJs containing the above types of glial membranes projections was calculated from 20 hemisegments for controls, and 15 hemisegments for draperΔ5 mutants.
Presynaptic debris was scored from type Ib boutons at muscles 6 and 7, abdominal segment A3. This quantification was performed using images of α-HRP labeled NMJs that were acquired with identical confocal settings, and the amount of debris scored blindly according to a subjective scale of 0–3. Number of NMJs analyzed are ten to 12 per sample (from six animals). To score presynaptic debris after spaced stimulation, intact larvae expressing channelrhodopsin-2 in motorneurons were subjected to spaced light stimulation as in (Ataman et al. [13]), fixed at 2 h (1.5 h stimulation, 30 min rest) (n = 18 for stimulated samples, n = 12 for unstimulated controls), and 18 h after stimulation (n = 6 for stimulated samples, n = 6 for unstimulated controls), and stained with α-HRP antibodies. Confocal images of NMJs at muscles 6 and 7 (A2 and A3) were acquired with identical settings, and two 8-µm diameter circles at the postsynaptic region of each NMJ branch were selected for analysis using NIH Image software. The number of synaptic boutons and ghost boutons were quantified at muscles 6 and 7 (A3) from preparations double stained with α-HRP and α-DLG (n≥10 NMJs per genotype). Data were represented in histograms as the average±SEM. Statistical significance of the data was obtained in pair-wise comparisons using the Student's t-test.
Live imagining of larvae was performed on either intact or dissected preps as Ataman et al. [13]. Briefly intact larvae were anesthetized using Sevoflurane (Baxter) and the dorsal muscles were then imaged through the cuticle using a 40× 1.2 NA objective on an Improvision spinning disk confocal microscope. Larvae were examined live by expression of UAS-mCD8GFP in motor neurons (pre-Gal4) or glia (gli-Gal4). Increased activity was induced in these larvae by expression of UAS-Channelrhodopsin2, and exposure to a pulsed 491-nm LED paradigm described in Ataman et al. [13] and Figure 1H. Larvae were examined every hour, every 4 h, or at 18-h intervals depending on the experiment. In order to visualize the debris, samples were converted to rainbow gradient color, and then contrast enhanced until the main arbor was saturated, as the debris is much dimmer than the presynaptic membrane.
Live imaging of glia was also performed in dissected preps, as Ataman et al. [13]. Briefly, larvae were dissected in 0.1 mM calcium Drosophila HL3-saline, and imaged on a Zeiss Pascal Confocal (Carl Zeiss) using either 25× or 40× water immersion objectives.
Total RNA was isolated from third instar body wall muscle preparations with Trizol (Invitrogen) and purified using the RNeasy Mini Kit (QIAGEN). First strand cDNA was synthesized using Superscript II (Invitrogen) enzyme and oligo (dT) 12–18 primer (Invitrogen). PCR was performed using the following Draper isoform specific primers to detect expression of Draper-I, Draper-II, or Draper-III: DrprIuECDF (5′-GGGTCCCCTATGTGATATGC-3′) and DrprIuECDR (5′-TTGTAGCACTCGCAGCTCTC-3′); DrprIIuF (5′-GAAAATATATAGCAAGATTTTGTTTCC-3′) and DrprIIuR (5′-TTCGTGTTGTCGAAGCACTC-3′); DrprIIIuF (5′-GTCATTAGACTTTTACACAGG c-3′) and DrprIIIuR (5′-CTAGCGTATAGAATCAGAC-3′). Plasmids containing the Draper isoforms (pUAST-DraperI, pUAST-DraperII, and pUAST-DraperIII) were used as controls for PCR amplification. PCR program was as follows: denature at 95°C for 1 min, anneal at 56°C for 30 s, extension at 72°C for 30 s (30 cycles total). PCR products were run on a 0.8% agarose gel and visualized by ethidium bromide stain.
|
10.1371/journal.pntd.0006689 | Understanding perceptions on 'Buruli' in northwestern Uganda: A biosocial investigation | An understudied disease, little research thus far has explored responses to Buruli ulcer and quests for therapy from biosocial perspective, despite reports that people seek biomedical treatment too late.
Taking an inductive approach and drawing on long-term ethnographic fieldwork in 2013–14, this article presents perspectives on this affliction of people living and working along the River Nile in northwest Uganda. Little is known biomedically about its presence, yet ‘Buruli’, as it is known locally, was and is a significant affliction in this region. Establishing a biosocial history of ‘Buruli’, largely obscured from biomedical perspectives, offers explanations for contemporary understandings, perceptions and practices.
We must move beyond over-simplifying and problematising ‘late presentation for treatment’ in public health, rather, develop biosocial approaches to understanding quests for therapy that take into account historical and contemporary contexts of health, healing and illness. Seeking to understand the context in which healthcare decisions are made, a biosocial approach enables greater depth and breadth of insight into the complexities of global and local public health priorities such as Buruli ulcer.
| Buruli ulcer, a neglected tropical disease, has been described as an emerging public health problem in parts of sub-Saharan Africa. One of the challenges highlighted by the World Health Organisation (WHO) is improving access to biomedical healthcare. A research priority is thus to determine local understandings of skin disorders such as Buruli ulcer, and social-cultural factors that influence health-seeking. This article explores perspectives on Buruli among fisherfolk in northwestern Uganda along the River Nile, where the ulcer has previously been documented. The findings are based on a long-term ethnographic study of health, healing and illness in this region, and integrate insights from biomedical and social sciences. This biosocial approach demonstrates that, rather than seeking therapy late, people in this region sought treatment from local herbalists promptly when signs of skin lesions appeared. This was not because of non-biomedical understandings of disease. The reasons why people continue to trust local herbalists as experts in Buruli can be found in the historical context of how ulcers have been understood and managed, and the broader context of quests for therapy in this region. While the findings relate to where this study took place, the lessons learnt and biosocial approach used could be usefully applied in other settings where Buruli ulcer is endemic, and for understanding the local context of other neglected diseases and global health priorities.
| A so-called neglected tropical diseases, Buruli ulcer occurs in rural areas with limited access to safe water, basic medical care and education [1]. Caused by Mycobacterium ulcerans, similar to mycobacterium that cause leprosy and tuberculosis, people affected develop nodules and other skin lesions typically on exposed areas of the body [1,2]. It is estimated that one third of early nodules resolve spontaneously [3]. However, severe long-term complications can develop from the toxin-producing bacteria with aggressive lesions causing permanent damage such as scarring, contractures and destruction of underlying bones. Outbreaks have occurred following ecological changes [4], and while the mode of transmission is not entirely clear, it has been associated with practices such as farming in swampy areas and along slow-moving water bodies, and various hygiene and wound care techniques [1,2,5,6]. The complexities that are not adequately understood through biological and epidemiological research clearly underscore the entanglement of biological and social dynamics of transmission.
Following a visit to West Africa in 1997, Dr Hiroyoshi Nakajima, then Director General of the World Health Organisation (WHO), declared Buruli ulcer an emerging disease and the Global Buruli Ulcer Initiative was launched in 1998 under WHO directive. Rather than interrupting transmission, which is poorly understood, the public health control strategy focuses on early detection and access to biomedical treatment [7]. However, there are implementation challenges. For instance, while the gold-standard is laboratory diagnosis, in practice a clinical diagnosis is often made based on presentation and exclusion of other skin diseases. And while the current recommendation for managing early lesions is an eight week course of antibiotics used in tuberculosis treatment and for severe lesions includes surgical intervention (amputation of the limb and skin grafting), the disease tends to occur in remote, rural areas where these facilities remain lacking.
Scope for control is further limited by paucity of data representing burden and patterns of disease. Reporting to the WHO has been in place since 2002, but evidence from West Africa where Buruli is highly endemic suggests significant under-reporting. In 1999, a national case search in Ghana identified 5619 patients with Buruli lesions (3725 active lesions and 1894 healed lesions) [8]. This is a substantial number given that from 2002 to 2012, between 2632 and 5867 cases were reported annually for Africa as a whole. Likewise in Nigeria, which typically has not reported cases, significant numbers were found in a case-finding study [9]. Epidemiological and clinical studies have thus pursued case-finding and mapping exercises.
Furthermore, it is often reported that people present late for biomedical treatment, when severe ulcers have developed. Thus, one aim set by the WHO was to: “reduce the proportion of category III [disseminated/severe] lesions from the 2012 average of 33% cases to below 25% by the end 2014 (sic)” [7, p.2]. Evaluations of access and barriers to biomedical healthcare report problematic healthcare seeking behaviours [10] providing explanations including stigma [11], knowledge and non-biomedical explanations of disease and forms of healing [12]. Yet, there has been little social science and anthropological research on this affliction. Noticeable exceptions include Grietens and colleagues’ [13] research unpicking the oversimplification between the role of local beliefs and treatment-seeking for Buruli in Cameroon, foregrounding an exploration of local knowledge. Whereas Giles-Vernick and colleagues [14] demonstrate additional insights to biomedical models of Buruli transmission by exploring ethno-ecological histories in Cameroon. More recently, anthropological insights have been incorporated into developing models for improving Buruli ulcer policy in Cameroon [15].
Northwestern Uganda is an intriguing area to study the historical and contemporary significance of Buruli ulcer. Historically, Buruli was significant in the region of northwestern Uganda along the River Nile. Ulcers were described in Uganda and Zaire (now Democratic Republic of Congo) by Sir Albert Cook in 1897 and Kleinschmidt in the early twentieth century [3]. Interestingly, a study of genetic diversity in Africa reports two main strains of Buruli with different lineages [16]. The Ugandan strain was one of the oldest in Africa (present for centuries), while the second strain (common in Gabon, Benin and Cameroon) appears to have been introduced in the 19th century–which the authors attribute to the upheaval of the neo-imperialism period. While Buruli primarily occurs in African countries, it was in fact first identified as Bairnsdale ulcer in Australia in 1948 and re-named from 1961 when many cases were reported in the then Buruli District (now Nakasongola District) of Uganda near Lake Kyoga [17,18]. During the 1960s and 70s, the Uganda Buruli Group described the disease among recent Rwandan refugees who settled near the River Nile [19]. At a similar time, Barker [17, 20] also associated the disease with swampy areas along the Nile, suggesting that the disease spread and became endemic in areas of Uganda after floods in 1962–64 created new sites of permanent swamps. Contemporary reported data suggests Buruli has declined as a public health concern in Uganda. In 2002, Uganda reported 117 cases of Buruli ulcer to the WHO (2632 from Africa, 3269 globally) compared to three in 2009 (5029 from Africa, 5084 globally) [21]. Since 2010, Uganda has not reported data, but previously the numbers were relatively low.
However, as literature from West Africa suggests, we clearly cannot rely on reported healthcare data to represent presence of disease let alone estimate prevalence [8,9].
This became evident during long-term inductive, iterative, ethnographic fieldwork in Moyo and Adjumani districts, northwestern Uganda, in 2009 and 2013–14, which this article is based on. In the two districts, while largely unseen at the health centres, Buruli ulcer was well-known among people along the river. What explains these diverging accounts?
This article addresses this question, taking as its starting point the perspectives of people vulnerable to (and affected by) neglected diseases such as Buruli ulcer. Echoing broader calls for biosocial approaches integrating insights from across the biological and social sciences [22, 23, 24, 25], this article builds on strands of interdisciplinary scholarship, bringing historical biosocial insight to understandings of afflictions and treatment practices. Drawing on long-term inductive, interdisciplinary fieldwork on neglected tropical diseases in Moyo and Adjumani districts in northwestern Uganda, this article presents an analysis of the social responses to an affliction known locally as ‘Buruli’ (italics are used throughout the article to distinguish the vernacular use of the term, as opposed to the common biomedical name, Buruli ulcer which refers to the biologically defined Mycobacterium ulcerans), along this stretch of the River Nile. This suggests a more complex picture, and highlights the disparities between the framing of global health priorities and perceptions on these priorities by people affected. It challenges the oversimplification and problematisation of healthcare-seeking, and drawing on historical and contemporary insights, demonstrates the pragmatic and empiric nature of quests for therapy.
This article is based on themes that emerged through long-term ethnographic fieldwork, in Moyo and Adjumani districts, Uganda, near the South Sudan border including three months of fieldwork in 2009, and a further 16 months in 2013–14. An inductive, iterative, interdisciplinary approach was adopted. The primary aim of the broader ethnographic research was to explore the context and everyday realities of neglected diseases among people living and working along the River Nile–a population deemed vulnerable to neglected tropical diseases–in order to understand the social responses to disease and public health, and through this establish wider implications for public health policy.
Ethnographic fieldwork included participant-observation, group discussions, semi-structured and open-ended unstructured interviews with key informants at fish landing sites (including fishermen, fish-processors, fishmongers, local council members, Beach Management Unit members and elders), along with health workers, local healers and district authorities (health and fisheries). The broader study included an epidemiological-parasitology survey conducted in collaboration with the District Vector Control Division of the Ministry of Health. This was of adults at twelve fish landing sites, one from each sub-county along the river and one island, investigating another neglected disease, schistosomiasis, (documented elsewhere [26]).This article presents findings that emerged from the ethnographic fieldwork related to Buruli ulcer, as documented in ethnographic fieldnotes, interviews and discussions.
Within this, twenty-one in-depth interviews were conducted with people who self-reported, incidentally in the survey and during further discussions and interviews, to have suffered from ulcers understood as Buruli. In other interviews and discussions, people discussed forms of local poisoning, some of which were related to skin ulcers. In-depth interviews were held with four male herbalists, identified by people locally, who treated suspected Buruli ulcer (Buruli), and numerous conversations including four semi-formal interviews were held with district health staff and healthcare workers on the subject. In addition, it was discussed in twelve semi-/in-formal group discussions with men and women at the landing sites. During conversations exploring understandings of ulcers and skin lesions, I sometimes showed people photographs from a poster (WHO, Global Buruli Ulcer Initiative), found in a health clinic cupboard. The photographs illustrated different forms of Buruli lesions, from nodules, papules and plaques, to oedema and various stages of ulcers. This elicited dynamic discussion on the causes of the different lesions as well as the types of ulcers that people had suffered or seen.
Discussions and interviews were held in either a local language–while I had some command of Madi to understand and participate in basic conversations, in-depth interviews were translated by an experienced local researcher who had conducted previous research with other research programmes- or carried out in English. Interviews were not recorded, but detailed notes taken and written up in full afterwards. Adults who had participated in the initial survey (not presented here) and follow-up interviews provided written consent. Rather than a one-off event, consent was an ongoing process and all adults who participated in additional interviews and discussions provided verbal consent. The London School of Economics and the National AIDS/HIV Research Committee (NARC) in Uganda for the Uganda National Council for Science and Technology (UNCST) granted ethical approval for the study (ARC141). UNCST granted research approval, and approval was also sought through both districts’ health offices.
This article presents an inductive thematic analysis of the ethnographic fieldnotes and interview transcripts. This was carried out manually, drawing out emerging themes on understandings of ulcers, healing practices sought and the relationship between local forms of therapy (erua Madi) and biomedical healthcare. These themes were strongly underpinned by a sense of historical legacies. Findings were triangulated with published data on Buruli and other ulcers in this region, by comparing the oral accounts and histories with available historical documentation, biomedical literature and other ethnographic accounts.
Moyo and Adjumani districts are situated along the River Nile bordering South Sudan. In the most recent census, 2014, the population of Adjumani was 225,251 (116,953 females, 108,298 males) and in Moyo was 139,012 (70,072 females, 68,940 males) [27]. The people are predominantly Madi (and identify as Christian, particularly Catholic, but also Protestant and, more recently, Pentecostal). However, in Obongi County of Moyo they identify with a number of neighbouring ethnic groups, including Lugbara or Kakwa related groups. A mix of languages are therefore used, but Madi is most widely spoken. The majority of people in Moyo and Adjumani are subsistence farmers, or fishing-farmers—along the river the land is particularly fertile and the fish business is important economically for men and women. People along the river are referred to as meri ti ba (people of the river) in Madi. It is difficult to assess the exact population along the river, with ongoing flux of people into and out of the fish business.
The northwestern region of Uganda has a long history of social upheaval, political and economic marginalisation [28,29]. Slave and ivory traders were active in the nineteenth-century, and subsequently the region was under various colonial and protectorate rule. During the colonial period, in now Moyo and Adjumani districts (then Madi sub district, named after the predominant ethnic group), there were a number of colonial public health campaigns, for instance for yaws and sleeping sickness. From the late 1970s, the time of decolonisation and Independence, through to the 2000s, there have been ongoing conflicts in the region causing an almost constant flux of refugee movements across the border with now South Sudan. During fieldwork, in December 2013, conflict broke out in neighbouring South Sudan leading to an influx of refugees into Adjumani and subsequently Moyo too.
Over time, people have experienced intermittent and long-standing presence of various NGOs including medical and humanitarian organisations. Aside from this, biomedical healthcare has largely been provided through both government and private facilities. Each district has a government hospital in the main town, which for some areas along the river is an hour’s motorbike ride away. Health centres at parish level are typically run by nursing staff and stock essential medicines. At sub-county level they also have some laboratory facilities. In addition, there are numerous private facilities, particularly drugstores, in trading centres. Thus in most areas along the river, there is some access to biomedical healthcare, although what is provided varies and is often limited. Alongside the public health system, there have been numerous disease-specific health programmes, for vaccine-preventable diseases, malaria and a number of the neglected tropical diseases. In contrast to other neglected diseases, such as intestinal helminths, onchocerciasis, human African trypanosomiasis and lymphatic filariasis, there has not to date been a specific public health programme for Buruli. Rather, the ‘strategy’ relies on patients self-presenting to the healthcare system—referred from the village health team or local health centre to a district or regional hospital (Arua, Gulu or Kampala). As well as biomedical healthcare facilities, therapies for various afflictions are provided by local herbalists and witchdoctors, as will be discussed in further detail below.
Findings from fieldwork illustrate how in reality, people rarely presented to the health centre for suspected Buruli in the first instance. This was brought to my attention during a meeting early on in the research when a doctor pointedly asked, ‘but what do you know about Buruli ulcer?’ It was rarely seen in the hospital however ‘deep in the village, it was found to be there’. During fieldwork, it became clear that for people living and working along the river, Buruli was a well-known affliction. Unlike global health rhetoric, Buruli was far from emerging; and unlike at the hospital, it was far from unknown or unseen. While there had been some adoption of biomedical practices and understandings of the disease, people predominantly continued to use treatment that they knew and trusted from local herbalists. For other diseases, such as schistosomiasis, herbs were used as a ‘last resort’ when or where medicines were not available, and people made claims on their ‘rights’ to healthcare when these needs were not being met. Yet for Buruli, interestingly this was not the case.
For instance, at the time of fieldwork, one health centre near a fishing site was treating an adult with active Buruli ulcer while another reported to have seen a case in the previous year, but otherwise it was rarely diagnosed. On the other hand, during fieldwork, local herbalists near the river reported seeing cases of Buruli every year or two. One even suggested he saw somewhere between two to seven cases per year, including in the previous twelve months. This leads on to the questions: how can we explain these diverging accounts? How, and to what extent do understandings of Buruli relate to Buruli ulcer? How, and why, do people seek particular therapies for suspected Buruli ulcer, or Buruli, and what does this mean for public health policies seeking to address neglected diseases such as Buruli ulcer?
In the following, the findings from the analysis are synthesised around three inter-related themes: the historical context of Buruli ulcer in Moyo and Adjumani districts; understandings of ulcers known as Buruli, and their relation to other forms of skin disease and afflictions; and erua Madi (Madi medicine) and biomedicine.
A review of various field studies carried out in the 1960s and 70s in now Moyo and Adjumani districts shows that Buruli ulcer was a significant public health concern [17, 19, 30]. Clancey and colleagues [19] reported four active cases from Moyo; and Lunn et al [30] reported ‘burnt out cases’ suggesting earlier infection that had resolved. While Barker wrote: “on the west side of the river [Moyo], where the land is hilly, the disease is confined to the river’s edge; but on the east side [Adjumani], where the land is flatter, the disease extends up to 10 miles from the river” (p.43) [17]. He proposed that there was no evidence to suggest the onset of Buruli from more than 10 years prior to the 1962 flood (p.872).
The findings presented in this article reflect this time period—a number of adults interviewed were born in the early and mid-twentieth century. Those who reported to have had Buruli were affected from the 1960’s onwards, with the highest proportion affected in the 1990s. Yet, oral accounts suggest that these forms of ulcers clearly had a much longer history. Most people we spoke to, including elders and herbalists who had grown up in Madi region during the 1950s and 60s, used the term Buruli (although ‘lupi lupi’, referring to a swelling, was also used and understood by other regional language groups). However, it seems that prior to this period, similar ulcers were known by other names. Lunn et al [30] noted the terms used for ulcers as ““juwe okoro” or “bile okoro”, meaning “the sore that heals in vain”” (p.278). A herbalist born in the 1920s recounted to me that when he was training, the wound that is now known as Buruli was previously called macodo—translated as ‘abscess’ in a Ma’di dictionary compiled in 1941 [31]. He described macodo as ‘a boil which swells bigger and is more serious’, reporting that it was treated in a similar way to Buruli today. Likewise, another herbalist, born in 1939, recalled:
Other skin lesions (nodules, boils, ulcers and wounds) were differentiated from Buruli and literature and documents from the twentieth century confirm that various forms of ulcer were endemic in northern Uganda. In the early twentieth century, ulcers were a preoccupation for the British administration in Uganda. Vaughan [32], for instance, describes the moral panic stemming from the high rates of ulcers among Bugandans, likely misinterpreted as syphilis, a sexually transmitted infection. On the other hand, in northern Uganda, treatment campaigns were for the non-sexually transmitted form of the disease, yaws (p.138). In a 1927 sleeping sickness campaign in Moyo it was reported that there was a “great number of cases of Yaws and Ulcers that came up for treatment” and during two months they administered nearly 2000 injections for these “often repulsive” ulcers [33]. Interestingly, during fieldwork in 2013/14, an elder described a disease that ‘bent the bone’ (seen with yaws) which he differentiated from Buruli, reporting that it had not been seen for a long time. The term Buruli, adopted by Madi to encompass particular ulcers, was thus differentiated from other skin lesions. While the vast majority of the swellings and ulcers discussed and identified as Buruli had not been diagnosed bio-medically, as I will demonstrate, there are consistencies in how Buruli and Buruli ulcer are identified.
People’s experiences and understandings of Buruli resonate with biomedical understandings of Buruli ulcer, and the majority of people, although unsure, used ideas of worms (obu) to explain it. This may be related to the focus on ‘worms’ in contemporary public health campaigns, particularly controlling intestinal helminths through mass treatment campaigns to ‘at risk populations’ including fisherfolk. In 2009, when I first visited this area of Uganda, fishermen often lamented, ‘we are the rubbish pit for worms!’
In 2014, a herbalist explained Buruli transmission:
Many supported the deduction that Buruli was found along the rivers or in muddy places. In this sense, it was very much associated with fishing areas and practices. One elder and Beach Management Unit member wondered if Buruli was acquired from fishermans’ tools used in the past, such as hooks and nails. A few people recounted how the disease came from Buruli district, as this 70 year old herbalist explained:
One hypothesis was that it was ingested by eating mudfish. When gutting mudfish, women reported seeing small white worms in the stomach or gills, which some deduced might be the cause of Buruli. This was supported by seeing small white eggs, presumed to be those of the worms, in the lesions when the herbalist treated Buruli. Interestingly, mudfish are found in the Nile and its smaller tributary rivers and in the rainy season men go with spears to these areas to catch the fish. Given that transmission of Buruli occurs along slow-moving water bodies, this may be an activity that makes men vulnerable to infection.
The empirical basis to these deductions is clear. Yet, as with scanty biological explanations there was still an air of uncertainty and many unknowns. One male elder herbalist reflected:
However, unlike in West Africa where magical-spiritual explanations appear to be much more prominent [11], in Moyo and Adjumani, these explanations did not appear to be as pervasive as one might expect. In contrast to witchcraft or local poisoning, herbalists were clear that Buruli could not be inflicted by another person.
Even in the case of severe, late ulcers, the majority of people identified these as Buruli, although the distinction was not clear cut. A minority reported that these ulcers can be caused by erua hwe, a form of local poisoning; however this was an exception rather than the norm. Herbalists reinforced that these lesions occurred when they had not been treated in time and denied they had ever been a form of witchcraft or local poisoning. However, one elder herbalist had this to say of late ulcers:
He went on to say that he had recently seen a woman with this form of affliction.
On the other hand, others described more distinction between the different afflictions. For instance, a 70 year old herbalist reported:
Another 52 year old man who had Buruli when he was 14 reported that he had not seen late ulcer and when asked about local poisoning and erua hwe, replied:
In light of these different explanations, a 62 year old woman who had experienced Buruli in her 20s, reflected on how treatment informed the distinctions:
Other forms of ulcer-causing poisoning also existed in the past. One elder fisher-farmer showed a scar above his ankle from an ulcer when he was 18 years old, and explained:
Thus, these accounts show how over time, ulcers and lesions have been differentiated by signs, symptoms and observed effectiveness of treatment, drawing on overlapping explanatory models.
Exploring perceptions, understandings and experiences of Buruli within this, when shown the photographs from the WHO poster all but one woman indicated that the nodule was jue, a boil. Seemingly, papules were rarely seen, with only one person suggesting it could be Buruli. On the other hand, people identified plaques, non-ulcerative oedema and early ulcers as Buruli. Rarely, early ulcer with indurated edges was associated with cancer. On the whole, most people who had had Buruli had experienced forms of early ulcer. On many occasions this wound developed after the herbalist began treatment. Some were affected once, others multiple times. Similar to epidemiological descriptions, the vast majority of people had experienced lesions on their legs which most commonly started with an itch, described as like a small prick of a thorn, developing within days into a small swelling or skin changes. Others first noticed a small nodule, and a minority experienced pain or other sensations.
Noticeably, people clearly recalled their experiences of Buruli. A 20 year old female fishmonger was affected when she was eight:
Likewise, a 36 year old fisherman recalled:
And a 28 year old female fishmonger recollected:
Not only do these narratives overlap with biomedical descriptions of Buruli, in addition, they demonstrate how quickly people sought ‘expert’ advice. Contrary to experiences at health centres, and perceptions on late presentation portrayed in public health literature, people in fact sought what they deemed appropriate healthcare relatively quickly. Advised by parents or other family members on where to seek treatment, people affected by ulcers, jue and Buruli largely sought therapy from a herbalist, and the majority did so within a few days of experiencing symptoms. Despite Buruli being understood to a large extent as a disease with biomedical causes, the experts, were local herbalists. It was the herbalists and the perceived efficacy of their treatment that people trusted.
Herbalists that treated Buruli learnt their skills, techniques and practices from elder herbalists. Their apprenticeship took years. Knowledge of herbs was learnt from their teacher, revealed in dreams, or through trial and error. While some herbs were collected from elsewhere, sometimes at the foot of a mountain, they were predominantly found growing wild around the home. One elder herbalist pointed to the common grasses around his house that he used, and laughing said, ‘you see, these people don’t know’. These local herbs are referred to as erua Madi (literally, ‘Madi medicine’), or erua abi dri more generally (‘medicine from ancestors’, or ‘medicine given by ancestors’) and are opposed to erua Mundro (‘European’s medicine’, or biomedicine).
When a herbalist was approached for advice he first assessed whether the lesion was Buruli, jue or another condition. One herbalist described examining for pitting oedema–observing an indentation after pressing the swelling with a thumb–a sign of Buruli. Another test involved making a cut at the site of the lesion (for example on the leg) and elsewhere (for example on the arm), demonstrating to the person that the colour of the blood from the area with Buruli was darker.
Herbalists’ treatment varied to a certain degree, but their general approach was similar. A 36 year old male fisher described his experience:
Herbalists often began by making small superficial cuts to the skin around where the problem was ‘to make the ‘germs’ collect in one place’. Typically, after a number of days when the swelling had accumulated, the herbalist made a deep cut to release pus immediately around the swelling. Mixtures of fresh or dried herbs were then applied topically. In addition, some herbalists made a drink from the herbs. Others specifically did not use herbal drinks for Buruli–only for local poison as herbal drinks precipitate diarrhoea and expel poison from inside the body and bloodstream.
This treatment was sometimes extensive, especially considering anaesthetic was not used. As one woman described during a discussion with female fishmongers:
Herbalists explained that in the past, such severe forms of Buruli requiring extensive treatment were more common. The deep cuts were completed in one sitting, as one herbalist said:
He explained that if it is not completed in one sitting, ‘it will continue to eat the flesh’, and the wound will spread, developing into a late ulcer. He went on to state that they had to be very careful managing the waste from cleaning the wound–‘so that no one can tamper with it’. Reflecting on how common it used to be for a number of people in one household to be affected, he suggested that perhaps this wasn’t done so rigorously in the past.
Depending on the severity, the wounds took weeks, months or years to heal. For a few people, Buruli had affected the use of the limb and it was common for people to suffer from pain long after the ulcers had healed, particularly when it rained. One male elder continued to make superficial cuts and apply herbs to his leg each period the rainy season began and his pain returned. Buruli had also affected some people’s day to day lives, limiting women’s ability to carry out daily chores, or children’s time spent in school. There were reports of people having died from late ulcer lesions, but most it seemed recovered and healed with little long-term sequelae except scarring, which people openly showed when the topic of Buruli came up during conversation. Some of these scars were the depressed scars of the ulcer itself, others were from the herbalist’s cuts.
Over time, there has been a continuing relationship with biomedicine surrounding treatment practices for Buruli. Along the river, some people who experienced Buruli had solely used erua Madi, with others using a combination of erua Madi and biomedical treatment. This reflects broader plurality in quests for therapy in this region. In part, decisions around treatment were influenced by who people went to for advice–normally a family member—before seeking treatment; and in part by their own experiences, or the circumstances that people found themselves in at the time. Only one 23 year old fisherman reported that he solely sought biomedical treatment–in this case, he was taken by his parents to a nearby health centre. Otherwise, people who had been affected, particularly those affected in the last 15 years, reported that they had initially visited a local herbalist for treatment (cutting) before attending a health centre for antibiotic injections. For instance, in 2013 a 30 year old woman first sought treatment from a local herbalist before concurrently receiving antibiotic injections and finally undergoing surgery. Another man suffered Buruli three times in the 1980s and 1990s. The first time he solely used erua Madi, however when the ulcer recurred he concluded that the local herbs had not sufficiently treated it and so sought biomedical treatment. At the second recurrence in the 1990s, during which time he was displaced due to conflict, he again used erua Madi.
It is noteworthy that both erua Madi and biomedicine can be used for Buruli. Contrastingly, for local poisoning only erua Madi can be used: if a person affected by local poisoning consumed erua Mundro their condition deteriorated or they experienced a potentially fatal reaction to the medicine. During an interview about Buruli, a 20 year old female fishmonger described a separate experience of local poisoning that was initially treated at the hospital:
Describing this adverse reaction to the biomedicine administered in the hospital, this experience was understood to be a sign that therefore the cause of her initial sickness was not biomedical. Rather, it was a form of poisoning, and this explanation was reinforced by the fact that when the biomedicine was stopped and the herbalist’s treatment started, the reaction subsided. Thus, through these empirical observations herbalists were deemed best placed to deal with particular conditions. Yet seeking erua Madi from local herbalists was not solely because of non-biomedical understandings of disease and illness. While erua Madi was used for Buruli, in contrast to local poisoning this was not because it was an interpersonal affliction. Buruli was still a Madi disease, because of the history of ulcers and treatment pre-dating the introduction of biomedicine.
In the case of ‘worms’ like schistosomiasis, people often reported that they used erua Madi as a last resort, because they were far from a health centre, the health centre was regularly out of stock of medicines, or they were not able to access the free distribution of praziquantel. Research on social responses to mass drug administration for schistosomiasis and other neglected tropical diseases in Moyo and Adjumani has demonstrated how demand for the public health programme and biomedicine in part draws on ideas of modernity and questions of citizenship–a demand for the state provision of resources where it has typically been lacking [34]. Even though the public health provision of Buruli ulcer treatment is similarly limited, such claims questioning the lack of biomedical treatment are not being made. In part this could be due to limited public health campaigns in contrast to other diseases. For instance there have been numerous, long-standing malaria interventions, such as the distribution of bednets and provision of anti-malarial treatment and Co-artem for malaria is widely accessed and used as soon as symptoms interpreted as malaria develop. More recently, hepatitis ‘emerged’ as a serious public health concern in 2010. A childhood vaccination programme has been introduced, yet there is demand for biomedical testing, vaccination and treatment beyond this, and a questioning of the limited availability of these resources [34]. Social responses to Buruli have clearly evolved in a different way to these other diseases.
During long-term fieldwork in the 1980s, Allen [34] describes a very similar situation, with Buruli understood as a Madi illness from impersonal causes requiring treatment from herbalists. Standard antibiotics at the time would not have been the current WHO-recommended regime for Buruli ulcer, and Allen [34] similarly notes that despite complications and fatalities, some cases of severe disease were cured by herbalists. Allen points out that this provided further empirical support for their expertise. Even earlier in the twentieth century it appears to have been a similar picture. Lunn et al’s study of Buruli ulcer in Madi District in the 1960s found that out of 39 new cases of disease, “On two occasions patients were found to be applying powdered herbs, a procedure which forms a thick dry crust over the ulcer bed” (p.278)[30]. The authors also reported evidence of ‘burnt-out’ cases: “some parents displayed their children proudly, affirming that their ulcers had healed without Western medicine” (p.279)[30]. Thus, there has not been an apparent dramatic change in practices over this 50 year period, and the perceived efficacy of herbalists’ treatment has persisted despite the general expansion of biomedicine during this time [29].
To some degree this is not surprising and there are a number of possible explanations. On the one hand, there is a perceived lack of efficacy of biomedical treatment. There has not been a concerted public health strategy to actively identify and treat Buruli and, from what I could ascertain, the broad-spectrum antibiotics widely available at local health centres were not the WHO-recommended regime (drugs which were available through HIV-TB services). On the other hand, there is a perceived efficacy of herbalists’ treatment with successful treatment of many nodules and ulcers providing evidence. This raises the question: what is it about herbalists’ treatment and practices that are effective? Interestingly, it has been reported that one third of early stage Buruli nodules resolve spontaneously and excision is estimated to have an 84% cure rate [3]. As described in this article, people tended to seek herbalists’ advice within a few days of initial symptoms, and herbalists treated nodules at this early stage. Furthermore, in Ghana, antimicrobial properties have been identified in the herbs used and hot poultices applied [35, 36]. Therefore, there are conceivable biomedical explanations for the effectiveness of herbalists’ practices.
But the effectiveness of herbalists’ treatment and practices goes beyond a consideration of biomedical plausibility, nor has it been without biomedical influence. Herbalists have incorporated biomedical knowledge, and their practices have adapted to new developments, technologies and biomedical threats. One herbalist explained how their practices had changed. Before a period of exile in the 1980s when people fled to southern Sudan, herbalists made their cuts using the head of an iron spear. Since razor blades became available, patients are required to bring their own, and whereas before the blades were reused, now they are for sole use, as one herbalist explained:
It was also not uncommon for herbalists to have a consultation record book, similar to hospital record-keeping, with details on the patient, sickness, and treatment given. There have been attempts to more formally professionalise herbalists [34] and herbalist associations have been established in the town. Yet the majority still practiced independently at their homes particularly in rural areas with little formal governance.
The relationship between herbalists and biomedical practitioners has not been without tensions. As a 70 year old male herbalist recalled:
While biomedicine was seen to offer antibiotics, as this herbalist explained, for Buruli it is the cutting and visible release of pus that is important, otherwise neither treatment will work. Cutting is a key feature of the herbalists’ treatment. This is interesting because some studies have suggested that a fear of surgery leads people to avoid hospital care and seek traditional healers [11, 37]. Yet from my discussion with people in Moyo and Adjumani, and seeing the scars of healed ulcers, a herbalist’s treatment was sometimes extensive, and treatment by a herbalist was preferred because of a perceived lack of intervention by medical staff at the health centre. People explained how the condition worsened if somebody attended the health centre as medical staff, following protocol, waited before lancing nodules or they only administered antibiotic injections. In one village, women described how healthcare workers ‘feared’ to lance a boil, preferring to refer patients to a herbalist (although the herbalist referred to denied being aware of this). In addition, antibiotics commonly available at health centres were likely not the most effective, depending on the microbial cause of the lesion. Furthermore, health centres’ drug stock is frequently limited, further undermining biomedicine as a credible source of treatment. On the other hand, erua Madi is empirically perceived as efficacious and the long history of herbalists managing jue and Buruli with erua Madi mean they are trusted as experts.
Thus, there are multiple reasons why people rarely presented through the biomedical health system–labelling this as ‘late presentation for treatment’ is clearly misleading. Village health teams were reportedly advised to report ‘wounds that don’t heal’ to health centre staff, and to ‘monitor and inform’ the health authorities of local herbalists treating such cases. But in reality, a healthcare worker explained:
Indeed, during a conversation, one woman even asked me:
Herbalists raised similar concerns in the past about the need for early diagnosis and treatment, but now, it was said, people were aware that if they sought treatment from a herbalist early they could be cured within a few days without the need for deep cuts. From this perspective, there was little need to question a lack of biomedical input or even seek it, at least initially and certainly not solely. In this respect, rather than ignoring initial symptoms of Buruli and jue, people were responding quickly and seeking out treatment that they deemed appropriate (whether solely from the herbalist or a combination of erua Madi and biomedicine), even if that doesn’t fit biomedical notions of appropriate treatment.
Accordingly, there has been an apparent reduction in severe Buruli. As a 70 year old herbalist remarked:
Another herbalist also reflected on the declining number of cases and clusters within households. Likewise, a district staff member reported that Buruli was ‘near eradication’, yet these patterns of disease were not documented, with limited, or no epidemiological data. It is not entirely clear what enabled this reported decline given that people are still involved in fishing and farming along the river where the disease is likely to be found.
Buruli remains significant though: these historical encounters shape contemporary responses. When I asked people along the river how they would advise others who developed Buruli, invariably the reply was ‘to go to a herbalist’. In this article I have described the situation in rural areas along the river, and responses may well be different in towns where people have access to the hospital and numerous private clinics. Indeed, when a young boy in a district town developed ‘jue’ his mother took him without question to a nearby private clinic for lancing and antibiotics. However, when a young fisherman at one landing site developed a painful swelling on his leg he adamantly refused to attend the health centre, even when I offered a lift:
Other people present confirmed: these sorts of afflictions, jue, Buruli, were treated at home not at the health centre.
Is Buruli ulcer a neglected disease in this region? Arguably, it depends on whose perspective is taken. Buruli ulcer remains a challenge: hospitals and health centres rely on people self-presenting, yet, for the reasons given people rarely do so. Far from being an ‘emerging public health concern’, for people living and working along the River Nile in Moyo and Adjumani, an area historically endemic for Buruli Ulcer, Buruli was a well-known affliction and similar conditions have been managed long before the introduction of biomedicine. It was neither emerging nor perceived as a particular threat.
The accounts presented in this article clearly show how statements of late presentation for biomedical treatment and healthcare-seeking are misleading and oversimplified. This historical biosocial analysis of Buruli in northwestern Uganda has elicited alternative, and deeper insights into contemporary perceptions and practices. The significance of the history of ulcers, herbalists and biomedicine in this region was evident–an aspect often understudied. Piecing together published accounts and oral histories shows how biomedical and historical documents echo people's accounts; from both perspectives, ulcers have continuously been reinterpreted. Understandings of Buruli are not necessarily contradictory to biomedical models, and do not exclusively explain healthcare-seeking. Understandings of ulcers and treatment practices in this region have developed over a long encounter with skin lesions, shaped by historical and contemporary encounters with biomedicine and long-established therapies. This enquiry illustrates the pragmatic, empiric nature of quests for therapy, and the early presentation to herbalists who are trusted to manage these lesions based on the longevity of the afflictions and healing practices. Establishing this social history goes some way to explain contemporary responses to ulcers, healing and healthcare.
This research builds on calls for biosocial approaches to global health priorities including neglected tropical diseases, and anthropological insights into local beliefs and treatment-seeking. Thus far, research that has explored non-biomedical understandings of disease and therapies for Buruli ulcer has largely not considered the broader historical and social context. Taking this into account elicits additional insight into why people seek healthcare in particular ways. Exploring responses to Buruli in northwestern Uganda shows how these lesions are part of the history of this region and elicits insights into understandings of afflictions, quests for therapy and encounters with biomedicine, which bear relevance for understanding contemporary perceptions and practices relating to global and local public health priorities. However, there are limitations. Firstly, this article is based on an analysis largely of accounts of previous quests for therapy for Buruli/Buruli ulcer–it was not possible to follow quests for therapy as they unfolded, nor was it possible to say for certain if the self-reported cases were Buruli ulcer, although the oral accounts reflect biomedical accounts. Secondly, it was beyond the scope of the study to carry out diagnostics, and therefore we cannot document biological presence of disease. Finally, the impact of Buruli in Uganda is arguably not comparable to that presented in research from highly endemic countries such as in West Africa, however, there are broader insights gained from reflecting on the approach presented in this article.
That is, what a biosocial approach to understanding healthcare-seeking for Buruli ulcer might look like, and the insights that this type of approach can bring.
This analysis thus raises additional questions that cannot as yet be answered: What are the biological explanations for the lesions understood as Buruli in this region? If prevalence has reduced, how, and why, has this come about? To answer these questions will require further inquiry that encompasses biological, social, epidemiological, ecological, environmental and historical insights.
|
10.1371/journal.ppat.1005383 | EBNA3C Directs Recruitment of RBPJ (CBF1) to Chromatin during the Process of Gene Repression in EBV Infected B Cells | It is well established that Epstein-Barr virus nuclear antigen 3C (EBNA3C) can act as a potent repressor of gene expression, but little is known about the sequence of events occurring during the repression process. To explore further the role of EBNA3C in gene repression–particularly in relation to histone modifications and cell factors involved–the three host genes previously reported as most robustly repressed by EBNA3C were investigated. COBLL1, a gene of unknown function, is regulated by EBNA3C alone and the two co-regulated disintegrin/metalloproteases, ADAM28 and ADAMDEC1 have been described previously as targets of both EBNA3A and EBNA3C. For the first time, EBNA3C was here shown to be the main regulator of all three genes early after infection of primary B cells. Using various EBV-recombinants, repression over orders of magnitude was seen only when EBNA3C was expressed. Unexpectedly, full repression was not achieved until 30 days after infection. This was accurately reproduced in established LCLs carrying EBV-recombinants conditional for EBNA3C function, demonstrating the utility of the conditional system to replicate events early after infection. Using this system, detailed chromatin immunoprecipitation analysis revealed that the initial repression was associated with loss of activation-associated histone modifications (H3K9ac, H3K27ac and H3K4me3) and was independent of recruitment of polycomb proteins and deposition of the repressive H3K27me3 modification, which were only observed later in repression. Most remarkable, and in contrast to current models of RBPJ in repression, was the observation that this DNA-binding factor accumulated at the EBNA3C-binding sites only when EBNA3C was functional. Transient reporter assays indicated that repression of these genes was dependent on the interaction between EBNA3C and RBPJ. This was confirmed with a novel EBV-recombinant encoding a mutant of EBNA3C unable to bind RBPJ, by showing this virus was incapable of repressing COBLL1 or ADAM28/ADAMDEC1 in newly infected primary B cells.
| The Epstein-Barr nuclear protein EBNA3C is a well-characterised repressor of host gene expression in B cells growth-transformed by EBV. It is also well established that EBNA3C can interact with the cellular factor RBPJ, a DNA-binding factor in the Notch signalling pathway conserved from worms to humans. However, prior to this study, little was known about the role of the interaction between these two proteins during the repression of host genes. We therefore chose three genes–the expression of which is very robustly repressed by EBNA3C –to explore the molecular interactions involved. Hitherto these genes had not been shown to require RBPJ for EBNA3C-mediated repression. We have described the sequence of events during repression and challenge a widely held assumption that if a protein interacts with RBPJ it would be recruited to DNA because of the intrinsic capacity of RBPJ to bind specific sequences. We show that interaction with RBPJ is essential for the repression of all three genes during the infection of B cells by EBV, but that RBPJ itself is only recruited to the genes when EBNA3C is functional. These data suggest an unexpectedly complex interaction of multiple proteins when EBNA3C prevents the expression of cellular genes.
| Epstein-Barr virus (EBV) is a large DNA virus that belongs to the gamma subfamily of herpes viruses and infects persistently >90% of the human population. Infection with EBV is aetiologically associated with several types of human cancer, including Burkitt lymphoma, Hodgkin lymphoma, peripheral natural killer/T-cell lymphoma, nasopharyngeal and gastric carcinoma [1]. Infection of B cells with EBV results in activation and transformation of resting cells into proliferating B blasts, in which the viral genome resides as an extra-chromosomal episome within the nucleus. In vivo, early after infection, all EBV latency-associated genes are expressed, producing six EBV nuclear antigens [EBNA1, 2, 3A, 3B, 3C and leader protein (LP)], three latent membrane proteins (LMP1, 2A and 2B), two small non-coding RNAs (EBER1 and 2) and micro-RNA transcripts from the BamHI A region (BARTs) [1,2]. In vitro, infection of primary resting B cells with EBV creates continuously proliferating lymphoblastoid cell lines (LCL) that constitutively express all latency-associated EBV genes [1].
The genes encoding EBNA3A, 3B and 3C are arranged in tandem in the EBV genome and share the same gene structure with a short 5’ exon and a long 3’ exon. The proteins originate, through alternative splicing, from the B cell specific EBNA2/LP/3A/3B/3C transcription unit resulting in very long mRNAs initiated primarily from the Cp promoter. There are only a few copies of EBNA3 mRNAs in LCLs, probably due to tight transcriptional regulation–for example it has been reported that less than 3 mRNA copies of EBNA3C per cell can be detected [3]–and associated with slow turnover of the proteins [4]. The EBNA3s form a family of transcriptional co-regulators that can cooperate to regulate host gene expression [5–7]. EBNA3 proteins do not bind DNA directly, but are assumed to be tethered to target genes by associating with DNA sequence-binding factors, an example being RBPJ (also known as RBP-jk, CBF1, CSL, Suppressor of Hairless and Lag1) [8–12].
RBPJ is a component of the Notch signalling pathway that was first discovered in Drosophila, but is highly conserved across species and has an important role in developmental processes in embryonic and adult tissue, e.g. cell lineage decisions (reviewed in [13,14]). In vertebrates–in the absence of active Notch signalling–RBPJ represses Notch target genes through interaction with TFIIA and TFIID to prevent transcription [15] and also recruitment of repressor complexes containing histone deacetylase 1 and 2 (HDAC1 and 2), silencing mediator of retinoid and thyroid hormone receptors (SMRT/NcoR), SMRT/HDAC1-associated repressor protein (SHARP/MINT/SPEN), CBF1-interacting co-repressor (CIR), C-terminal binding protein (CtBP), CtBP-interacting protein (CtIP) and KyoT2 [16–20]. Ligand binding to the Notch receptors induces a series of proteolytic cleavages of the receptor resulting in the release of Notch intracellular domain (NICD) from the cell membrane [21–23]. NICD translocates into the nucleus where it binds to RBPJ via its RBPJ-associated molecule (RAM) domain WΦP (Φ = hydrophobic residue) motif (WFP) [24] and via its ankyrin repeats [25–27]. Binding of NICD to RBPJ disrupts the association with repressor complexes [16] and additional binding to the strong co-activator Mastermind [28–30] leads to formation of a stable activating complex and full activation of the repressed Notch signalling target genes.
EBNA2 also binds to RBPJ via a RAM domain WWP motif [31–34]. EBNA2, one of the first viral genes expressed after infection of B cells and a transcriptional transactivator of the other latent viral genes as well as cellular genes, operationally resembles NICD [35]. All the EBNA3s share a highly conserved N-terminal homology domain (HD) that contains RBPJ binding sites [9,11]. EBNA2 and EBNA3s, however, form mutually exclusive complexes through competitive binding to the same binding site on RBPJ [8,36]. EBNA3/RBPJ complexes were shown to disrupt DNA binding of RBPJ in vitro, in electrophoretic mobility shift assays [8,9,37] and to repress EBNA2-mediated activation in transient reporter assays [8,11,37,38]. This repression is dependent on the ability of EBNA3 to bind to RBPJ. Mutation of four core residues within the HD of EBNA3C from 209TFGC to 209AAAA (HDmut) that affect binding to RBPJ produced a protein that did not disrupt RBPJ/DNA binding and that failed to repress EBNA2-mediated transcriptional activation in transient reporter assays [11,39]. The HDmut EBNA3C also failed to sustain LCL proliferation when transfected into LCL with conditional EBNA3C after inactivation of EBNA3C [40,41]. In addition to the earlier identified core 209TFGC motif [11], Calderwood and colleagues more recently identified a RAM-like motif (227WTP) in EBNA3C but not EBNA3A and EBNA3B [42]. The W227S mutant EBNA3C successfully repressed EBNA2-mediated transcriptional activation in transient reporter assays and sustained LCL proliferation in back-complementation assays, however, both 209AAAA and W227S mutations were required for an effective loss of RBPJ binding as determined by co-immunoprecipitation [42].
Originally, a model was proposed in which EBNA2 acts as a viral analogue of NCID, producing transcriptional activation when bound to RBPJ; this is counteracted by competitive binding of EBNA3s to RBPJ and destabilisation of RBPJ binding to DNA [1,43,44]. An alternative model was then proposed in which EBNA3s directly recruit repressors to RBPJ that remains statically bound to its responsive elements [45]. When targeted directly to DNA by fusion with the DNA-binding domain of Gal4, all EBNA3 proteins exhibit strong repressor activity in reporter assays [37,46,47]. Moreover, EBNA3C can interact with cellular factors that are involved in transcriptional repression, these include HDAC1, HDAC2, CtBP, Sin3A and NcoR [48–50]–this would be consistent with the repressor recruitment model. However, it is fair to say that currently the role of RBPJ in gene regulation by EBNA3C remains largely unknown.
From previous microarray analyses in EBNA3A knockout (KO) LCL [5], BL31 infected with EBNA3A or EBNA3C KO viruses [6], BJAB cells stably expressing EBNA3C [51] and EBNA3C-conditional LCL [52] it is known that EBNA3A and EBNA3C repress ADAM28 and ADAMDEC1, two members of a disintegrin and metalloprotease (ADAM) family that are encoded in adjacent genomic loci. McClellan and colleagues identified an intergenic EBNA3 binding site that loops to the transcription start site (TSS) of both genes only in the presence of EBNA3C and repression involved reduced levels of activation-associated H3K9/14ac mark and increased levels of the repressive H3K27me3 mark within ADAM28 and at the TSS of ADAMDEC1 [51,53]. However, these observations were obtained from stable transfectants of EBNA3C in the EBV-negative B cell lymphoma line BJAB in the absence of the other latent viral gene products and nothing is known about the temporal sequence of events at these regulatory sites and TSS, or the factors that are involved in EBNA3C-mediated gene repression early after infection of primary B cells with EBV. In addition to confirming repression of both ADAMs, a microarray study of EBNA3C-conditional LCL identified COBLL1 as the gene most robustly repressed by EBNA3C (S1 Table, [52]). The function of the COBLL1 gene product is unknown and it has not previously been characterised as an EBNA3C repressed gene. Based on the previous studies and the microarrays, we selected ADAM28, ADAMDEC1 and COBLL1 in order to explore in more detail the temporal sequence of events and factors that are involved in EBNA3C-mediated gene regulation.
Here, we show that all three genes are highly repressed in B cells following infection of primary CD19+ cells with EBV, only when EBNA3C is expressed and functional. Using LCLs conditional for EBNA3C function we could show for a first time that this system can be used efficiently to replicate EBNA3C-mediated changes in gene expression very similar to those seen early after infection of primary B cells with EBV. Using the conditional system we were able to explore further the temporal changes in epigenetic marks at regulatory elements and TSSs leading to repression of transcription and showed that it involved two-steps, rapid initial loss of activation-associated histone marks that led to repression of mRNA expression, followed by recruitment of polycomb proteins and increases of repressive histone H3K27me3 mark. Furthermore, we show that RBPJ is only recruited when EBNA3C is functional and that repression is absolutely dependent on the ability of EBNA3C to bind to RBPJ. This is the first time that EBNA3C-mediated transcriptional repression has been described in such detail and it provides novel insights into temporal sequence of events occurring early after infection and the dynamic role of RBPJ in EBV-mediated gene repression.
Interrogation of Affymetrix Exon 1.0 ST microarray analysis from the EBNA3C conditional system (3CHT) indicated that COBLL1, ADAM28 and ADAMDEC1 required expression of functional EBNA3C for very significant levels of repression in LCL (S1 Table; http://www.epstein-barrvirus.org.uk). In order to establish that ADAM28, ADAMDEC1 and COBLL1 are regulated by EBNA3C during viral infection of B cells–and to determine whether EBNA3A and/or EBNA3B are involved in the regulation–primary CD19+ cells were infected with previously characterised wild type (B95.8-BAC) EBV or recombinant EBNA3 KO or revertant (Rev) viruses (Fig 1A–1C) [54]. Infections with wild type, Rev and EBNA3B KO viruses resulted in a reduction of both ADAM28 (2–3 log fold) and ADAMDEC1 (1–2 log fold) levels of mRNA, over a period of 30 days after infection. In the absence of EBNA3C (3CKO) and to a lesser extent EBNA3A (3AKO) there was a failure to repress ADAM28 (Fig 1A) and ADAMDEC1 (Fig 1B)–this is consistent with reports derived from stable cell lines [51]. All infections with EBNA3C competent viruses led to a remarkable 3–4 log fold reduction of COBLL1 mRNA over the same period of time, but this was not seen after infection with 3CKO virus–here the levels detected in primary B cells were maintained (Fig 1C). These differences in gene expression between the various virus infections were not observed for the two control genes ALAS1 (S1A Fig) and GNB2L1 (S1B Fig), neither of which are known to be targets of EBNA3 proteins or EBV.
After establishing that all three genes were robustly repressed after infection of primary B cells with EBV, which confirmed previous findings from stable cell lines that ADAM28 and ADAMDEC1 were repressed by EBNA3C and EBNA3A and identifying that COBLL1 was repressed by EBNA3C alone, we wanted to determine whether it was possible to recapitulate this repression in LCLs carrying EBV-recombinants conditional for EBNA3C (3CHT). In this cell line, EBNA3C activity is conditional on the presence of 4-hydroxytamoxifen (HT) and proliferation of the cells does not decrease in its absence due to the homozygous deletion of p16INK4A, a primary target of EBNA3C [52]. This cell line could therefore be established having never expressed functional EBNA3C (3CHT A13) and the expression of ADAM28, ADAMDEC1 and COBLL1 in this cell line is similar to uninfected primary B cells. Activation of EBNA3C by the addition of HT to these cells (+HT) resulted in a 2-log fold repression of ADAM28 (Fig 2A), ADAMDEC1 (Fig 2B) and a 4-log fold repression of COBLL1 (Fig 2C), which was not seen when EBNA3C was kept inactive (-HT) over this 60-day period. The repression of all three genes was fully reversible. Inactivation of EBNA3C, by washing out HT on day 30, led to an increase in expression of mRNAs corresponding to ADAM28, ADAMDEC1 and COBLL1 up to levels similar to those at the start of the time-course. The observed repression of ADAM28, ADAMDEC1 and COBLL1 in the 3CHT system was a direct consequence of the HT-induced activation of EBNA3C, because adding HT to a non-conditional EBNA3C KO LCL, grown out from the primary B cell infection (see below), did not change the expression levels of any of the three genes (S1C–S1E Fig).
The repression of COBLL1 by EBNA3C could also be observed at the protein level. Over the 3CHT A13 60-day time-course, COBLL1 protein levels quickly disappeared. They were barely detectable three days after activation of EBNA3C (by the addition of HT) and were completely undetectable thereafter. However, COBLL1 protein did not reappear until about 20 days after inactivation of EBNA3C (Fig 2D). In addition, a rare LCL that grew from an infection of primary B cells with EBNA3C KO virus showed that only EBNA3C-deficient LCLs expressed COBLL1 (S2A Fig). Consistent with previous published data [52], cells from the EBNA3C KO virus infection underwent the expected crisis around 3–4 weeks post-infection but in a single experiment a sub-population survived and grew into a stable LCL. Immunoblot analysis revealed–in addition to high levels of COBLL1 –low levels of retinoblastoma protein (Rb), phosphorylated Rb (P-Rb) and loss of p16INK4a in this unusual LCL (S2A Fig)–we and others have seen this type of clonal selection, more frequently, with cells infected with EBNA3A KO virus [5,55]. It should be noted that protein expression data for ADAM28 and ADAMDEC1 was not included because, in our hands, the commercial antibodies that we tested did not produce convincing or reproducible results.
The results obtained from the 3CHT A13 time-course were highly reproducible, not only in the same cell line, but also in 3CHT C19, another 3CHT cell line created by an independent 3CHT recombinant virus clone on the p16-null background, but grown out in the presence of HT and then washed (S1F–S1H Fig). The dynamic range of COBLL1 repression was remarkable, following a similar highly exponential repression profile in both 3CHT A13 and C19 conditional cell lines and also in newly infected primary B cells (S3A Fig). Repression of ADAM28 and ADAMDEC1 was rather more variable between cell lines, but showed a similar exponential repression profile (S3B and S3C Fig). Taken together these results showed for the first time, that the EBNA3C conditional cell lines could be used to recapitulate efficiently EBNA3C-mediated gene repression observed in B cells early after infection with EBV.
Having shown that the EBNA3C conditional system could be efficiently used to replicate gene expression changes seen early after EBV infection of primary B cells, we wanted to determine the sequence of events that led to such robust repression of COBLL1 and the ADAM28-ADAMDEC1 locus by employing chromatin immunoprecipitations (ChIP) on samples harvested throughout the 3CHT A13 time-course. COBLL1 is located on chromosome 2, has multiple transcript variants and a CpG island around the promoter region. ChIP coupled to high throughput DNA sequencing (ChIP-seq) using LCLs expressing tandem-affinity purification (TAP) tagged EBNA3s ([56] and K. Paschos et al., manuscript in preparation) identified a single intragenic EBNA3A and EBNA3C binding site, hereafter called the COBLL1 peak (Fig 3A). This binding site was confirmed in the LCLs used here by ChIP-qPCR (S4 Fig). ChIP analysis on samples from the 3CHT A13 time-course and probed across the COBLL1 locus, showed that after activation of conditional EBNA3C, there was a sustained decrease in the activation-associated histone marks H3K9ac, H3K27ac and H3K4me3 and an increase in the repressive histone mark H3K27me3 primarily at the TSS of COBLL1 (Fig 3B–3E). These changes were not observed at GAPDH or Myoglobin, two controls for expressed and repressed genes, respectively. Next, since H3K27me3 is catalysed by polycomb repressive complex 2 (PRC2) and polycomb complexes were previously found to be involved in the repression of BCL2L11 (BIM) [57] and CDKN2A (p16INK4A) [55] we explored whether they also play a role in the repression of COBLL1. ChIP for PRC2-family member SUZ12 showed increased enrichment at the TSS of COBLL1 as it is repressed (Fig 3F). In contrast and rather unexpected, ChIP for the polycomb repressive complex 1 (PRC1) family member BMI1 revealed recruitment of BMI1 to the COBLL1 peak, but not the TSS, with highest BMI1 levels nine days after EBNA3C activation (Fig 3G). Finally, since the EBNA3 proteins cannot bind directly to DNA and RBPJ is the most well characterised DNA binding factor to which they have all been reported to bind, ChIP for RBPJ was performed. This revealed that RBPJ accumulated on the COBLL1 peak, but only when EBNA3C was functional–with the highest levels appearing six days after activation of EBNA3C by HT (Fig 3H).
ADAM28 and ADAMDEC1 are encoded on chromosome 8 and also have multiple transcript variants, but no distinct CpG island. Previous studies by McClellan and colleagues [51] identified an intergenic EBNA3A and EBNA3C binding site (hereafter called the ADAM peak), which was also detected in our ChIP-seq performed on EBNA3A-TAP and EBNA3C-TAP LCLs (Fig 4A) and confirmed by ChIP-qPCR (S4 Fig). ChIP for activation associated histone marks on samples of the 3CHT A13 time-course and across the ADAM locus again showed a loss of H3K9ac, H3K27ac and H3K4me3 largely at the TSS of ADAMDEC1, but also ADAM28, when EBNA3C was made functional by the addition of HT (Fig 4B–4D). There was an increase in the repressive H3K27me3 mark across the ADAM28-ADAMDEC1 locus, but at considerably lower levels than seen at the COBLL1 TSS (Fig 4E). This is consistent with previous data from McClellan and colleagues that showed less H3K9/14ac and increased H3K27me3 in a stable EBNA3C expressing BJAB cell line [51]. Unlike at the COBLL1 locus, no increase in SUZ12 enrichment could be detected across the ADAM28-ADAMDEC1 locus (Fig 4F). However, similar to COBLL1, ChIP for BMI1 revealed recruitment to the ADAM peak (but again not to either TSS) with highest levels nine days after EBNA3C activation (Fig 4G). Although repression of this locus had not been previously described as being RBPJ-dependent, as with COBLL1, RBPJ enrichment also increased at the ADAM peak only when functional EBNA3C was induced, with highest levels appearing six days after activation (Fig 4H).
In order to confirm these temporal changes of histone marks and factor recruitment at both the COBLL1 and ADAM28-ADAMDEC1 loci, ChIP samples taken during a biological replicate time-course–using the 3CHT C19 LCL–were analysed in a similar way to the 3CHT A13 samples and showed very similar results (S5 and S6 Figs). Unfortunately, we were unable to reproducibly perform ChIP for EBNA3C using a commercial polyclonal antibody against EBNA3C that also precipitates EBNA3A and EBNA3B [53]. In order to reliably ChIP for EBNA3C during an extended time-course experiment, the conditional EBNA3C would also need to be TAP-tagged, but these viruses are not currently available.
Immunoblot analysis showed that there were no consistent changes to the levels of EBNA3A, EBNA3B, BMI1, SUZ12 and RBPJ proteins during either the 3CHT A13 or 3CHT C19 time-courses (S2B and S2C Fig).
The similarity of the two time-course experiments allowed a more detailed analysis of the temporal development of histone marks and factor recruitment to regulatory elements. For this, ChIP enrichment levels relative to input from both 3CHT A13 and 3CHT C19 time-courses were normalised. Activation-associated histone marks were expressed relative to the first time point–HT (day 3) and repressive histone marks relative to the last time point +HT (day 30). This revealed that at the TSS of all three genes loss of the activation-associated histone marks H3K9ac, H3K27ac and H3K4me3 preceded any increase in the repressive histone mark H3K27me3 (Fig 5A–5C top). Furthermore, comparison between the changes in histone marks and corresponding mRNA expression levels (Fig 5A–5C bottom) of each gene over time revealed that the initial repression of mRNA expression was caused by loss of all three activation-associated histone modifications (H3K9ac, H3K27ac and H3K4me3) and was independent of the appearance of the repressive H3K27me3 modification, which was only deposited after most mRNA was depleted.
At the TSS of COBLL1, maximal SUZ12 enrichment levels were reached by day nine, which precedes the substantial increase in H3K27me3 at this site about 15 days after activation of EBNA3C (Fig 6A). Analysis of the BMI1 recruitment to both ADAM peak and COBLL1 peak showed that maximal BMI1 levels were reached nine days after activation of EBNA3C (Fig 6B). However, these high BMI1 levels were not maintained and relative enrichment levels of BMI1 subsequently dropped, but they remained significantly higher than in cells where EBNA3C was kept inactive. Interestingly, comparing the recruitment profile of SUZ12 to the TSS of COBLL1 with that of BMI1 to the COBLL1 peak, it appeared that both increased simultaneously at the distinct sites, but in contrast to BMI1, SUZ12 levels were maintained at a high level for at least 60 days (Fig 3F). Analysis of the accumulation of RBPJ at both the ADAM and COBLL1 peaks revealed again what appeared to be a transient recruitment of RBPJ with maximal enrichment 3–6 days after the addition of HT (Fig 6C). This preceded the recruitment of BMI1 to the two EBNA3-binding sites.
Taken together, these analyses showed consistently that activation-associated histone marks were removed first, consistent with this being the main cause for repression of mRNA expression, before repressive marks were established. Furthermore, the dynamic recruitment of both RBPJ and BMI1 was dependent on functional EBNA3C, with RBPJ recruitment probably preceding BMI1 and EBNA3C involved in recruiting both.
We were interested to see whether it was possible to recapitulate the repression of COBLL1 and ADAM28 in transient reporter assays. Therefore, initially, the promoter region of COBLL1 was cloned upstream of a luciferase cassette either in the presence or absence of the COBLL1 peak inserted downstream of luciferase (Fig 7A). The ideal B cell line for transient reporter assays is the readily transfectable EBV-negative Burkitt’s lymphoma cell line DG75 [58]. However, luciferase activity of the COBLL1 construct was very low in this cell line, which made it impossible to study repression, perhaps because DG75 cells do not express endogenous COBLL1. Therefore, a panel of transfectable B cell lines was screened for luciferase activity of the COBLL1 construct and it was found to be robust in the EBV-positive, but EBNA3C-null cell line Raji [59]. This expresses endogenous COBLL1. The presence of COBLL1 peak in the plasmid led to an increase in luciferase activity (~10 fold) relative to the construct that only has the promoter region of COBLL1 –indicating that, in the absence of EBNA3C, the COBLL1 peak acts as an enhancer (Fig 7B). Co-transfection of expression plasmids for the EBNA3s along with the luciferase constructs that contain the COBLL1 peak showed that EBNA3C repressed luciferase activity, but neither EBNA3A nor EBNA3B had this effect (Fig 7C). This was consistent with the results from the primary B cell infections presented above, confirming EBNA3C as sole repressor of COBLL1. Co-transfection of the COBLL1 reporter with an expression plasmid encoding a mutant EBNA3C that is unable to bind to RBPJ based on the double mutant described in Calderwood et al. (see Introduction and Materials and Methods), failed to repress luciferase activity (Fig 7D).
In order to recapitulate the repression of ADAM28 a similar approach was used in the EBV-negative Burkitt’s lymphoma cell line DG75 [58], here luciferase activity of the ADAM28 construct was robust (Fig 8A). Again the presence of the ADAM peak included downstream of the luciferase gene led to an increase in luciferase activity, but this was not as substantial as for COBLL1 (<3 fold) (Fig 8B). For this construct, co-transfection of plasmids expressing either EBNA3A or EBNA3C, but not those expressing EBNA3B, resulted in a reduction in luciferase activity (Fig 8C)–again consistent with results from the primary B cell infection experiments. As for the COBLL1 reporters, the repression of the ADAM reporter was dependent on the presence of RBPJ, since neither EBNA3A nor EBNA3C induced a reduction in luciferase activity in RBPJ-null DG75 cells (SM224.9 [60], Figs 8D and S2D). Moreover, co-transfection of the ADAM reporter with the plasmid expressing the EBNA3C RBPJ-binding mutant also failed to repress luciferase activity (Fig 8E).
These results showed that transient reporter assays can be used to recapitulate the repression of COBLL1 by EBNA3C and the repression of ADAM28 by both EBNA3A and EBNA3C and that the ability of EBNA3s to bind and recruit RBPJ was likely to be important for the repression of both loci in the context of latent EBV infection.
In order to determine whether binding of EBNA3C to RBPJ is necessary for the repression of the endogenous COBLL1 and ADAM28-ADAMDEC1 locus in the context of infection, a new EBV recombinant encoding the RBPJ binding mutant of EBNA3C (RBPJ BM EBNA3C) was constructed. The RBPJ BM EBNA3C was based on the double mutant described by Calderwood and colleagues [42] and comprised the newly identified W227S mutation and the previously identified mutation of residues 209TFGC→AAAA ([11], see Introduction and Fig 9A). NotI and SalI restriction sites were introduced in order to allow restriction digest confirmation of successful mutagenesis (S7A Fig) and rescued BACs from HEK293 virus producing clones (S7B Fig) that maintained general BAC integrity. DNA sequencing of rescued episomes confirmed the mutations that had been engineered.
Infection of primary CD19+ cells with this RBPJ BM EBNA3C-recombinant virus resulted in outgrowth and establishment of an LCL. This was unexpected because previous back-complementation experiments using similar RBPJ BM EBNA3C transfected into LCL with conditional EBNA3C, failed to rescue LCL proliferation after inactivation of the conditional EBNA3C [40,42]. Cell proliferation, measured by the incorporation of thymidine analogue EdU 36 days after primary B cell infection, showed that 22.9% of RBPJ BM EBNA3C cells were synthesising DNA, which is double the 11.7% of cells infected with EBNA3C KO virus, but considerably less than 55% of cells infected with wild type or revertant viruses (Fig 9B). Immunoblot analysis of the established RBPJ BM EBNA3C LCL 56 days after primary B cell infection showed similar EBNA3A, EBNA3B, EBNA3C, EBNALP and RBPJ levels compared with wild type or EBNA3C revertant LCLs (Fig 9C). EBNA2 expression, and probably as a consequence LMP1 expression, appeared to be significantly increased in the RBPJ BM EBNA3C LCL, even in comparison to EBNA3C knockout LCL. This suggests that the ability of EBNA3C to interact with RBPJ is important for the regulation of viral genes in the context of infection. The inability of RBPJ BM EBNA3C to bind to RBPJ was confirmed by immunoprecipitation of RBPJ from these LCLs. This very efficiently pulled down wild type EBNA3C, but only trace amounts of RBPJ BM EBNA3C (Fig 9D). RBPJ was immunoprecipitated efficiently from both LCLs.
Finally, in order to determine whether binding of EBNA3C to RBPJ was necessary for the repression of the endogenous COBLL1 and ADAM28-ADAMDEC1 locus, RNA samples taken every 5 days from the time of infection of primary B cells with the recombinant RBPJ BM EBNA3C virus were analysed. Consistent with the results of the luciferase reporter assays and the ChIP studies, infection with the RBPJ BM EBNA3C virus was unable to repress ADAM28, ADAMDEC1 or COBLL1 (Fig 10A–10C). Expression levels of all three genes were similar compared to changes seen after infection with EBNA3C KO virus, whereas infection with EBNA3C revertant or wild type viruses resulted in robust repression of all three genes as seen in the previous primary B cell infections (Fig 1). As before, these differences in gene expression between the various viruses were not seen for the control gene ALAS1 (S1I Fig).
In conclusion, these results showed that the ability of EBNA3C to interact with RBPJ is not only essential for repression in transient reporter assays, but also for the repression of the endogenous COBLL1 and ADAM28-ADAMDEC1 locus in the context of viral infection.
Although it is well established that EBNA3C is essential for transformation of normal primary B cells and for repression of host tumour suppressor genes (e.g. BCL2L11 and CDKN2A), the precise molecular mechanisms by which EBNA3C regulates gene expression remain largely unknown. Here, by using two genomic loci very robustly repressed by EBNA3C, we have explored some of the molecular interactions involved in EBNA3C-mediated gene repression. This has produced new insights into the temporal sequence of events during the repression and challenges existing models based on DNA-binding transcription factors remaining relatively static on chromatin.
Most surprising was the dynamic recruitment of and/or stabilisation of what we take to be RBPJ/EBNA3C complexes on both COBLL1 and ADAM peaks. This is in contrast to the paradigm that RBPJ is stably bound to its DNA recognition sequences and that, in the absence of Notch signalling, recruits repressors that are replaced by activators upon signalling [13,61]. Furthermore, it is in contrast to previous reports that suggest EBNA3C disrupts RBPJ binding to DNA in order to prevent transactivation by EBNA2 [8,9,37]. It is tempting to speculate that interaction of EBNA3C with RBPJ increases the binding to–or stabilises RBPJ/EBNA3C complexes on–regulatory elements that control expression of these EBNA3C target genes. Recent studies in Drosophila [62] and mammalian cells [63] have revealed that activation of Notch signalling can induce de novo binding or increased binding of RBPJ mainly at regulatory elements. A similar observation was made for some EBNA2 target genes, where EBNA2 expression appeared to increase the occupancy of RBPJ at these genes during activation [64]. However, increased binding of RBPJ has not, to our knowledge, been reported during gene repression. It might be that EBNA3C, through interaction with one or more other transcription factors (S8 and S9 Figs), increases the on-rate in the dynamic equilibrium of RBPJ binding to its recognition sites. Using publically available transcription factor binding prediction software (PROMO [65,66] and Patch 1.0 [BIOBASE]), it was not possible to identify any strictly canonical RBPJ binding sites within 1kb around either the ADAM peak or the COBLL1 peak, but if a more relaxed interpretation was used, several possible sites were found at both EBNA3C peaks. It is not possible to determine which, if any, of these are responsible for targeting RBPJ and it is more than likely that other transcription factors are also involved. Alternatively a combination of EBNA3C and some other factor could redirect RBPJ to cryptic sites. Previous studies found that EBNA3 binding sites coincided with various transcription factors, e.g. BATF, BCL11A, IRF4, PAX5 and RUNX3 [53,67,68], all of which seem to co-occupy the EBNA3 binding sites at COBLL1 and ADAM28/ADAMDEC1 (S8B and S9B Figs). Any of these could act as a co-factor in directing RBPJ/EBNA3C complexes to these particular loci.
This raises the question of precisely what role RBPJ plays in repression here, or more generally in the context of infections with other gamma-herpesviruses such as KSHV [43,44,69]. Our best guess is that the interaction between RBPJ and EBNA3C is needed for the assembly of a multi-protein platform of co-repressors (see Introduction and below) that is unable to efficiently assemble in the absence of either RBPJ or EBNA3C. It is possible that the assembly of these multi-protein complexes can mask the epitope detected by the anti-RBPJ antibody in ChIP experiments, which might be an explanation for what appeared to be lower RBPJ occupancy at ADAM peak and COBLL1 peak at later time-points after activation of EBNA3C. Recently RBPJ has been found to be retained on mitotic chromatin–book-marking the transcriptional state of genes through cell division–and also to interact with CTCF, which might be involved in formation of higher-order chromosome structures [70]. We cannot exclude either of these functions being important here.
The functional importance of RBPJ for the repression of ADAM28 and COBLL1 was indicated in transient reporter assays using RBPJ BM EBNA3C and RBPJ knockout cells. In these assays EBNA3C binding appeared to convert enhancer-like elements into repressor elements. The magnitude of repression in these transient reporter assays was far less than the repression seen in the context of viral infection and the host chromosomes. Consequently we constructed the RBPJ BM EBNA3C recombinant virus–based on the most recent assessment of RBPJ binding sites in EBNA3C. The primary B cell infection with this virus unequivocally demonstrated that the RBPJ BM EBNA3C virus is completely unable to repress ADAM28, ADAMDEC1 and COBLL1, providing compelling evidence that the EBNA3C:RBPJ interaction is essential in the context of viral infection. The only caveat here is that we cannot formally exclude the possibility that these two mutations in EBNA3C alter other yet to be described interactions required for gene repression. However, the failure of wild type EBNA3C to repress ADAM28 luciferase constructs in the RBPJ-null DG75 cells also showed that presence of RBPJ is essential for repression. The RBPJ BM EBNA3C virus was able to establish a stable LCL (in culture >2 months) that, although it proliferated slowly, was in contrast to the previous back-complementation studies that suggested the interaction was essential to rescue LCLs when EBNA3C was inactivated [40–42]. Further studies are required to characterise the importance of the EBNA3C:RBPJ interaction and to determine which other factors are required for the dynamic binding of these complexes to specific regulatory elements in response to EBNA3C.
The repression of ADAM28/ADAMDEC1 and COBLL1 was highly reproducible and very similar between primary B cell infection and EBNA3C conditional LCL after activation of EBNA3C, which further validated the full functionality of the 3CHT fusion proteins. The kinetics followed a highly exponential repression profile over the first two weeks after infection or activation of EBNA3C (S3 Fig), but the fully repressed state was only achieved after 30 days or even later. Full de-repression, through inactivation of EBNA3C, required a similar time period. Currently, we do not understand why these changes in gene expression take place over such a long period of time. At least for COBLL1 the initial repression was relatively rapid, with a 10-fold drop in gene expression by day three and a complete disappearance of COBLL1 protein in immunoblot by day six. However, reactivation of COBLL1 expression and reappearance of COBLL1 protein from the fully repressed state took at least 20 days after inactivation of EBNA3C. One possible explanation is that it takes that long for repressive histone marks to be fully established during repression or fully removed during reactivation.
The comparison between histone marks and mRNA expression suggested that the initial repression of both loci is independent of polycomb protein recruitment, but requires removal of activation-associated histone acetylation and the H3K4me3 mark. There are 18 human proteins with deacetylase activity that are grouped into four families according to their homology (HDAC family 1–4) [71]. EBNA3C has been shown to bind to and recruit the class one HDAC family members 1 and 2, which makes both of them likely candidates involved in the initial repression [48,49]. There are more than 30 proteins in the Jumonji C family of demethylases that are able to remove mono-, di- or trimethylation on lysine residues and at least four of them, KDM5A/B/C/D, have been shown to be able to catalyse the removal of H3K4me3 [72–77]. Multiprotein complexes composed of both HDAC1 and HDAC2 together with KDM5A/Sin3 or KDM5C/NcoR/REST have been identified [76,78,79]. Besides HDAC1 and HDAC2, EBNA3C can bind to both Sin3, NcoR and CtBP [49,50], so it seems likely that EBNA3C recruits one of these multifunctional complexes to remove histone acetylation and H3K4me3 in the initial phase of repression. Furthermore RBPJ can independently recruit similar complexes of repressors (see Introduction), adding further support to the idea that by physically interacting EBNA3C and RBPJ synergise in the early phase of repression.
Following the initial repression, PRC1 and PRC2 were recruited probably to maintain or further extend the repressive state by depositing H3K27me3. In more recent studies the classical sequential recruitment model in which PRC2-induced modifications recruit PRC1 has been challenged. It has been shown that PRC1 can be recruited independently from PRC2 and the H3K27me3 modification [80–83]. Furthermore, PRC1 can actually recruit PRC2 through the deposition of H2AK119Ub [84–86]. In our study, however, the ChIP analysis revealed that–in contrast to the classical and more recent models of polycomb repressive complex recruitment–it appeared that PRC1 and PRC2 complexes were recruited not only in the same time frame, but also to different genomic loci. BMI1, as part of PRC1, was found at the EBNA3C binding site located in the regulatory elements of ADAM28/ADAMDEC1 and COBLL1, whereas the PRC2 subunit SUZ12 was found at the TSS of COBLL1. No direct SUZ12 recruitment could be detected at various sites across the ADAM28/ADAMDEC1 locus although H3K27me3 levels increased at these sites, albeit to much lower levels than at the TSS of COBLL1. Recruitment of SUZ12 to a discrete site might not have been detected because of the choice of primers, this is however unlikely, because a previous study in human embryonic stem cells identified that 95% of SUZ12 binding sites localised within 1kb of TSS and 40% were within 1kb of CpG islands [87]. Further studies verified this and showed that CpG islands could recruit PRC2 and led to the establishment of H3K27me3 [88–90]. The TSS of both ADAM28 and ADAMDEC1 were included in the ChIP analysis. Furthermore, there is a CpG island around the TSS of COBLL1, but not at the ADAM28/ADAMDEC1 locus. This might explain the direct recruitment of SUZ12 and the much higher H3K27me3 levels at the TSS of COBLL1 relative to the ADAM28/ADAMDEC1 locus. It was very surprising that SUZ12 and BMI1 were recruited to two distinct sites at COBLL1. Polycomb complexes have been shown to mediate the formation of higher-order chromosome structures [91–93] (reviewed in [94]). So perhaps chromatin looping between the COBLL1 peak and the TSS of COBLL1 would bring BMI1 (PRC1) and SUZ12 (PRC2) together. We attempted chromosome conformation capture to analyse this locus but we could not reproducibly show looping between COBLL1 peak and the TSS. We do not know the reason for this, however, the same technique has been successfully used to show repressive loop formation between ADAM peak and the TSS of ADAM28 and ADAMDEC1 when EBNA3C was expressed [53] and we have used it to show looping between a distal enhancer and TSS during EBNA3A/3C-mediated activation of a micro-RNA cluster [56].
It is currently unclear whether BMI1 and SUZ12 are recruited by direct interaction with EBNA3C or if this represents a default mechanism of gene repression at these two genomic loci once activation-associated histone marks are removed. A similar two-step model has been proposed recently for the EBNA3A-mediated repression of CXCL9 and CXCL10 [45]. At the CXCL9/CXCL10 locus, EBNA3A binding to the regulatory sites displaced EBNA2 resulting in an initial state of repression (or de-activation), which was subsequently maintained or further extended by the recruitment of polycomb proteins. This is very similar to what we have observed, however, no occupancy of EBNA2 has been reported on the ADAM or COBLL1 peaks (see below). Furthermore, it is currently unclear why EBNA3A plays only a subsidiary role in the regulation of ADAM28 and ADAMDEC1 compared to EBNA3C, or why EBNA3A is found on the COBLL1 peak (S4 Fig) even though in the primary B cell infection the presence of EBNA3A is clearly not required in the repression of COBLL1. This is unlikely to be an artefact of our TAP-tagged LCL cell lines, because two recently published ChIP-seq experiments using anti-HA antibody in EBNA3A-HA [68] or EBNA3C-HA [67] LCL obtained similar results and detected both EBNA3A and EBNA3C at ADAM and COBLL1 peaks (S8A and S9A Figs). Furthermore, confirming our RBPJ ChIP-qPCR results, ChIP-seq for RBPJ in the IB4 LCL [95] revealed steady-state occupancy of RBPJ on both sites (S8A and S9A Figs). Similarly, ChIP-seq in latency III expressing BL cell line MutuIII using the sheep polyclonal anti-EBNA3C antibody, which also precipitates EBNA3A and EBNA3B, identified the same peaks at both loci, but EBNA2 ChIP-seq revealed no binding at these loci ([53], S8B and S9B Figs).
It is worth noting that RP11-624C23.1 –number four in the list of EBNA3C repressed genes found in the microarray transcriptome analysis (S1 Table)–is a long non-coding RNA with four isoforms of different lengths that run across the ADAM28/ADAMDEC1 locus on the negative strand (S9B Fig). Together with ADAM28 and ADAMDEC1, RP11-624C23.1 is repressed by EBNA3C (http://www.epstein-barrvirus.org.uk), again consistent with the whole locus being co-ordinately regulated.
In summary, a detailed analysis of the mechanisms involved in EBNA3C-mediated gene repression of two genomic loci has provided novel insights into the temporal sequence of events during the repression of transcription and the dynamics of factor recruitment. First, it seems that histone marks associated with activation are removed in an initial step of the repression before repressive marks are deposited. Second, the sequential model of polycomb recruitment was not observed, but rather both PRC1 and PRC2 appeared to be recruited at the same time, but to different sites. Third, the paradigm that RBPJ is stably bound on DNA will need to be reassessed to accommodate a more dynamic recruitment and/or stabilisation model of RBPJ/EBNA3C complexes in repression, as has been proposed for RBPJ-mediated activation [62–64]. Although it is clearly essential, the role of RBPJ in EBNA3C-mediated repression examined here has still to be defined. Finally, the EBNA3C mutant incapable of binding RBPJ was able to sustain cell proliferation and to establish a stable LCL, suggesting a robust interaction between EBNA3C and RBPJ is not an absolute requirement for B cell proliferation.
Recombinant EBV KOs, Revs and wild type have been described previously [54]. Virus production and B cell isolation were performed as previously described [52,55]. Primary B cells were isolated from anonymous buffy coat residues (UK Blood Transfusion Service). B cell purity was assessed to be >90% CD20+ using anti-CD20-APC (eBioscience) staining and flow cytometric analysis. Three million primary B cells in 1.5 ml were infected with 0.5 ml of supernatant containing a total of 1–3x105 Raji green units [55]. RNA from three million uninfected primary B cells was taken on the day of infection and infected cells were incubated in RPMI 1640 (Life Technologies) supplemented with 15% foetal bovine serum, penicillin, and streptomycin at 37°C and 5% CO2. For a period of 30 days, every three days—or five days for the second primary B cell infection with RBPJ binding mutant virus (see below) - 0.5 ml of cells were harvested for RNA extraction and replaced by fresh medium. After this, cells were, where possible, grown into LCL and analysed by immunoblotting.
All cells were cultured in RPMI 1640 medium (Life Technologies) supplemented with 10% foetal bovine serum, penicillin, and streptomycin either in absence or presence of 400nM 4-hydroxytamoxifen (HT, Sigma) at 37°C and 10% CO2. For the time-course experiments EBNA3C conditional LCLs (3CHT) on the p16-null background were used. These have been described previously [52,55]. 3CHT A13 (established in the absence of HT) were used in a time-course experiment over 60 days with samples taken for RNA, protein and ChIP every three days over the first 30 days and every ten days until day 60. Cells were counted and diluted to 3x105 cells/ml at every time-point until day 30 and three times a week subsequently, but seeded at 3x105 cells/ml the day prior to harvesting. After harvesting cells on day 30, half of the +HT culture was centrifuged and the medium replaced by fresh medium without HT and cultured without HT until day 60 (washed). 3CHT C19 (established in the presence of HT, washed and subsequently grown without HT in the medium for more than three months before the experiment was started) were used in a time-course experiment over 30 days. Cells were counted and split to 3x105 cells/ml the day before harvesting samples for RNA, protein and ChIP.
RT-qPCR was performed essentially as described previously [57]. RNA from 4.5x106 cells was extracted using the RNeasy mini kit (Qiagen) and 10ng of cDNA was used for each qPCR reaction. GAPDH or GNB2L1 were used as housekeeping genes as indicated and gene expression was expressed relative to primary B cells or LCL -HT on d0. The sequences of the primers used in this study are listed in S2 Table.
Immunoblotting was performed essentially as described previously [54,55,57]. A total amount of 30 μg of RIPA protein extract was separated on 12, 10 or 7.5% SDS-PAGE, as appropriate, using a mini-PROTEAN II cell (BioRad) and transferred onto Protran nitrocellulose membrane. Antibodies used in this study are listed in S3 Table.
ChIPs for histone modifications and SUZ12 were performed as described previously [57]. For anti-Flag, BMI1 and RBPJ ChIPs, 4.5x106 cells were incubated for 20 min in 1 ml swelling buffer (25 mM HEPES pH 7.8, 1.5 mM MgCl2, 10 mM KCl, 0.1% NP-40, 1 mM DTT, 1 mM PMSF, 1 μg/ml aprotinin and 1 μg/ml pepstatin A). Nuclei were resuspended in 1 ml sonication buffer (50 mM HEPES pH 7.8, 140 mM NaCl, 1 mM EDTA, 1% Triton-X-100, 0.1% sodium deoxycholate, 0.1% SDS, 1 mM PMSF, 1 μg/ml aprotinin and 1 μg/ml pepstatin A) and sonicated for one hour using a Covaris M220 (75 W peak power, 26 duty cycle, 200 cycles/burst and 6°C set temperature). Thereafter, the ChIP assay kit from Millipore (17–295) was used, according to the manufacturer’s protocol. DNA was cleaned using QIAquick PCR purification Kit (Qiagen) and was assayed by qPCR on QuantStudio 7 Flex (Life technologies). Input DNA was 5% of DNA used in immunoprecipitations and diluted to 2.5% prior to PCR quantification. Enrichment relative to input was calculated using four 5-fold-dilution series and error bars calculated as standard deviations from triplicate PCR reactions for both input and IP. Antibodies used are listed in S3 Table and sequences of the primers used in these assays are listed in S4 Table.
Genomic DNA extracted from GM12878 LCL cells using Blood & Cell Culture DNA Midi kit (Qiagen) was used to PCR amplify the promoter region 1 kb upstream of ADAM28 and COBLL1 (short transcripts), the ADAM peak (1 kb around the EBNA3 binding peak at ADAM28-ADAMDEC1) or the COBLL1 peak (1.5 kb around the EBNA3 binding peak at COBLL1) (Primers are listed in S5 Table). The promoter regions were cloned upstream of the luciferase gene in pGL3-basic vector using the MluI restriction site. ADAM peak and COBLL1 peak were cloned downstream of the luciferase gene using the SalI restriction site. All vectors were screened for the correct orientation and were sequence verified. DG75 (for ADAM28 constructs) or Raji cells (for COBLL1 constructs) were electroporated with 1 μg of pGL3-luciferase vectors, 1 μg pSV-beta-galactosidase and varying amounts of pCDNA3-EBNA3 expression plasmids. Total amounts of DNA were balanced using an empty pCDNA3 expression plasmid. Electroporations were performed as described previously for DG75 [96]. For Raji a voltage of 240V was used and all electroporations were harvested after 48h. Luciferase and beta-galactosidase assays were performed as described previously [96] and measured on FLUOstar Omega (BMG Labtech). Beta-galactosidase activity was used to normalise luciferase activities for transfection efficiencies.
For the creation of the RBPJ binding mutant (BM) EBNA3C recombinant virus, the N terminus of EBNA3C was cloned from the B95.8 EBV-BAC [97] by XbaI digestion at position 98,398 and BglII at site 99,749 (relative to GenBank entry V01555.2) into a modified pBlueScrit II SK+. The two in previous studies identified RBPJ binding site of EBN3C were mutated to generate the RBPJ binding mutant EBNA3C. In-Fusion PCR mutagenesis (Clonetech) was employed to first substitute EBNA3C residues 209TFGC for 209AAAA while introducing a NotI recognition site (forward primer: 5’-GCGGCCGCAGCTCAAAATGCGGCACGAACT-3’, reverse primer: 5’-AGCTGCGGCCGCGGCAGTTAACATGATGCTGT-3’). PCR products were purified (Diffinity RapidTip2) and circularised using In-Fusion cloning (Clonetech). Plasmid DNA was screened by NotI restriction digest before introducing the W227S mutation together with a SalI recognition site (forward primer: 5’-GCCACCGTGTCGACACCACCCCATGCTGGACCAA-3’, reverse primer: 5’-CGACACGGTGGCAGAGAAGGTGT-3’). The RBPJ binding mutant fragment of EBNA3C was subcloned into the shuttle plasmid pKovKanΔCm [98] and verified by DNA sequencing. The recombinant EBV was created by RecA based homologous recombination between the B95.8 EBV-BAC and the shuttle plasmid as previously described [98]. At each stage of recombineering BAC DNA was isolated and validated by restriction digest and pulsed-field gel electrophoresis. RBPJ binding mutant EBNA3C virus producing 293 cell clones were established as previously described [54]. Episome rescue of EBV BACs from 293 producing cell lines was performed as previously described for low molecular weight DNA [99].
IPs were performed essentially as described previously [96]. Briefly, RBPJ was immunoprecipitated from protein extracts of 107 wild type EBNA3C or RBPJ BM EBNA3C LCLs for two hours at 4°C using RBPJ rat monoclonal antibody 1F1. Then, 30 μl of protein G-Sepharose beads were added and incubated under rotation for 1h at 4°C, washed four times in IP buffer and immunoprecipitated proteins were resolved by SDS-PAGE and probed for EBNA3C (A10) or RBPJ (ab25949).
A cell proliferation assay–based on measuring the incorporation of EdU at day 36 after primary B cell infection–was performed as described previously for established cell lines [52].
|
10.1371/journal.pntd.0005591 | An individual-level meta-analysis assessing the impact of community-level sanitation access on child stunting, anemia, and diarrhea: Evidence from DHS and MICS surveys | A lack of access to sanitation is an important risk factor child health, facilitating fecal-oral transmission of pathogens including soil-transmitted helminthes and various causes of diarrheal disease. We conducted a meta-analysis of cross-sectional surveys to determine the impact that community-level sanitation access has on child health for children with and without household sanitation access.
Using 301 two-stage demographic health surveys and multiple indicator cluster surveys conducted between 1990 and 2015 we calculated the sanitation access in the community as the proportion of households in the sampled cluster that had household access to any type of sanitation facility. We then conducted exact matching of children based on various predictors of living in a community with high access to sanitation. Using logistic regression with the matched group as a random intercept we examined the association between the child health outcomes of stunted growth, any anemia, moderate or severe anemia, and diarrhea in the previous two weeks and the exposure of living in a community with varying degrees of community-level sanitation access. For children with household-level sanitation access, living in a community with 100% sanitation access was associated with lowered odds of stunting (adjusted odds ratio [AOR] = 0.97, 95%; confidence interval (CI) = 0.94–1.00; n = 14,153 matched groups, 1,175,167 children), any anemia (AOR = 0.73; 95% CI = 0.67–0.78; n = 5,319 matched groups, 299,033 children), moderate or severe anemia (AOR = 0.72, 95% CI = 0.68–0.77; n = 5,319 matched groups, 299,033 children) and diarrhea (AOR = 0.94; 95% CI = 0.91–0.97); n = 16,379 matched groups, 1,603,731 children) compared to living in a community with < 30% sanitation access. For children without household-level sanitation access, living in communities with 0% sanitation access was associated with higher odds of stunting (AOR = 1.04, 95% CI = 1.02–1.06; n = 14,153 matched groups, 1,175,167 children), any anemia (AOR = 1.05, 95% CI = 1.00–1.09; n = 5,319 matched groups, 299,033 children), moderate or severe anemia (AOR = 1.04, 95% CI = 1.00–1.09; n = 5,319 matched groups, 299,033 children) but not diarrhea (AOR = 1.00, 95% CI = 0.98–1.02; n = 16,379 matched groups, 1,603,731 children) compared to children without household-level sanitation access living in communities with 1–30% sanitation access.
Community-level sanitation access is associated with improved child health outcomes independent of household-level sanitation access. The proportion of children living in communities with 100% sanitation access throughout the world is appallingly low. Ensuring sanitation access to all by 2030 will greatly improve child health.
| A lack of access to a sanitation facility, i.e. a toilet and/or latrine, leads to numerous health challenges such as parasitic worms and environmental enteropathy. Parasitic worms are transmitted through human feces and cause multiple health complications in children including anemia and child growth stunting. Environmental enteropathy occurs with repeated and long-term inflammation of the small intestine which then reduces nutrient uptake and can cause child growth stunting, anemia and diarrhea. One-sixth of the world population has no access to any type of sanitation facility, and are therefore at higher risk of these challenges. Scientific literature on the impacts of sanitation typically examines household access to sanitation rather than community-level access to sanitation. We used national survey data to assess the impact that community-level access to sanitation has on child health, both for children with access to a sanitation facility and children without access to a sanitation facility. We found that a lack of sanitation access in the community is a significant risk factor for anemia and child growth stunting, but not for incidence of diarrhea. This risk decreases if a child has access to a sanitation facility, but even among those children with a sanitation facility poor sanitation access in the community is still a risk factor for anemia, child growth stunting and diarrhea. In addition to improving household access to adequate sanitation, community-level sanitation access needs to be addressed to improve child health. These results will add impetus to the Sustainable Development Goal to ensure sanitation access for all by 2030.
| An estimated 1 billion people live without access to any type of sanitation facility, i.e. a toilet or latrine [1]. This lack of sanitation access fails to contain human feces, which are responsible for transmission of various diarrheal diseases as well as soil-transmitted helminthes (STH) primarily through the fecal-oral route where fecal matter is ingested via water, dirt or food [2]. Diarrheal diseases kill millions of children each year [3], and for those who survive present the problem of malnutrition and developmental delays [4]. STH cause malnutrition and stunting in addition to developmental delays [5]. Furthermore hookworm (Necator americanus or Ancylostoma duodenale) are known risk factors for anemia [6]. Infections with Ascaris lumbricoides (roundworm) and Trichuris trichiura (whipworm) may also be risk factors for anemia although the evidence is inconclusive [7].
The prevalence of anemia is high in lower-income countries, estimated at 47% of children in 2005 [8], though recent reports suggest the prevalence is decreasing [9]. Due to the importance of iron to various cellular functions including immune system functionality [10,11], iron deficiency anemia is implicated as a cause of mortality for millions of children under five years of age each year [12,13]. Beyond a cause of mortality, anemia also decreases cognitive function [14–16], and energy levels which leads to decreased productivity and economic well-being [17,18]. For subsistence farmers in lower-income countries decreased productivity can in turn lead to low crop yields and food insecurity, perpetuating a vicious cycle of malnutrition.
Through containment and disposal of human feces, individual-level access to sanitation is known to decrease both diarrheal disease and STH infection [19–23]. A previous examination of survey data 1986–2007 found decreased risk of child mortality, diarrhea and stunting for children living in households with access to improved sanitation [24]. However, limiting sanitation to a household-level risk factor while ignoring the community-effect may greatly underestimate the impact that sanitation has on human health [25]. Poor sanitation in the community leads to increased exposure to fecal matter for all in that community, a significant risk factor for environmental enteropathy and subsequent child malnutrition [26]. Indeed, in India the behavior of open defecation was associated with reductions in child growth in an ecological analysis [27], and in Cambodia community-sanitation behavior was associated with increased child growth more prominently than household-sanitation behavior [28]. Numerous community-randomized controlled trials of total sanitation campaigns have suggested that increasing access to sanitation can improve child health [28–31], while others have found little to no effect of these interventions on child health [32–34].
Herein we present a study estimating the impact of community-level access to sanitation on child health as measured through child growth, anemia, and diarrhea symptoms using survey data compiled into an individual-level meta-analysis.
We sought to measure the impact that living in a community with 100% sanitation access has on the outcomes of child growth stunting among children aged 12–59 months, anemia among children under 5 years of age, and diarrhea in the previous two weeks from nationally-representative surveys. To do so we pooled surveys to create an individual-level meta-analysis [35].
We included multiple indicator cluster surveys (MICS), demographic and health surveys (DHS), malaria indicator surveys (MIS), and AIDS indicator surveys (AIS) that were nationally-representative and publicly available as of July 2016. As part of original survey protocol all data were anonymized prior to download from repositories to protect participant privacy.
Anthropomorphic data are regularly collected in nationally-representative surveys. In these surveys height for age z-scores are computed for children under 5 years of age based upon World Health Organization growth reference standards. We classified children as stunted or not based upon the child’s height for age z-score being less than 2 standard deviations of the WHO growth reference standard. The outcome of stunting was available for 267 of 301 datasets. Nationally-representative surveys typically use the HemoCue system to measure hemoglobin levels for children age 5 and under and adjust these values for altitude. Depending upon the level of hemoglobin in the blood anemia is classified as none (≥12.0 g/dl), mild (10.0–11.9 g/dl), moderate (7.0–9.9 g/dl), and severe (< 7.0 g/dl). We conducted analyses with two separate anemia outcomes, children with any anemia (mild, moderate, or severe) and children with moderate to severe anemia. The anemia outcomes were available for 104 of 301 datasets. Caregivers of children under five are also asked whether their child has had any commonly occurring illnesses such as fever, diarrhea, or cough. We classified children with diarrhea as those whose caregivers reported them having diarrhea in the previous 2 weeks, and children without diarrhea as those whose caregivers reported them not having diarrhea in the previous 2 weeks. The outcome of diarrhea was available for 281 of 301 datasets.
In order to estimate the incremental effect of increasing community-level sanitation access on the outcomes of child growth stunting and anemia among children we classified children as living in households with any type of sanitation facility (unimproved or improved), or not having any access to a sanitation facility. If households reported sharing a sanitation facility with others they were classified as having any type of sanitation facility. We defined community as the survey sampling area or cluster, and calculated the proportion of households having any sanitation facility (unimproved or improved) to serve as a measure of community-level sanitation access. We excluded datasets where > 95% of children live in communities with 100% sanitation access from any further analyses.
Children in households with sanitation facilities or in communities with high sanitation access are likely to be predisposed to less risk of stunting and anemia, independent of sanitation access. To account for this selection bias and potential confounding we used two separate methodologies. First, we stratified our analyses by children in households with any sanitation access and children in households without any sanitation access. Second we used exact matching on community-level measures to circumvent the inherent selection bias of living in communities with more access to sanitation. Using the MatchIt package [36] in R version 3.2.3 [37] we matched children on numerous community-level and other covariates. To do so, we first took the cluster mean of child-level immunization coverage (3 doses of diphtheria, pertussis and tetanus). We then took the cluster mean of household wealth quintile and household access to a water source that was not considered surface water (rivers, dams, ponds, lakes or unprotected springs). Once these cluster-level estimates were estimated we categorized estimates of cluster-level immunizations into tertiles, community-level wealth above and below the median, and community-level access to a non-surface water source above and below the median. In addition to the community-level measures we matched on household-level wealth (dichotomized into rich or poor) and mother’s education (dichotomized into completed primary or not). The exact matching was conducted in accordance with the following equation: mijkl = β0 + βCi + χHj + δPk + ϕSl where mijk is a matched group for child i in household j in cluster k in survey l, Ci is an estimate of the mother’s education, Hj is an estimate of household wealth, Pk is a vector of cluster characteristics and Sl is a survey dummy. The matching procedures and all covariates were selected a priori.
Pooling all datasets to create an individual-level meta-analysis we first examined the relationship between the outcomes and community-level sanitation access through a Lowess smoothing figure. To account for observable non-linearity in the exposure of interest we attempted to fit a cubic spline, however the spline was unable to account for the large decrease in the odds of the outcomes when going from 99% sanitation access to 100% sanitation access. We therefore categorized community-level sanitation access at 0%, 1–30%, 31–60%, 61–99%, and 100% to both align with the knots of the cubic spline (0.6 and 0.99) and to provide an appropriate comparison group (1–30%).
Second, we calculated the unadjusted association between the exposures and outcomes of interest. For the unadjusted analysis we included the dataset and household sanitation access as covariates and adjusted the standard errors for correlated data at the survey cluster level. Finally, we used a generalized linear model with the matched group as a random intercept and a logit link to assess an adjusted association between the exposures and outcomes of interest. We included the following covariates to decrease the potential for confounding, with variable selection determined a priori: household sanitation access, urban or rural, child’s age in years, mother’s education (quantified as none, some, and completed primary or higher), household wealth quintile, insecticide treated mosquito net (ITN) coverage (no ITN in household, household owns ITN but child did not use previous night, and child used ITN previous night), child’s weight for height (no wasting, 0–2 standard deviations below reference, >2 standard deviations below reference), child has a health or immunization card (no, yes), child immunizations (none, some, or all according to WHO standards), previous birth interval (< 24 months or not), birth order (firstborn, second born, third born, or later), mother’s age of the child in 5 year increments (i.e., 15–19, 20–24, etc.), household size in terms of number of people (<6, 6–15, >15), whether the household uses an open water source (defined as a river, stream, pond, or unprotected spring), national gross domestic product retrieved from the World Bank database for the year of the survey as a continuous variable, and dataset as a dummy variable.
The general model we use to assess the relationship between the outcomes and the exposure of interest is given by the following equations:
yijklm|πijklm~Binomial(1,πijklm)logit(πijklm)=β1Sanj×β2Sank+χCijk+δHj+κSl+ζmζm~N(0,ψ)
where πijklm is a dichotomous outcome for child i in household j in cluster k in survey l in matched group m, Sanj is whether the household has access to any sanitation or not, Sank is the level of sanitation access in the community, Cijk is a vector of child characteristics, Hj is a vector of household characteristics, Sl is a vector of survey characteristics and ζm is a random intercept for matched group m that is assumed to be normally distributed with a mean of zero. All analyses were conducted in Stata version 13.1.
We identified 301 publicly available two-stage cluster surveys from 93 separate countries beginning in 1991 and conducted as recently as 2015. Table 1 gives descriptive statistics on household and community-level access to sanitation, as well as the outcomes of stunting, anemia and diarrhea before matching. While access to sanitation has reportedly increased throughout lower-income countries, the proportion of children under 5 years of age living in communities with 100% sanitation access remains low throughout much of the world (Fig 1).
Before matching, these 301 datasets contained anthropomorphic information for 1,592,914 children under 5 years of age, measured levels of anemia for 424,334 children under 5 years of age, and reported symptoms of diarrhea for 2,140,805 (S1 File). The matched datasets contained anthropomorphic information for 1,197,371 children from 233 datasets, measured levels of anemia for 299,560 children from 93 datasets, and reported symptoms of diarrhea for 1,616,619 children from 247 datasets. (See S1 Data for Prisma framework).
Among children living in households with sanitation access, living in a community with 100% sanitation access is associated with lower odds of stunting (Table 2). The lower odds of being stunted is only observed at 100% sanitation access; there was no effect of increasing community-level sanitation access for children in households with a sanitation facility located in clusters with < 100% sanitation access (Fig 2). Among children living in households without sanitation access, living in communities with zero sanitation access was associated with higher odds of stunting compared to children living in communities with 1–60% sanitation access (Table 2). Among children living in communities with high access to sanitation (60% or more of households with sanitation access) not having household-level sanitation access was associated with higher odds of stunting compared to children living in communities with 1–60% sanitation access (Fig 2; Table 2).
For the outcomes of any anemia as well as moderate or severe anemia, increasing community-level access to sanitation is associated with lower odds of anemia for children in households with sanitation access as well as children in households without sanitation access (Fig 2; any anemia Table 3; moderate or severe anemia Table 4). Increasing protection for all children occurred with increasing community-level sanitation access.
For the outcome of diarrhea symptoms in the previous two weeks, increasing community-level access to sanitation is not associated with lower odds of diarrhea for children in households without access to sanitation (Fig 2, Table 5). For children in households with access to sanitation living in a community with 100% sanitation access was associated with a lower odds of diarrhea (Fig 2, Table 5).
Children living in houses with access to a sanitation facility was associated with lower odds of stunting and any anemia at all levels of community-access to sanitation compared to children in houses with no access to a sanitation facility (Table 6). Living in a house with access to a sanitation facility was associated with lower odds of the outcomes of moderate or severe anemia and diarrhea compared to living in a house with no access to a sanitation facility only when community-sanitation access was higher.
We found that community-level access to sanitation is associated with lower odds of stunting and anemia for children independent of household-level sanitation access, and lower odds of diarrhea for children in houses with a sanitation facility. For children with sanitation access our analyses suggest that further gains in reducing the risk of stunting, anemia and diarrhea can be made as their communities move toward universal sanitation access. For children without household-level sanitation access our analyses suggest that community-level sanitation in addition to household-level sanitation is an important factor in child health.
Unexpectedly for children without individual-level access to sanitation, living in a community with higher access to sanitation (60–99%) was not beneficial compared to living in a community with no access to sanitation in terms of both stunting and diarrhea. (It was beneficial for the outcome of anemia). We suspect that lacking a sanitation facility when the majority of neighbors have one is an indicator of vulnerability and for an outcome such as stunting with a multi-factorial causal etiology the vulnerability may represent a risk factor. In contrast for the outcome of anemia a significant benefit was observed for this particular population.
Diarrhea in the previous two weeks was not associated with community-level access to sanitation, except for those children living in communities with 100% sanitation access. Also household-level access to sanitation was only associated with lower odds of diarrhea when community-level sanitation exceeded 60%. These findings that found improved sanitation at the household level to be associated with lowered risk of diarrhea [24]. Unmeasured confounding is a primary threat to these types of analyses, and the lack of impact may be due to unmeasured risk factors. Furthermore, there is great uncertainty around the validity of self-reported diarrhea in surveys [38], and the subsequent misclassification error may lead to an underestimation of the impact of sanitation on diarrhea. Decreased fecal matter in the environment is likely to decrease circulation of diarrhea-causing agents, however there was no way to account for handwashing behavior in this analysis which is suggested to drive the relationship between diarrheal disease and sanitation [39,40].
The association between higher community-level sanitation access and the outcomes of anemia and stunting (at lower levels of community-level access) are consistent with the theory that environmental enteropathy is a significant risk factor for child malnutrition and health [26]. A recent modeling analysis and literature review suggests that community-level sanitation acts through a type of “herd-immunity” mechanism [41], and an observational study demonstrated the protective nature of herd-immunity from sanitation in rural Ecuador [42]. These analyses confirm that a lack of sanitation at the community level poses a risk to members of that community, independent of household sanitation access and that the greatest gains occur as communities achieve universal access to sanitation.
Our findings are in line with the scientific understanding of how fecal-oral transmission of various pathogens impact child health [26,41]. The measurement of sanitation access at the level of primary sampling unit of nationally-representative surveys is an innovation that improves upon previous analyses of survey data that only measure sanitation access at the district level [27], or consider sanitation as a household-level risk factor [24]. Still, these findings should be treated cautiously for a number of reasons. First we greatly simplified sanitation access as having a sanitation facility or not. The sanitation ladder is much more nuanced [43], with the greatest benefits to health coming from improving sanitation beyond a simple pit latrine. The simplification of sanitation access to having a facility or not allowed for its measurement at community level. Second, survey data are subject to error including recall and information error. Both outcomes included in this analysis were measured by survey personnel, and are not likely to be associated with the exposure of interest. However responses in survey questions about sanitation access may have suffered from social-desirability bias. Finally the use of the primary sampling unit as the community is not a perfect measure of community, given that primary sampling units may comprise various villages. Given the comprehensive nature of the datasets used and the random sampling of children selected we do not anticipate any publication or reporting bias to threaten the validity of these results.
These results suggest that the greatest gains in health from sanitation are made when communities achieve universal access to sanitation. Until access to sanitation is universal within a population, even those with access carry risk derived from those without access to sanitation. Access to sanitation was included in the Millennium Development Goals as target 7.C, with the goal of reducing by half the population without access to safe drinking water and basic sanitation. Progress was minimal; the target was missed by nearly 1 billion people [1]. These data show that poor community-level sanitation access is a significant risk factor for child growth stunting and anemia, both for children living in households with access to sanitation and for children living in households without. The number of children living in communities where any households lack sanitation access is alarmingly high throughout the world, and efforts must be made to achieve the Sustainable Development Goal of eliminating open defecation by 2030.
|
10.1371/journal.pcbi.1005631 | Grandmothering and cognitive resources are required for the emergence of menopause and extensive post-reproductive lifespan | Menopause, the permanent cessation of ovulation, occurs in humans well before the end of the expected lifespan, leading to an extensive post-reproductive period which remains a puzzle for evolutionary biologists. All human populations display this particularity; thus, it is difficult to empirically evaluate the conditions for its emergence. In this study, we used artificial neural networks to model the emergence and evolution of allocation decisions related to reproduction in simulated populations. When allocation decisions were allowed to freely evolve, both menopause and extensive post-reproductive life-span emerged under some ecological conditions. This result allowed us to test various hypotheses about the required conditions for the emergence of menopause and extensive post-reproductive life-span. Our findings did not support the Maternal Hypothesis (menopause has evolved to avoid the risk of dying in childbirth, which is higher in older women). In contrast, results supported a shared prediction from the Grandmother Hypothesis and the Embodied Capital Model. Indeed, we found that extensive post-reproductive lifespan allows resource reallocation to increase fertility of the children and survival of the grandchildren. Furthermore, neural capital development and the skill intensiveness of the foraging niche, rather than strength, played a major role in shaping the age profile of somatic and cognitive senescence in our simulated populations. This result supports the Embodied Capital Model rather than the Grand-Mother Hypothesis. Finally, in simulated populations where menopause had already evolved, we found that reduced post-reproductive lifespan lead to reduced children’s fertility and grandchildren’s survival. The results are discussed in the context of the evolutionary emergence of menopause and extensive post-reproductive life-span.
| In all human populations, regardless of environmental and socioeconomic conditions, menopause occurs in women well before the end of their expected lifespan. Conversely, extensive post-reproductive life-span is rare in other species; except in some cetaceans. Evolutionary theory predicts that menopause and extensive post-reproductive lifespan should emerge and persist in populations only if it is advantageous for gene transmission. Identifying this advantage is a long-standing issue. We provide a better understanding by demonstrating that humans’ cognitive abilities, in association with grand-mothering, are required for the emergence of this pattern. Indeed, cognitive abilities allow accumulation of skills and experience over the lifespan, thus providing an advantage for resource acquisition. These surplus resources can then be used to increase the number of offspring or be transmitted to existing offspring and grandoffspring. Stopping reproduction during aging allows allocating more resources to assist offspring and grandoffspring, thus increasing children’s fertility and grandchildren’s survival.
| Menopause, the permanent cessation of ovulation, occurs in women well before the end of their expected lifespan; reproductive senescence occurs substantially earlier than somatic senescence, leading to a particularly long post-reproductive life [1]. This is a rather uniform pattern across traditional and modern human societies. For example, if a man or a woman reaches age 45, he or she can expect to live at least an additional two decades [2–5]. However, and remarkably consistently across populations, reproductive senescence in women is largely completed by age 45 [6]. Extensive post-reproductive life-span (PRLS) in humans is thus not a consequence of modern improvements to nutrition, hygiene or medicine. Rather, reproductive cessation occurring approximately twenty years before the end of the expected lifespan appears as a constant feature of human biology [5, 7]. Among other species, only pilot and killer whales also exhibit extensive female PRLS. For instance, female killer whales can live into the 90s although they usually stop reproducing around age 40 [8–10]. However, patterns of reproductive and somatic senescence in killer whales differ from those of humans in some other ways, especially for males. Indeed, males rarely live beyond 50 years. Moreover, they do not undergo reproductive cessation [11]. In contrast, observations in traditional human populations have suggested that men may often undergo reproductive cessation once their wives reach menopause [12].
Understanding the conditions involved in the evolution of menopause and extensive PRLS is a long-standing challenge for biologists. First, an early end to reproduction seems contrary to maximizing Darwinian fitness. Second, the selective advantage associated with long life after the end of reproduction is not trivial. Various hypotheses have been proposed (for a review see [13]), including the Maternal hypothesis (MH), the Grand-mother Hypothesis (GMH), and the Embodied capital model (ECM). The MH is the idea that menopause has evolved in humans to avoid the risk of dying at childbirth, which is higher in older women, and to ensure the survival of the last offspring [14,15]. This hypothesis might thus explain why ageing women stop reproduction. However, as it relies on costs but not on benefits, the MH seems unlikely to explain alone the particularly long duration of PRLS observed in women. Indeed, whereas age-related costs of reproduction may explain early end of reproduction, it cannot explain why additional life after reproduction may be advantageous. Furthermore, death in childbirth may not be common enough to constitute a sufficient cost [16]. According to GMH [7] and ECM [12], both menopause and long life after reproduction may have evolved as two parts of the same allocation strategy consisting of ceasing to allocate resources to direct reproduction (i.e. producing new children) to favor indirect reproduction (i.e parental or grandparental care). Indeed, menopause and extensive PRLS may allow additional resource allocation to grandoffspring care and, therefore, increased fertility of the children and survival of the grandchildren. There are two main differences between the GMH and the ECM. The first resides in the specific causal hypotheses involved [17]. Indeed, according to the GMH, strength (e.g. proxied by body size) is the primary determinant of resource production [18–21]. Children productivity is low because foraging requires strength. As human growth is particularly slow, benefit of grand-mothering for grand-children survival and fertility is high, generating selection for older women to increase longevity [21]. According to the ECM, neural capital development and the skill intensiveness of the human foraging niche play the major role in shaping the age profile of resource production and transfers. In traditional societies, a peak of resource production is reached approximately twenty years after the peak of strength (mid-twenties) [12]. This is because earlier-life investments in neural capital lead to later-life energetic returns from such investments, with the consequences that individuals still acquire more resources than they need for survival until age 70 [12]. These extra resources could be used either for direct reproduction or for indirect reproduction. However, if the cost of reproduction increases with age (for instance, due to physiological constraints), it may be more advantageous to use these resources for increasing condition and fertility of the children and grandchildren, rather than increasing the number of children. The second difference between GMH and ECM resides in the fact the ECM is a two-sex model, whereas males may not be considered in the GMH. Indeed, as the traditional hunter-gatherer pattern of production, reproduction, and parental investment depends fundamentally on a cooperative division of labor between men and women, the ECM predicts that both aging women and men may stop producing new children to allocate resources to existing children and grand-children.
To test the MH [22, 23] and the GMH [24, 25], empirical studies have compared the fitness of children and grandchildren of women who experienced different durations of post-reproductive life-span. However, it is unclear if these studies help to understand the emergence or maintenance of menopause and extensive PRLS [26]. Indeed, the conditions favoring their maintenance are not the same as the conditions favoring their emergence. This is because female reproductive strategies in a population alter the social environment and determine the benefits of a trait. This change affects competition for reproductive resources and the average relatedness between interacting individuals [26]. Thus, the evolution of menopause and PRLS should not be studied outside of its ecological context or without considering the feedback between the evolution of this trait and the resulting ecology. To empirically study the evolutionary emergence of extensive PRLS, the fitness of rare mutant females who experience menopause should be compared to the fitness of resident females who do not. However, this is a possibility neither in humans, as menopause and extensive PRLS is already present in all populations [26], nor in our closest relative species, as reproductive senescence in midlife seems to be absent in non-human primates [27]. Regarding the ECM, the prediction that both aging women and men may stop producing new children to allocate resources to existing children and grand-children has been already supported by observations in traditional human populations [12]. However, the relation between neural capital development and skill intensiveness of the foraging niche on the one hand, and the duration of PRLS on the other hand, have not been demonstrated yet.
Here, we tested the MH, GMH and ECM for both the emergence and the persistence of menopause and extensive PRLS using a modeling approach based on life-history theory. Life history theory is the idea that living organisms must divide the total energetic potential available to them over their lifetime to perform different tasks, mainly survival, growth, direct reproduction, and parental care [28, 29]. As this energetic potential is limited, trade-offs occur among these tasks, resulting in different life-history strategies. The first trade-off occurs between immediate and future reproduction (via investment in growth and survival). The second trade-off occurs between the quantity and quality of offspring (i.e., having more offspring versus a larger investment in each of them). Modeling the evolution of allocation strategies should allow investigating the conditions for a switch from allocation to direct reproduction to allocation to indirect reproduction, i.e. for the emergence and persistence of both menopause and extensive PRLS. However, it requires a comprehensive model that considers both all of the allocation decisions that an individual has to make during his or her life, and how these decisions are shaped by complex interactions between genes, environment, and the internal state of the individual at the time when he has to make the decision.
We used Artificial Neural Networks (ANNs [30]) to simulate the evolution of resource allocation strategies, including all types of complex, even unforeseen, trade-offs in populations subject to diverse ecological conditions. Allocation decisions were allowed to freely evolve, and menopause and extensive PRLS emerged under some ecological conditions. We then tested for the following predictions: (1) Under the MH, menopause (and thus extensive PRLS) should not be observed without including age-dependent risk of dying at childbirth in the model; (2) Under both the GMH and the ECM, menopause and extensive PRLS should not emerge, whatever the ecological conditions, if resource transfers to grand-offspring are not allowed; (3) Under the ECM only, cognitive resources, because of delayed benefits of investment in neural development, should be a required condition for the emergence of menopause and extensive PRLS. Note that the ECM, as mentioned before, also predicts that both aging women and men may stop producing new children to allocate resources to existing children and grand-children, a prediction which has been supported by observations in traditional human populations [12]. Due to methodological issues (see limitation section), we did not test this prediction here. We rather focused on the relation between cognitive resources and extensive PRLS, which has never been tested before. Finally, we also tested whether MH, GMH and ECM may explain the persistence of extensive PRLS in simulated populations where this trait has already evolved. In these populations, GMH predicts that mother death at the age of menopause or delayed menopause of the mother should lead to decreased fertility of the children and/or decreased grandchildren survival. MH predicts that, under the same conditions, survival of the children should be decreased.
With the exception of the flow rate of available resources, all of the ecological parameters included in the model (α, the skill intensiveness of the foraging niche; β, the rate of skills acquisition; γ2, the difficulty of acquiring resources in the environment; δ, the depletion rate of somatic and cognitive capital; and σ2, the dangerousness of the environment) somehow influenced the duration of PRLS (Fig 1). Lower values for α were associated with shorter PRLS regardless of the values of the other parameters. However, high values of α were not always sufficient to generate a duration of PRLS higher than 5 time units, suggesting the presence of interactions with other parameters. The highest durations of PRLS observed (>5 time units) were associated with parameters of intermediate (for β) or high values (for δ and γ2), suggesting multiple complex interactions among them.
The maximization procedure confirmed that at least one combination of ecological parameters (α = 0.91; β = 0.47; γ2 = 157; δ = 0.87; σ2 = 27) lead to menopause and extensive PRLS with our model. Among all tested combinations of parameters (see Material and method section), this set, referred to as EP*, led to evolution of the longest duration of PRLS in the simulated population. With EP*, 1,121 individuals were born and died during the final 2,000 time units of the simulation process. Their average duration of PRLS was 21.92 (+/- 3.04) time units. A total of 78.1% of these individuals had exactly the same allocation strategy (S1 Fig), as defined by the combination of synaptic weights of the artificial neural network (see the Material and Methods section). Their average duration of PRLS was 18.84 (+/- 0.60). The typical life history of an individual with this allocation strategy achieving reproduction was the following (Fig 2): The first resource allocation was to growth, survival and maintenance until the quantity of somatic capital reached the value of 0.6 (at t = 18). Then, the first reproductive event occurred and somatic senescence started, thus suggesting that resources were allocated to reproduction at the expense of investment in the quality of somatic capital. Investment in maternal care for a given child was maximal following birth and then decreased over time. A second reproductive event occurred at t = 28, again at the expense of the quality of somatic capital. At t = 32, a grandchild was born. Then, the individual started to allocate resources for maternal care for both the first child, who has given birth, and the second child, who is not autonomous yet. The resulting increase in maternal care occurred at the expense of the quality of somatic capital but also at the expense of investment in direct reproduction. Indeed, no individual along the simulation process gave birth to an additional child after the birth of a grandchild. Despite the decrease of the quality of somatic capital, the quality of both cognitive capital and survival probability remained stable until t = 42. Then, they started to decrease until death, which occurred at t = 47.
Whatever the ecological parameters used, menopause and extensive PRLS did not emerge in the simulated populations when grandoffspring care was not allowed. In that case, the mean duration of PRLS obtained after applying the maximization procedure was 1.6 time units, a 92.6% reduction compared to the mean duration of PRLS (21.9 time units) obtained when grand-mothering was a possible option. Similarly, menopause and extensive PRLS did not emerge when cognitive resources were not differentiated from somatic resources in the model (i.e. both resources are interchangeable and had the same properties, including no delayed benefits). In that case, the mean duration of PRLS obtained after applying the maximization procedure was 2.1 time units, a 90.6% reduction compared to the mean duration of PRLS obtained with the full model. Finally, allowing resource transfers between siblings lead to no substantial changes in the results (duration of the PRLS of 22.04 with the full model, 1.71 without grand-mothering, and 2.03 without delayed benefit of investment in cognition).
In populations where menopause and extensive PRLS had already evolved, condition 1 (death at the age of menopause) had no detectable effect on the survival of the first-generation children (G1), although these children had reduced fertility. In the subsequent generations, the survival and fertility of the manipulated individuals with condition 1 were lower than the control (Fig 3) and they decreased in frequency (Fig 4). Manipulated individuals with condition 2 (delayed menopause, i.e. one additional reproductive event at the age of menopause) were significantly more frequent in the population than control individuals at G1, which was expected given the nature of the condition. Then, they decreased in frequency and were significantly less frequent than the control individuals from G5 to G10 (Fig 4). The probability to survive until reproduction was significantly lower for the manipulated individuals with condition 2 than for the control individuals, from G1 to G10. At G1, this difference was explained by a low probability of survival (0.23) for the last child, who was born at the mother’s expected age of menopause. Conversely, the other children had a probability of surviving until reproduction of 0.48, which is equal to those of the control individuals. The fertility of the manipulated individuals with condition 2 was significantly higher than that of the control individuals at G0, as expected given the nature of the condition. Then, however, fertility of the manipulated individuals was not significantly different from that of the control individuals from G1 to G10 (Fig 5). Finally, the lifespan of the manipulated individuals with condition 2 was significantly shorter than that of control individuals (on average 42 units of time rather than 47, p-value: 0.003).
Studying the correlations between the ecological parameters used for the simulations and the resulting duration of PRLS revealed that high values of α (i.e. skill intensiveness of the ecological niche) were necessary to generate duration of PRLS higher than 5 time units. This result supports the ECM [12]. However, high values of α were not sufficient to generate a duration of PRLS higher than 5 time units, suggesting the presence of complex interactions with other parameters. Therefore, studying the correlations between ecological parameters and the resulting duration of PRLS was insufficient to clearly understand the conditions favoring the emergence of menopause and extensive PRLS. We thus used the maximization procedure to investigate the evolution of life history traits in the simulated populations and to identify required conditions for emergence of menopause and extensive PRLS.
When allocation decisions were allowed to freely evolve in a simulated population, menopause and extensive PRLS emerged under at least one set of ecological parameters (Fig 2). The patterns of somatic and reproductive senescence obtained were strikingly similar in some ways to those observed in traditional human populations [12]. In particular, we observed a cognitive senescence beginning about twenty units of time after somatic senescence, and stable productivity until cognitive senescence begins. In contrast, some other characteristics of the evolved strategy were less realistic when compared to observations in traditional human populations (e.g. number of offspring per individual, inter-birth intervals, see Fig 2). However, note that we did not aim here to simulate precisely all the aspects of a human life-cycle. Indeed, there are substantial differences in the timing of life-history between human populations around the world, and all this variability cannot be captured here. Moreover, there is no indication that the trait values observed now in hunter-gatherers (mainly living in marginal habitats), reflect the values in the ancestral hunter-gatherers, at a time when menopause evolved. For these reasons, we have designed the maximization procedure to optimize the ecological parameters in order to obtain the longest PRLS under various simulated conditions. This approach allowed us to identify some factors which are required for the emergence of extensive PRLS, whatever the ecological parameters used, the species considered, and the other characteristics of the allocation strategy. An advantage from this approach is that our findings apply to any species with menopause and extensive PRLS, not only humans. Another is that we assumed a minimal number of physiological or environmental constraints.
In particular, menopause and extensive PRLS evolved without imposing a starting condition with the presence of a somatic senescence. Rather, somatic senescence, reproductive senescence (i.e. menopause) and extensive PRLS evolved as an allocation strategy. Similarly, no prior assumption was made on an increase of the cost of direct reproduction with age. To explain reproductive senescence, MH assumes that the cost of direct reproduction increases with age due to the higher risk of dying at childbirth [14, 15]. ECM also assumes increasing costs of reproduction due to physiological constraints (e.g. decreasing oocyte quality), although Kaplan et al. [12] recognized that additional costs to late-life reproduction beyond physiological costs (e.g. reduced future productivity from maternal depletion) may exist. Here, the cost of direct reproduction is only defined by the amount of resources allocated for direct reproduction and for parental care, which are allowed to freely evolve. However, when individuals were forced to reproduce at the age of menopause, their own lifespan was significantly reduced, and the child had a higher probability of dying before achieving reproduction, compared to previous children. Late reproduction is thus costly for survival and weakly advantageous for gene transmission, as assumed by MH and ECM. However, this is a result of an evolved allocation strategy rather than the consequence of pre-existing physiological constraints. Similarly, mortality was only a probabilistic consequence of a reduced quantity of resources invested in survival. Extrinsic mortality was not included in the model, as it can be considered that evolved organisms exert some control over many possible causes of mortality (e.g., by altering patterns of travel to avoid predators, by investing in immune functions, etc.; see [31]). Most types of mortality could thus be seen as the result of an allocation strategy.
Investigating how patterns of reproductive senescence were shaped by the evolution of allocation decisions under different simulated conditions allowed us to test three hypotheses (MH, GMH and ECM) for the emergence and the persistence of menopause and extensive PRLS. By supporting key assumptions from the GMH and ECM (but not the MH), our results support the idea that both grand-mothering (GMH) and cognitive resources (ECM) are required for the emergence of menopause and extensive PRLS.
We also support the importance of GMH, but not MH, in explaining the persistence of extensive PRLS in populations where this trait has already evolved. Indeed, in a population where extensive PRLS had already evolved, when maternal mortality was enforced at the age of menopause (i.e., on average 4.3 time units after the second childbirth), the children’s fertility was affected, but not their survival until reproduction (Fig 3). This non-reduced survival of motherless children did not result from allocare [37], as children without their mother could not receive resources from other individuals. Rather, it was the result of an evolved strategy consisting in prioritizing survival rather than fertility when facing a lack of resources. In contrast, grand-mothering is likely pivotal to maintain extensive post-reproductive life-span once it has evolved. Indeed, when the grand-mothering effect was suppressed at the age of menopause (the grandmother was forced to die) or reduced (the grandmother was forced to have an additional child so that parental resources were reduced for any given child), this was associated with a reduced fitness and the corresponding strategy decreased in frequency (Figs 3 and 5). This is consistent with several empirical studies [23–25, 28–42].
The main limitations in this study were due to the use of a one-sex model. Up to now, no validated and reliable method has been published to use neural networks in the context of a two-sex diploid model. We hope that further methodological developments will allow overcoming this limitation in the near future. It would make possible to complement this study by testing another key prediction of the ECM, i.e. both aging women and men may stop producing new children to reallocate resources to existing children and grand-children. Note however that this prediction has been already supported by observations in traditional populations [12]. In contrast, the relation between cognitive resources and duration of the PRLS had not been previously tested. Using a two-sex model would have also allowed testing the reproductive conflict hypothesis [23, 39, 43]. The idea is that, when old and young women are co-breeding in the same family unit, as in patrilocal societies, menopause could be the result of a limitation in resources due to competition. Relatedly, some authors suggested that, in this context of intra-familial competition, younger females should benefit from a decisive advantage as compared to older females [25,43,44] due to asymmetric relatedness. Indeed, the daughter-in-law is not related to the children of her mother-in-law, but the mother-in-law is related to the children of her daughter-in-law. Testing of these hypotheses require using a two-sex model, as they are explicitly based on relatedness within a family. Therefore, it cannot be excluded here that these processes, in addition with grand-mothering and cognitive resources, may have also played a role in the emergence or persistence of menopause and extensive PRLS in humans.
More generally, future developments may be envisaged to make the model more realistic. For instance, this may include taking into account migration and patterns of patri or matri-locality (i.e. the individuals can invest for their kin only if they are co-resident), modelling resource transfers between non-kin or distant kin, considering separately different kind of resources (e.g. time and energy), or allowing different degradation rates for somatic and cognitive capital. Indeed, in the absence of any published evidence that the respective degradation rates of somatic and cognitive capital are different, for the sake of simplicity, we assumed that these two rates are equal. Note that we speak here of physiological degradation rates, which are different from observed rates of decrease in performance. Indeed, there is published evidence that age-related decline in physical strength follows a very different trajectory than age-related decline in various cognitive abilities [35, 45]. However, age-related decline in performance depends both on the physiological degradation rate and on investment in maintenance. Finally, we assumed here a maximum of five children living simultaneously. This assumption was also imposed by computational limits. However, this is unlikely to have affected the results, as no individuals gave birth to more than three children in our simulated populations using the EP* parameters.
To conclude, we hope to stimulate further interest to use artificial neural networks (or any other adequate tool) to study the evolution of allocation decisions to address these questions, as well as many other issues in evolutionary biology. Indeed, allocation decisions are central to various long-standing questions in this field (e.g., the evolution of senescence, cognition, social interactions,…), and modelling their evolution may result in significant improvement.
The model (Fig 6) was coded in C++. The neural networks were fully connected multi-layer perceptrons with a single hidden layer of 5 neurons. The inputs to the networks were information on the internal state and social environment perceived by the individual. The outputs were the proportions of resources allocated to each function.
Preliminary exploration showed that increasing the amount of available resources at each time unit, all else being equal, lead only to a proportional increase in population size, without changing the average duration of post-reproductive period. This parameter was established at 20,000, resulting in population sizes of at least 500 individuals. A random value was attributed to each of the five other parameters (α, β and δ were drawn from a uniform distribution between 0 and 1 and γ2 and σ2 from a uniform distribution between 0 and 200), and a simulated population with an initial size of 1,000 individuals was allowed to evolve during 10,000 time units with these parameters. PRLS was measured as the average time interval between the last reproduction and death, calculated over the individuals who were born and died during the final 2,000 units of time. This process was repeated 100 times to detect the influence of each parameter on the variation of PRLS.
The maximizing function “rbga” (package “genalg” [47]), implemented in R v3.2 [48], was used to test whether at least one combination of ecological parameters lead to extensive PRLS with our model. To this end, we identified the combination of ecological parameters values able to promote the evolution of the longest PRLS, and we measured the average PRLS duration in a population which has evolved under these conditions. For each set of parameters (α, β, δ, γ2 and σ2), a population with an initial size of 1,000 individuals was simulated and was allowed to evolve during 10,000 units of time (or 20,000 units of time, without changing the results). PRLS was the variable to be maximized in the space of parameter values. For each combination of ecological parameters, a combination of synaptic weights evolved (i.e., became the most frequent in the population). This procedure thus allowed identifying both the combination of ecological parameter values which led to the longest PRLS (referred to as EP*), and the associated synaptic weights (referred to as the best weights).
To observe and describe the allocation strategy corresponding to the best weights, a population with an initial size of 1,000 individuals was simulated using the EP*, with all individuals having the best synaptic weights, without possible mutations. With these conditions, the demographic characteristics were allowed to freely evolve during 10,000 units of time (or 20,000 units of time, without changing the results). Although individuals with the same synaptic weights necessarily have a similar strategy, decisions could vary based on the local perceived conditions. This step allowed a reduction in the inter-individual variation of realized life histories.
To test the GMH, the same procedure was performed with the outputs corresponding to allocation in maternal care for a child who had already reproduced set to 0; grand-parenting was thus no longer a possible allocation option. Indeed, as mentioned before, grand-parenting was modeled in this study by allowing the individuals to adapt their parental investment for a given child depending on its own number of children, rather than allowing direct resource transfers to grandoffspring. If the GMH is determinant to explain the emergence of extensive PRLS, we thus expected that it cannot emerge under this condition, whatever the combination of ecological parameters. To test the ECM, the procedure was performed after removing delayed benefits of investing in cognition from the model (i.e. the integral term and the β parameter were removed from Eq 3). Indeed, without delayed benefits of investment, cognitive resources were not differentiated from somatic resources in the model (i.e. both resources are interchangeable and had the same properties).Strong delayed benefits of investment are a specificity of cognitive capital [12].Indeed, investing in neural development at time t promotes accumulation of skills and experience all along the life. Returns from cognitive capital can thus continue to increase (not only to be maintained) even after stopping investment in it. This is not the case for somatic capital. Indeed, although investment in somatic capital at time t can provide benefits later (for resource production, protection,…), these benefits will not increase without further investment. Therefore, this procedure allowed testing for the relation between cognitive resources and the duration of PRLS. Indeed, without delayed benefits of investment, cognitive resources were not differentiated from somatic resources in the model (i.e. both resources are interchangeable and had the same properties). In addition, as resource transfers are sometimes also provided by older siblings in humans, we also tested whether allowing transfers between siblings change the results.
To identify the costs of suppressed or reduced PRLS in a population where extensive PRLS has already evolved, we simulated 200 populations with initial size of 1,000 individuals, where all individuals had the same allocation strategy (best synaptic weights). We allowed each population to evolve using the EP* during 20,000 units of time, without possible mutations. 100 populations were attributed to condition 1: death at age of menopause, and the 100 other populations were attributed to condition 2: one additional reproductive event at age of menopause. At t = 10,000, we applied the condition to half of the individuals in each population. The condition was heritable and was also applied to their offspring at each generation. No condition was applied to the other individuals and their offspring (control). For each population, the proportion of individuals who received the condition among the successive generations, up to the 10th, was tested for a significant departure from the expected frequency (0.5) using two-sided binomial tests (R-based function binom.test). The average fertility and the proportion of individuals who survived until reproduction were compared between the control and condition across the first ten generations using two-sided student tests. The data fitted the requirements for these tests.
|
10.1371/journal.ppat.1006394 | Molecular basis for the binding and modulation of V-ATPase by a bacterial effector protein | Intracellular pathogenic bacteria evade the immune response by replicating within host cells. Legionella pneumophila, the causative agent of Legionnaires’ Disease, makes use of numerous effector proteins to construct a niche supportive of its replication within phagocytic cells. The L. pneumophila effector SidK was identified in a screen for proteins that reduce the activity of the proton pumping vacuolar-type ATPases (V-ATPases) when expressed in the yeast Saccharomyces cerevisae. SidK is secreted by L. pneumophila in the early stages of infection and by binding to and inhibiting the V-ATPase, SidK reduces phagosomal acidification and promotes survival of the bacterium inside macrophages. We determined crystal structures of the N-terminal region of SidK at 2.3 Å resolution and used single particle electron cryomicroscopy (cryo-EM) to determine structures of V-ATPase:SidK complexes at ~6.8 Å resolution. SidK is a flexible and elongated protein composed of an α-helical region that interacts with subunit A of the V-ATPase and a second region of unknown function that is flexibly-tethered to the first. SidK binds V-ATPase strongly by interacting via two α-helical bundles at its N terminus with subunit A. In vitro activity assays show that SidK does not inhibit the V-ATPase completely, but reduces its activity by ~40%, consistent with the partial V-ATPase deficiency phenotype its expression causes in yeast. The cryo-EM analysis shows that SidK reduces the flexibility of the A-subunit that is in the ‘open’ conformation. Fluorescence experiments indicate that SidK binding decreases the affinity of V-ATPase for a fluorescent analogue of ATP. Together, these results reveal the structural basis for the fine-tuning of V-ATPase activity by SidK.
| V-ATPase-driven acidification of lysosomes in phagocytic cells activates enzymes important for killing of phagocytized pathogens. Successful pathogens can subvert host defenses by secreting effectors that target V-ATPases to inhibit lysosomal acidification or lysosomal fusion with other cell compartments. This study reveals the structure of the V-ATPase:SidK complex, an assembly formed from the interaction of host and pathogen proteins involved in the infection of phagocytic white blood cells by Legionella pneumophila. The structure and activity of the V-ATPase is altered upon SidK binding, providing insight into the infection strategy used by L. pneumophila and possibly other intravacuolar pathogens.
| Acidification of intracellular compartments by vacuolar-type ATPases (V-ATPases) is crucial for numerous biological processes [1,2]. These processes include glycosylation in the Golgi [3,4], loading of neurotransmitters in secretory vesicles [5,6], protein trafficking in endosomes [7–9], and amino acid sensing in lysosomes [10,11]. V-ATPases pump protons across a phospholipid membrane using energy from the hydrolysis of adenosine triphosphate (ATP) to adenosine diphosphate (ADP) and inorganic phosphate [1,12]. In the yeast Saccharomyces cerevisiae the complex is composed of subunits A3B3CDE3FG3Hac8c′c″def [13], where subunits denoted by upper case letters form the soluble catalytic V1 region while subunits denoted by lower case letters form the membrane-embedded VO region. ATP hydrolysis occurs in the V1 region, where three catalytic heterodimers of A- and B-subunits assemble into a pseudo-symmetric trimer of AB heterodimers [14–16]. Each AB heterodimer contains a catalytic nucleotide-binding site and each is found in a different conformation [17] termed ‘tight’, ‘loose’, and ‘open’ with bound ATP, bound ADP and phosphate, and no nucleotide expected to be bound, respectively. Conformational changes in the AB heterodimers are coupled to proton translocation across the VO region by a rotary catalytic mechanism [18,19] where the rotor subcomplex, consisting of subunits DFc8c′c″d, turns relative to the rest of the complex. Under certain conditions, the V1 region can dissociate from the VO region to inhibit ATP hydrolysis [20–23] and prevent proton translocation. Aside from dissociation of the complex, little is known about how the activity of V-ATPases is regulated.
V-ATPase activity has a central role in the clearance of material phagocytosed by immune cells. Killing of pathogens by phagocytic white blood cells, such as macrophages, occurs in phagolysosomes [24], which are acidified by V-ATPases [25]. This acidification leads to the activation of enzymes that help to destroy phagocytized material. Some intracellular bacteria subvert this process by secreting effectors that inhibit either the formation or acidification of phagolysosomes [25–27]. The protein SidK, secreted by Legionella pneumophila [28,29], interacts with the V-ATPase to inhibit acidification of the phagolysosome in the early stages of infection [25]. However, the molecular basis of this interaction and inhibition remain unclear. In this study, we determined the structure of the N-terminal domain of SidK at 2.3 Å resolution by X-ray crystallography, revealing SidK to be a flexible and elongated protein. Electron cryomicroscopy (cryo-EM) has recently emerged as a powerful method to analyse the structure of the V-ATPase at subnanometer resolution [13,30,31]. Although SidK normally binds to the human V-ATPase, to date it has only been possible to perform sub-nanometer resolution cryo-EM with V-ATPase from S. cerevisiae [30] or the insect Manducca sexta [31] due to abundance of the enzyme. Therefore, we used the S. cerevisiae V-ATPase to determine the structure of a V-ATPase:SidK3 complex at 6.8 Å resolution by single particle cryo-EM, showing that SidK binds the N-terminal region of the V-ATPase A-subunit. SidK binding to the V-ATPase inhibits V-ATPase activity by ~40%. Consistent with this subtle fine-tuning of V-ATPase activity, the structures do not reveal significant conformation rearrangement on binding. Instead, SidK binding reduces A-subunit flexibility and decreases the affinity of V-ATPase for the fluorescent ATP analogue TNP-ATP. The full neutralization of the Legionella containing vacuole (LCV) in the early stages of infection [32] therefore likely involves other effectors and the extended C-terminal domain of SidK seen in our model suggests that there are additional roles for this part of the protein.
In order to gain insight into the role of SidK in the intracellular survival of L. pneumophila, intact SidK (575 residues) and different truncations of the protein were expressed heterologously in Escherichia coli, purified, and crystallization trials were performed. One SidK construct, consisting of residues 16–278 (SidK-N), yielded two crystal forms that diffracted X-rays to 2.3 to 2.4 Å resolution and the structure of SidK-N was determined from these crystals. The first crystal form contained one molecule of SidK(16–278) in the asymmetric unit and the final model included amino acids 16–274. The protein has an elongated and slightly bent α-helical structure consisting of three α-helical bundles (Fig 1A). The three α-helical bundles contain four, four, and three α-helices, respectively, and are connected by loops in the bundles. The first α-helical bundle (α1–2, α5–6) also has an extension nearly perpendicular to the bundle axis that consists of two additional short α-helices (α3-α4) (Fig 1A, green arrowheads). Bioinformatic analysis of the SidK structure showed that the arrangements of α-helices in the second and third α-helical bundles are similar to the arrangement of α-helices observed in many other proteins, including DOCK2 (PDB ID 3A98) and endo-α-N-acetylgalactoseaminidase [33]. However, the N-terminal bundle with its extensions, which from previous analysis was proposed to be the region that interacts with the V-ATPase [25], does not resemble any other known protein structures. The second crystal form of SidK(16–278) contained two molecules in the asymmetric unit related by a non-crystallographic two-fold symmetry (Fig 1B). In this crystal form SidK-N exists as a dimer, with the two molecules swapping their second and third α-helical bundles (Fig 1A and 1B, red arrowheads and oval). This domain swapping is accomplished by a rotation of ~180° relative to the first α-helical bundle (Fig 1C) that occurs through conformational changes within the short linker connecting the first and second α-helical bundles (residues 117–125). The domain swapping retains the interface between the first and second α-helical bundles seen in the first crystal form and the domain-swapped monomer is similar to the monomer in the first crystal form (Fig 1A and 1B). The monomeric SidK predominates in solution, even at high concentrations, and therefore the elongated monomer structure is expected to be the functional state of the protein. Only the N-terminal region of SidK interacts with the V-ATPase and consequently a highly-elongated SidK structure was unexpected. However, the structure on its own did not provide clues into how SidK interacts with or inhibits the V-ATPase.
In order to understand how SidK interacts with the V-ATPase, purified full length SidK (residues 1 to 573) and detergent solubilized and purified intact V-ATPase from S. cerevisiae [23] were incubated together and the resulting complex analysed by cryo-EM (S1 Fig). In the absence of ATP, V-ATPase adopts three distinct conformations that correspond to different rotational positions of its rotor subcomplex relative to the rest of the enzyme [30]. Particle images were subjected to 3-D classification in order to separate these different conformations of the V-ATPase, as well as different occupancies of SidK binding to the V-ATPase (S2 Fig). The cryo-EM maps showed additional density corresponding to SidK attached to the A-subunits of the V-ATPase (Fig 2A, green arrowheads). While V-ATPase particles decorated with SidK were found in all three rotational states of the enzyme, the highest-quality cryo-EM map was obtained for V-ATPase in rotational state 1 fully decorated with three copies of SidK (Fig 2A and S2 Fig) and reached a resolution of 6.8 Å (S3 and S4 Figs). The C-terminal region of SidK had a local resolution worse than the rest of the complex (S4A Fig, 'C-term'). The distribution of particles in the different rotational states differed slightly from the distribution observed in the absence of SidK (S4B Fig): 49%, 23%, and 28% for rotational states 1, 2 and 3 with SidK bound versus 47%, 36%, and 17% without SidK bound [30]. A pseudo atomic model of the V-ATPase:SidK3 assembly was constructed using the cryo-EM map to guide flexible fitting of the crystal structure of SidK determined here, an earlier model of V-ATPase subunits A3B3CDE3FG3 [30], and a model of the VO region subunits ac8c′c″def [13] (Fig 2A). Fitting of SidK into each of the three corresponding densities in the map required flexing the crystal structure only at the interface between α-helical bundles I and II. The flexibility between the first and second α-helical bundles is consistent with the flexibility in the linker between these bundles observed by X-ray crystallography. Bound to the V-ATPase, the C terminus of SidK extends toward the expected position of the membrane (Fig 2A). The flexing of the SidK model required to fit it into the cryo-EM map introduced a slight bend in the linker region around residues Lys123-Ser124. The fitting shows that SidK binds the V-ATPase A-subunit primarily via its first α-helical bundle (Fig 2B, blue bracket) and the three copies of SidK each interact mostly with the corresponding N-terminal region of one of the three A-subunits [25]. Mapping the contact surfaces in all three SidK:A-subunit pairs in the cryo-EM map to the SidK crystal structure, the main contact surface appears to involve residues Gly24, Tyr28, Phe62, Ser85, and Trp122 from the first α-helical bundle of SidK. Although the SidK crystal structure spans almost the entire region that binds the V-ATPase, it is missing the first 16 residues of the protein, which the cryo-EM map shows to form a hook-like feature that penetrates the non-catalytic interfaces between AB heterodimers (Fig 2C, 'hook'). The C-terminal region of SidK is poorly resolved in the map and has lower density than SidK-N, suggesting flexible tethering between the N- and C-terminal regions (Fig 2A and 2C, grey density). Surprisingly, comparison of the V-ATPase:SidK3 and V-ATPase [30] structures revealed no major conformational rearrangements in the V-ATPase upon SidK binding (S4C–S4E Fig). Furthermore, purification and structural analysis of substoichiometric V-ATPase:SidK assemblies (S2C Fig) did not show major conformational differences in the V-ATPase with substoichiometric SidK, compared to V-ATPase alone and V-ATPase that is fully-decorated by SidK.
The main interaction between SidK and V-ATPase involves the ‘non-homologous regions’ of the A-subunits (Fig 2B). These regions are found in the catalytic A-subunits of V-ATPases, but not in the corresponding catalytic β-subunits of ATP synthases, suggesting that SidK inhibition is V-ATPase-specific. To test this hypothesis we performed in vitro ATPase inhibition assays with V-ATPase and the F-type ATP synthase. These assays showed that SidK inhibits V-ATPase by ~40% (Fig 3A) but does not detectably inhibit the F-type ATP synthase (Fig 3B). Partial inhibition of the V-ATPase by SidK is not surprising, as complete inhibition of V-ATPase activity is well known to kill mammalian cells [34], which would not be advantageous for L. pneumophila infection. A construct consisting of residues 10 to 414 could inhibit V-ATPase to the same extent as the full length SidK (Fig 3A), showing that the N-terminal ‘hook’ feature seen in Fig 2C is not essential for inhibition by SidK. SidK was first identified as a V-ATPase binding protein because its expression in yeast reduced growth in liquid medium buffered to pH 7.0 [25], a characteristic of V-ATPase deficiency. We found that yeast expressing SidK from a plasmid were still able to grow on solid rich medium supplemented with 4 mM ZnCl2, indicating residual V-ATPase activity and consistent with the partial inhibition seen in in vitro assays (Fig 3A). Residues from the first α-helical bundle of SidK that appear to interact with V-ATPase (Fig 2B, Gly24, Tyr28, Phe62, Ser85, and Trp122) are highly conserved in SidK orthologs from other Legionella species, with Ser85 substituted to Ala in some distant Legionella species [35]. α-helical bundles II and III of SidK had almost no exposed conserved residues, except for Glu220. In order to test the importance of SidK residues as well as the model for V-ATPase binding presented above, SidK point mutations G24E, Y28A, F62A, S85E, and W122A were generated in both yeast and bacterial expression vectors. Using the bacterial expression vectors, mutant SidK was purified and all of the constructs were found to have melting temperatures within 0.5°C of the 54.1°C Tm found for wild type SidK. Mutant SidK constructs in yeast expression vectors were used to transform yeast, and yeast growth was monitored at pH 5.5 and 7.0, the latter requiring fully functional V-ATPase for optimal growth. This assay showed that the point mutations F62A and S85E are sufficient to prevent SidK from inhibiting V-ATPase and allow normal yeast growth (Fig 3C and 3D). Mutants Y28A and W122A were indistinguishable from wild type SidK while G24E expressed at a lower level than the other SidK constructs and consequently its effect on V-ATPase activity could not be characterized (Fig 3D). The SidK mutants F62A and S85E (S5A Fig). However, unlike wild type SidK (S1A Fig), could not be co-purified with V-ATPase, suggesting significantly decreased affinity for the V-ATPase. The F62A and S85E mutations also prevent SidK from inhibiting ATPase activity of the V-ATPase in vitro (S5B Fig). For these ATPase assays, which required large quantities of purified V-ATPase and simultaneous availability of wild type SidK and mutant SidK at identical concentration, SidK samples were frozen before use. Possibly as a result of this freezing, wild type SidK only inhibited V-ATPase by ~30%, not the ~40% inhibition seen with freshly purified wild type SidK (Fig 3A). The ability of the F62A and S85E mutations to prevent SidK from inhibiting V-ATPase supports the proposed model of SidK:V-ATPase complex formation.
As described, the V-ATPase:SidK3 structure by itself did not offer an explanation for the ~40% inhibition of V-ATPase activity that occurs on SidK binding. However, comparison of a map of V-ATPase alone [30] with the V-ATPase:SidK3 structure showed a subtle difference. In the earlier cryo-EM maps of the V-ATPase alone, the C-terminal region of the A-subunit in the ‘open’ conformation had a lower density relative to A-subunits in the ‘tight’ or ‘loose’ conformations (Fig 4A upper, yellow densities). The cryo-EM map of V-ATPase alone was determined with identical specimen preparation, imaging, and image processing conditions. The decreased density indicates mobility of this protein domain. In comparison, the C-terminal region of the ‘open’ A-subunit in the V-ATPase:SidK3 cryo-EM map is well defined (Fig 4A lower, yellow densities, and 4B, surface versus mesh), indicating a more rigid structure when SidK is bound. This higher density for the ‘open’ A-subunit with SidK bound compared to the ‘open’ A-subunit without SidK is seen in all three rotational states of the enzyme (Fig 4C). Furthermore, subtraction of the V-ATPase map from the V-ATPase:SidK3 map, although dominated by the presence of SidK, shows residual density in all three rotational states that is consistent with a more rigid A-subunit with SidK bound (S6 Fig). Estimation of the local resolution of the density maps also shows higher relative resolution estimates for the ‘open’ A-subunit C-terminal region in V-ATPase:SidK3 than in V-ATPase alone (S4A Fig), further suggesting reduced mobility upon SidK binding.
The ‘open’ A-subunit is poised to bind nucleotide and flexibility is essential for rotary catalysis in the V-ATPase and other rotary ATPases [30,36]. The binding of SidK to the A-subunit and the effect of SidK on the flexibility of this subunit suggested that SidK may alter binding of nucleotide to the AB heterodimers. The effect of SidK on the affinity of V-ATPase for ATP was investigated using the soluble V1 subcomplex purified from S. cerevisiae. As expected, the purified V1 subcomplex had no detectable ATPase activity [22]. However, the V1 subcomplex was still able to bind nucleotide as well as SidK (Fig 4D and S1A Fig), suggesting that the catalytic A3B3 hexamer of the V1 subcomplex remains in a conformation similar to the intact V-ATPase assembly and that the auto-inhibition of the V1 region does not interfere with accessibility of the active site. The binding affinity of nucleotide to the V1 subcomplex was determined using the fluorescent TNP-ATP. Although the affinities of a protein for ATP and TNP-ATP may be different [37], this fluorescent analogue of ATP has been used successfully to compare changes in nucleotide binding in a variety of experiments [37–39]. TNP-ATP was titrated into a solution containing the purified V1 assembly and fluorescence was measured (Fig 4D). The increase in fluorescence was modeled to calculate the equilibrium dissociation constant Kd (Fig 4D) according to Eqs 1 and 2 (see Methods). The calculated Kd of V1:TNP-ATP (160 +/- 50 nM) was approximately two-fold lower than the Kd of V1:SidK:TNP-ATP (280 +/- 60 nM), indicating that SidK decreases the affinity of the V-ATPase soluble catalytic region for TNP-ATP. This decrease in nucleotide affinity is consistent with the decreased flexibility of the A-subunit of V-ATPase upon SidK binding. Removal of the C-terminal residues 415–573 of SidK did not significantly affect V-ATPase inhibition by the SidK mutants (Fig 3A), but truncations of the N-terminal region of the protein has been demonstrated to prevent V-ATPase inhibition [25]. These data show that binding of the N-terminal domain of SidK to the V-ATPase is responsible for its partial yet specific inhibition of V-ATPase activity. The role of the C-terminal domain of SidK in vivo remains unclear.
In the data presented here, we determined the structure of residues 16–278 of SidK by X-ray crystallography. We showed that the N-terminal α-helical bundle of the protein binds with sufficient affinity to the catalytic A-subunit of V-ATPase to determine a cryo-EM structure of the V-ATPase:SidK3 complex. The interaction between SidK and the ‘non-homologous’ region of the V-ATPase A-subunit explains the specificity of SidK to V-type rotary ATPases. However, we observed that binding of purified SidK to purified and detergent solubilized yeast V-ATPase in vitro leads to only a ~40% inhibition of V-ATPase activity. It is possible that SidK induces a larger inhibitory effect on the macrophage V-ATPase than on the yeast V-ATPase. However, sequence alignment (S4F Fig) shows that the A-subunit from both human and yeast V-ATPase are highly conserved in their binding site for SidK, suggesting that at least binding affinity is similar.
The partial inhibition of V-ATPase is consistent with the observed effects of SidK binding on V-ATPase structure and TNP-ATP binding affinity. Complete inhibition of purified V-ATPase by bafilomycin shows that the V1 region (where ATP hydrolysis occurs) and VO region (where bafilomycin binds) are fully coupled in the preparation of the enzyme used in these studies. The 40% inhibition appears to contradict the previous observation that purified SidK is a potent inhibitor of ATPase activity in yeast membrane vesicles [25]. However, the current in vitro assays with purified components [40] are more precise than the assays used previously, and complete inhibition of V-ATPase by an intracellular pathogen would be unexpected as it would ultimately kill the host cell. The amount of SidK expressed and translocated into the host cell cytoplasm could affect how SidK influences the host cell but it seems unlikely that inhibition would ever exceed the 40% observed in the in vitro assay without compromising the utility of the infected cells for the pathogen. Early in infection, L. pneumophila maintains a neutral pH in the Legionella containing vacuole (LCV) [32]. This stage of infection corresponds to the period where SidK expression is high [25]. However, it was also found that L. pneumophila with SidK deleted is still infectious and did not exhibit a detectable growth defect in either mouse bone marrow-derived macrophages or Dictyostelium discoideum [25]. This observation supports the idea that, as with most Legionella effectors, there are multiple functionally-redundant effectors that serve to neutralize the pH of the LCV during the early stages of infection. This idea is also consistent with the amount of inhibition of V-ATPase caused by SidK, which likely functions along with other, currently unknown, factors to control the pH of the LCV. SidK is not expressed at late phase of L. pneumophila infection, which may allow V-ATPase to lower the pH in the LCV, a condition that appears to benefit intracellular bacterial replication [32]. Thus, the coordinated expression of multiple effector proteins that affect the pH of the LCV would provide L. pneumophila with a level of control over LCV pH that would not be possible with a single effector protein that is a potent V-ATPase inhibitor.
A recent report showed that SidK binds the V-ATPase with a Kd of ~3.5 nM [41]. This high affinity suggests that even translocated at a low level in the host cell, SidK will bind to available V-ATPase complexes and associate with the LCV, which is known to possess V-ATPase at its membrane [42]. V-ATPase has been proposed to interact with numerous other proteins in cells, including ARNO and Arf6 [7], actin [43], aldolase [44], and ragulator [11]. It is possible that SidK binding to V-ATPase alters one of these interactions leading to downstream consequences in addition to V-ATPase inhibition. Although SidK clearly causes partial inhibition of the V-ATPase upon binding, it does so without causing significant structural changes in its target. Numerous other proteins inhibit rotary ATPases, but their binding usually involves more dramatic alteration of rotary ATPase structure and function. Inhibitory Factor 1 (IF1) prevents ATP hydrolysis in the mitochondria F-type ATP synthase [45] and the ε-subunit of the bacterial F-type ATP synthase inhibits ATP hydrolysis in that enzyme [46]. However, unlike IF1 and the ε-subunit, SidK binds to an interface on the A-subunit that is far from the catalytic site of the enzyme. The SidK binding site also differs from that of PA1b from Pisum sativum, which inhibits insect V-ATPases by binding to the c-subunits [47]. It is tempting to speculate that the C-terminal regions of SidK, which extend toward the membrane in the V-ATPase:SidK3 structure, may play additional roles in infection. This hypothesis is particularly appealing because the majority of the intact SidK protein is not necessary for interaction with the V-ATPase and is accessible to serve some other function. The inability of SidK to cause significant structural changes in V-ATPase is consistent with the need for V-ATPase activity as the infection proceeds. Two of five point mutations tested were sufficient to prevent inhibition of V-ATPase by SidK. Three of the five mutations engineered to test the proposed SidK:V-ATPase interface did not alter SidK’s affect on yeast growth. For the G24E mutation, this null result is explained by reduced expression of the construct. Mutations Y28A and W122A appear to be simply insufficient to abrogate inhibition by SidK. Given their important role in the defense against intracellular pathogens, V-ATPases are likely modulated by other intracellular pathogens during infection. The hybrid structural analysis procedure described here is uniquely capable of understanding the formation of these host-pathogen protein complexes.
Truncated constructs of SidK (Q5ZWW6_LEGPH) were designed and created taking into account the secondary structure and disorder predictions [48,49]. Two constructs of SidK were cloned for crystallization trials: the full-length protein (residues 2–573) and a truncated N-terminal domain (residues 16–278). The gene Lpg0968 was PCR-amplified from the genomic DNA of L. pneumophila strain Philadelphia 1 using the following primers: forward (2–573) 5’- TACTTCCAATCCAATGCCTCTTTTATCAAGGTAGGTATAAAAATGGG -3’, reverse (2–573) 5’- TTATCCACTTCCAATGTTA AAGGCTTAGGCTTTCTTCCTGTACTTT-3’; forward (16–278) 5’- TACTTCCAATCCAATGCCGAGCAATATCATAGTCAAGTAGTCGGT -3’, reverse (16–278) 5’-TTATCCACTTCCAATGTTATTTGCTTAAAGCATTTAATTTTTCGTTTTC-3’. The PCR products were cloned by Ligation Independent Cloning (LIC) into the LIC vector pMCSG7 [50] with the standard protocol [50]. The pMCSG7 vector encodes an N-terminal 6×histidine tag, separated from the target gene by a TEV protease cleavage site. The constructs produced, pCX0016 (SidK2-573) and pCX0038 (SidK16-278), were verified by DNA sequencing.
Competent BL21(DE3) pLysS cells (EMD Millipore Corp., Billerica, MA, USA) were transformed with the expression vectors described above. Single colonies were used to inoculate LB media (50 mL, 100 μg/mL ampicillin) and were grown overnight at 37°C. 1L TB media (100 μg/mL ampicillin) were inoculated with 50 mL of the overnight cultures and grown at 37°C with shaking at 220 RPM. Isopropyl β-D-1-thiogalactopyranoside (IPTG) was added to 0.5 mM when the OD600 of the cultures reached 2.0, and cells were grown for another 16 h at 18°C before being harvested by centrifugation for 20 min at 5,000 g. SidK(16–278) labeled with selenomethionine was expressed in methionine auxotroph B834(DE3) (Novagen). For selenomethionine labeling cells were grown in 1 L M9 medium supplemented with methionine at 37°C. When the OD600 reached 1.0 cells were centrifuged and resuspended in 1 L of M9 medium and the culture was grown at 37°C for another 4 h to deplete remaining methionine. After depletion, 50 mg/L of selenomethionine was added to the cell culture 30 min prior to the addition of IPTG.
Cell pellets were resuspended in buffer A (50 mM Tris pH 7.8, 400 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine (TCEP)) with 1 mM p-aminobenzamidine and lysed with a high-pressure cell disrupter TS series Benchtop (Constant Systems Ltd, UK) at 35 psi. After centrifugation for 30 min at 16,000 g, the supernatant was loaded onto 5 mL of TALON Metal Affinity Resin (Clontech Laboratories, Inc., USA), incubated for 2 h at 4°C, and washed with 20 resin volumes of buffer A. Protein was eluted with buffer A with 100 mM imidazole. Protein was then dialyzed against 2 L of buffer B (20 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) pH 8.0, 150 mM NaCl, 0.5 mM TCEP), and cleaved with TEV protease (1:50 (w:w) ratio, 16 h at 4°C) to remove the 6×histidine tag. Protein was purified further by gel filtration with an ENrich SEC 70 10 x 300 column (Bio-Rad Laboratories Inc., USA) in buffer C (20 mM HEPES pH 8.0, 150 mM NaCl, 0.5 mM TCEP). The protein eluted in two peaks: ~ 10% as an apparent dimer and ~ 90% as a monomer. The dimer peak fractions were pooled and concentrated to 18 mg/mL for the crystallization screening. After initial unsuccessful crystallization trials, the monomer peak fractions were subjected to reductive methylation using the Reductive Alkylation Kit (Hampton Research, Aliso Viejo, CA, USA), purified again by gel filtration, and the peak fractions were concentrated to 30 mg/mL for crystallization screening. SidK mutations were made by site directed mutagenesis (New England Biolabs). Protein melting temperatures were measured with the fluorescent dye Sypro Orange (Molecular probes) using an Applied Biosystems StepOnePlus Real Time PCR Instrument (Life technologies) according to the standard TmTool protocol developed by the manufacturer.
SidK(16–278) dimer fractions produced well-diffracting crystals during crystallization screening with a Crystal Gryphon robot (Art Robbins Instruments, USA) using a Hampton Research Crystal Screen. The best crystals grew at 20°C with the well solution containing 0.1 M sodium cacodylate trihydrate pH 6.5, 0.2 M magnesium acetate tetrahydrate, and 20% (w/v) polyethylene glycol 8,000. The best crystals of Se-Met derivatized SidK(16–278) dimer were obtained at 15°C by microseeding with well solution containing 0.1 M 2-(N-morpholino)ethanesulfonic acid (MES) pH 6.5, 0.2 M magnesium acetate tetrahydrate, 0.13 M succinic acid pH 7.0 (Hampton Research) and 18% (w/v) polyethylene glycol 8,000. The initial crystallization conditions for the methylated monomer of SidK(16–278) were also identified using the Hampton Research Crystal Screen and vapor diffusion method. The best crystals were obtained at 20°C with wells containing 0.1 M Tris hydrochloride pH 8.5, 0.2 M lithium sulfate monohydrate, and 30% (w/v) polyethylene glycol 4,000. For data collection, crystals were soaked in a cryo-protectant (reservoir solution supplemented with 15% glycerol) and flash-cooled in liquid nitrogen. The X-ray diffraction data were collected at 100 K on the 08ID-CMCF (SidK-N dimer) and CMCF-BM 08B1-1 [SidK(16–278) methylated monomer] beamlines at the Canadian Light Source (Saskatoon, SK, CA) [51]. Diffraction data from the Se-Met crystals were collected at the Se absorption edge wavelength of 0. 9788 Å. The native and Se-Met datasets were processed and scaled with XDS [52].
The positions of heavy atoms were found with SHELXD [53] and the initial model of the SidK(16–278) dimer was built with Phenix AutoBuild [54]. The model was then refined against a higher-resolution native dataset using the PHENIX software package [54] combined with manual rebuilding using Coot [55]. The structure of the SidK(16–278) monomer was solved using the molecular replacement program Phaser [56]. Asp125-Pro235 of SidK(16–278), was used as the initial model. This portion of the molecule was fixed as a partial solution, and then the N-terminal fragment of SidK(16–124) was used as a search ensemble. The resulting model was refined with phenix.refine and manually rebuilt in Coot. The structures were validated with MolProbity [57]. The details of data collection and refinement statistics are given in Table 1. Search for structural homologs of SidK (16–278) was performed with the DALI server [58], PDBeFold [59] (http://www.ebi.ac.uk/msd-srv/ssm), and deconSTRUCT [60].
V-ATPase and V1 were purified from S. cerevisiae as described previously [23] via a C-terminal 3×FLAG tag on the A-subunit. The SidK gene was cloned into a pET28 plasmid containing an N-terminal 6×histidine tag followed by a tobacco etch virus (TEV) cleavage site. BL21 Codon+ cells containing the SidK plasmid were grown at 37°C with shaking (225 RPM) in 1–4 L of LB media supplemented with 0.4% (w/v) glucose and 50 mg/L kanamycin. At an OD600 of 0.7, protein expression was induced with 1 mM IPTG and cells were grown overnight at 16°C. All subsequent steps were performed at 4°C. Cells were harvested by centrifugation at 5,000 g and lysed by sonication in TBS (50 mM Tris-HCl pH 7.4, 0.3 M NaCl) containing 0.001% (w/v) phenylmethanesulfonylfluoride (PMSF). Cell lysate was centrifuged at 38,000 g and the supernatant was loaded onto a HisTrap Ni-NTA column (GE Healthcare). The HisTrap column was washed in TBS containing 25 mM imidazole and SidK was eluted in TBS containing 0.3 M imidazole. SidK was mixed with TEV protease and dialyzed overnight in 2 L of TBS buffer containing 1 mM dithiothreitol (DTT). Cleaved protein was dialyzed in 2×1 L of TBS and passed through a HisTrap column. The HisTrap column was washed with TBS containing 25 mM imidazole and the flow through and wash were collected. Fractions corresponding to SidK were pooled and exchanged into ion exchange buffer (50 mM Tris-HCl pH 7.4, 1 mM ethylenediaminetetraacetic acid (EDTA)) by concentration and dilution in a centrifuged concentrating device (EMD Millipore). SidK was loaded onto a HiTrap Q anion exchange column (GE Healthcare) and eluted with a gradient of 0 to 200 mM NaCl. Fractions corresponding to SidK were pooled and exchanged into TBS containing 5 mM DTT by concentration and dilution in a centrifuged concentrating device (EMD Millipore).
To purify substoichiometric V-ATPase:SidK assemblies, SidK from after TEV-cleavage was added to detergent-solubilized yeast vacuolar membranes and the V-ATPase:SidK complex was purified as described previously [23] except with twice the amount of washing. To purify fully-bound V-ATPase:SidK3 assemblies, TEV-cleaved SidK purified from the second HisTrap column (see protein purification above) was added to detergent-solubilized yeast vacuolar membranes and the protein complex was purified as described above.
ATPase activity assays were performed as described previously [23] in assay buffer (50 mM Tris-HCl pH 8, 0.05% [w/v] n-dodecyl β-D-maltoside (DDM), 3 mM MgCl2, 1 mM DTT, 0.2 mM NADH, 10 U pyruvate kinase, 25 U L-lactic dehydrogenase, 1 mM phosphoenolpyruvate, 2 mM ATP). Proteins were incubated on ice until assaying. 160 μL reactions were performed at room temperature using a 96-well plate in a Spectramax M2 UV/visible light plate reader (Molecular Devices). Four replicate experiments were conducted for enzyme with SidK or enzyme with buffer using the same stock of freshly purified V-ATPase or frozen ATP synthase. The final concentrations of enzyme and SidK were ~1 nM and ~30 nM, respectively. The concentration of bafilomycin used was 6 μM. In order to assay multiple constructs of SidK simultaneously for S5 Fig, samples had to be frozen and thawed to coordinate experiments, which decreases the ability of SidK to inhibit V-ATPase.
Yeast strain BY4741 [61] was transformed by the standard lithium acetate method [62] with plasmid p425GPD [63] carrying wild-type SidK or SidK mutants. The resulting strains were grown to saturation overnight at 30°C in Leucine dropout minimal medium (pH 5.5) and diluted at 1:40 into either minimal medium at pH 5.5 or minimal medium buffered with 50 mM MES and 50 mM MOP to pH 7.0. The cultures were incubated with vigorous shaking for 24 hrs at 30°C and yeast growth was monitored spectrophotometrically at 600 nm. Yeast cell lysates for western blotting analysis were prepared as described previously [25].
Fluorescence emission spectra of 2',3'-O-(2,4,6-trinitrophenyl) adenosine 5'-triphosphate (TNP-ATP) from 485 nm to 600 nm (10 nm slit width) were recorded with a Quantamaster QM-80 spectrofluorometer (Photon Technology International) using an excitation wavelength of 465 nm (4 nm slit width). The temperature was maintained at 10.0°C with a Peltier unit. Samples (0.5 mL) contained 0.25 μM V1, with or without 2.5 μM SidK, and varying amounts of TNP-ATP in assay buffer (25 mM Tris-HCl, pH 7.9, 0.15 M NaCl, 1 mM MgCl2, and 0.5 mM DTT). For each titration reading, 20 μL of the reaction sample was removed and replaced with 10 μL of trinitrophenyl-ATP (TNP-ATP) at different concentrations and 10 μL of 0.5 μM V1 in 2× assay buffer with or without 5.0 μM SidK. Fluorescence curves were corrected by baseline subtraction using the curve for [TNP-ATP] = 0. No significant fluorescence from the protein was observed. The increase in fluorescence due to addition of TNP-ATP was modelled in GNUplot (www.gnuplot.info) with the fit command using the equation [64,65]
I=[ES][ET]Ib+m([S]−[ES])
(1)
Where I is the fluorescence intensity, Ib is the fluorescence intensity of total enzyme bound to substrate, [ET] is the total enzyme concentration, [ST] is the total substrate concentration, and m is the increase in fluorescence per unit increase in free substrate. [ES] is the concentration of enzyme bound to substrate given by:
[ES]=0.5{[ET]+[ST]+Kd−([ET]+[ST]+Kd)2−4[ET][ST]}
(2)
where Kd is the equilibrium dissociation constant given by Kd = [E][S]/[ES]. The m parameter was estimated from fitting a straight line through the points corresponding to the four highest [TNP-ATP] values in each titration dataset. Each titration dataset was then modeled with fixed m to calculate Ib and Kd. The experimental setup described here is similar to the study by Kubala et al. [65]. However, the tight binding affinity between TNP-ATP and the V1 subcomplex allowed parameters to be estimated from the titration data instead of from separate experiments. This experimental approach is advantageous in situations where protein availability is limited, which was the case in this study. Five replicate experiments were performed for each condition (with and without SidK) using the same batch of freshly purified V1 subcomplex.
3 μL of ~10 mg/mL V-ATPase:SidK sample was applied to nanofabricated holey carbon grids [66] previously glow discharged in air for 2 mins. Excess sample was blotted away and the grid was plunge-frozen in a liquid propane-ethane mixture using a modified Vitrobot Mark III grid preparation robot (FEI company). Vitrified samples were imaged with a FEI Tecnai TF20 electron microscope operating at 200 kV and 34,483× magnification, resulting in a pixel size of 1.45 Å/pixel. Images were collected on a Gatan K2 Summit direct detector device operating in counting mode. 15 s movies were recorded at 2 frames/s and 5 e-/pixel/s. Movie frames were aligned using alignframes_lmbfgs and averaged with shiftframes [67]. Contrast transfer function (CTF) parameters of the averaged images were measured with CTFFIND3 [68] and corrected for magnification anisotropy using star_fixmaganiso [69]. Candidate particle image coordinates were identified automatically in averaged images with TMaCS [70] using templates that were 2D projections of an existing map of the V-ATPase that had been low-pass filtered to 20 Å [23]. Candidate particle images were extracted from the raw movies and corrected for local drift using alignparts_lmbfgs [67]. The aligned and averaged particle images were corrected for magnification anisotropy with correctmaganisotropy_fspace_list [69] and processed by 2D and 3D classification and 3D refinement in Relion [71]. To minimize the possible influence of SidK on the classification algorithm, image classification was performed with masked maps to focus classification on the V1 region.
Maps were visualized and segmented in UCSF Chimera [72]. Atomic models of the V-ATPase (PDB 3J9T, 3J9U, 3J9V) were docked rigidly into the density maps using UCSF Chimera and flexibly fit into the maps with Molecular Dynamics Flexible Fitting (MDFF) [73]. Flexible fitting in MDFF was performed with implicit solvent and using backbone atoms only while restraining secondary structure. The atomic model of SidK was docked into the V-ATPase:SidK3 map using UCSF Chimera by rigidly docking alpha-helical bundles I and II/III into the map as independent domains. The domains were joined in UCSF Chimera and the geometry of the short connecting loop between the domains was optimized using Modeller [74]. The secondary structure of SidK was predicted using the online server JPred [75].
Where appropriate, results were analyzed using an unpaired, two-tailed Student's t-test (TTEST function in Microsoft Excel 2007) to calculate a p-value. Values less than 0.05 were deemed statistically significant.
|
10.1371/journal.ppat.1004510 | Succinate Dehydrogenase is the Regulator of Respiration in Mycobacterium tuberculosis | In chronic infection, Mycobacterium tuberculosis bacilli are thought to enter a metabolic program that provides sufficient energy for maintenance of the protonmotive force, but is insufficient to meet the demands of cellular growth. We sought to understand this metabolic downshift genetically by targeting succinate dehydrogenase, the enzyme which couples the growth processes controlled by the TCA cycle with the energy production resulting from the electron transport chain. M. tuberculosis contains two operons which are predicted to encode succinate dehydrogenase enzymes (sdh-1 and sdh-2); we found that deletion of Sdh1 contributes to an inability to survive long term stationary phase. Stable isotope labeling and mass spectrometry revealed that Sdh1 functions as a succinate dehydrogenase during aerobic growth, and that Sdh2 is dispensable for this catalysis, but partially overlapping activities ensure that the loss of one enzyme can incompletely compensate for loss of the other. Deletion of Sdh1 disturbs the rate of respiration via the mycobacterial electron transport chain, resulting in an increased proportion of reduced electron carrier (menaquinol) which leads to increased oxygen consumption. The loss of respiratory control leads to an inability to recover from stationary phase. We propose a model in which succinate dehydrogenase is a governor of cellular respiration in the adaptation to low oxygen environments.
| This work establishes the principle that Mycobacterium tuberculosis undergoes a metabolic remodeling as oxygen concentrations fall that serves to decrease its rate of oxygen consumption and therefore oxidative phosphorylation. Furthermore, cells can be stimulated to respire, even in low oxygen conditions, by providing reducing equivalents to the respiratory chain by either genetic manipulation (deletion of succinate dehydrogenase) or by exogenous addition of reducing agents such as DTT. Thus, activation of persister cells may be accomplished by increasing their respiration rate in low oxygen conditions. These findings will inform the design of novel drug screens which should seek enhancers of cellular respiration to find compounds which will serve to shorten the duration of TB chemotherapy.
| The World Health Organization has estimated the prevalence of Tuberculosis (TB) in the human population to be nearly two billion people. Although only a fraction of those individuals will ever display symptoms, TB is still a significant cause of worldwide mortality and was responsible for 1.3 million deaths in 2012 [1]. The organism responsible for this disease, Mycobacterium tuberculosis, owes its unqualified success as a pathogen to the ability to survive and persist in a human host, where it can evade immune surveillance and establish a sub-clinical infection. These latently infecting bacilli have the potential for reactivation in certain circumstances, as is commonly seen in HIV-induced immunosuppression [2]. In addition to immune evasion mechanisms found in some other chronic pathogens, M. tuberculosis appears to evade immunity by adopting a metabolically active but quiescent state during which cell division is limited [3]. In fact, the current antibiotic therapy regimen recommended by the WHO is multiphasic and is modeled around the presence of tolerant persister cells that are not cleared in the initial two months of treatment. The reliable occurrence of this subpopulation in clinical investigations has led to the addition of a continuation phase to the antibiotic course, which can last four months or more. Currently, the physiological adaptation which enables this organism to persist remains an area of active research, but targeting persisters should considerably improve the outcome of therapeutic efforts.
The inability to physically isolate a persister subpopulation without perturbing its labile state has prompted the adoption of a number of approaches to gain insight into the basis of the phenomenon. These models, which recapitulate a slowly- or non-dividing state in vitro, have revealed a number of interesting clues to persister physiology. It is important to note that M. tuberculosis is widely considered to be an obligate aerobe with the important stipulation that even though division is limited in anaerobic conditions, bacterial cultures can remain viable for decades [4]. A model developed by Wayne was instrumental in delineating the oxygen set points which result in cessation of division below 1% dissolved oxygen (DO), and a decline in survival below 0.06% DO, thus providing a framework to understand dormancy [5]. More recently, Gengenbacher et al. found that when quiescence is initiated through nutrient starvation the bioenergetic remodeling results in a decrease of ATP to one-fifth its log phase level, a concentration which apparently is reflective of maintenance of the protonmotive force [6]. Watanabe et al. subsequently verified these results and further noted that the depletion of ATP correlated with an apparent dearth of NAD+, at very low dilution rates in continuous culture [7]. The information gleaned from these studies directly informs the mechanistic descriptions of new TB drugs, including the diarylquinolone, Bedaquiline, a newly approved ATP-synthase inhibitor which is effective against dormant mycobacteria [8].
Although a number of studies have examined the transcriptional response of dormant cells, direct genetic evidence of metabolic genes essential for growth rate transitions was reported from studies of the abundance of specific mutants in transposon insertion libraries following alteration of the dilution rate in continuous culture. Among enzymes with a bioenergetic function, genes involved in energy metabolism (a putative succinate dehydrogenase), and a number of oxidoreductases were found to be important for this transition suggesting that the resumption of growth requires the benefits of oxidative phosphorylation [9]. It is difficult to point to a specific physiological adaptation which would be responsible for survival without knowledge of which of the diverse in vivo microenvironments might harbor persistent mycobacteria, but some groups have tried screening approaches aimed at addressing these questions in specific tissues [10], [11]. In terms of bioenergetic capacity, these studies revealed that one member of an operon containing a putative succinate dehydrogenase appeared to be essential for in vivo mycobacterial survival in a mouse model during the chronic phase of infection, a finding that was subsequently repeated using an analogous transposon-based screen [10], [12].
Oxidative flux through the TCA cycle is directly coupled to the electron transport chain via the oxidation of succinate and the corresponding reduction of membrane-localized quinones. Disruption of this activity would be a good strategy for control of growth in the energy limiting conditions that M. tuberculosis is thought to encounter in vivo [13]. This is an important consideration because ATP generation by oxidative phosphorylation is energetically much more efficient than ATP generation by substrate-level phosphorylation. M. tuberculosis has two operons annotated as succinate: quinone oxidoreductase - as well as a putative quinol:fumarate oxidoreductase which should be capable of succinate oxidation (see Annotation in Text S1). To date, the functional activity of these complexes has not been investigated, so we sought to understand their role in the transition from aerobic growth to persistence in M. tuberculosis. To this end, we targeted the two operons with homology to succinate dehydrogenase, which are encoded by Rv0247c-Rv0249c and Rv3316–Rv3319 (Figure 1) (or according to the convention of Berney et. al. as sdh1 and sdh2, respectively) [14], for further study.
In this work, we employed a combination of genetic, physiological and biochemical approaches to dissect the roles of Sdh1 and Sdh2 in the metabolic shiftdown of M. tuberculosis during adaptation to hypoxia. We report that Sdh1 (and not Sdh2) is the primary aerobic succinate dehydrogenase of M. tuberculosis. Deletion of this enzyme resulted in a number of bioenergetic deficiencies such as a major deficit in viability during stationary phase or during the chronic phase of infection in C3HeB/FeJ mice. The cause of this energetic insolvency was a peculiar mismanagement of oxygen consumption due to an imbalance in the redox state of the menaquinone pool. The Δsdh1 mutant consumed oxygen with close to perfect uncoupled kinetics, whereas wild type (wt) M. tuberculosis enacted an oxygen conservation strategy. The respiratory rate was dependent on the redox state of the menaquinone pool and respiration could be stimulated in non-respiring cells by adding exogenous reductant.
To determine the role of each enzyme complex, we prepared strains with null deletions of each in attenuated (mc26230) and virulent (H37Rv) strains of M. tuberculosis using specialized transduction (Table S1 and Complementation in Text S1) [15], [16]. For safety reasons, we relied on null mutants of attenuated strains for all assays in which virulence was not a primary focus. The resulting mutant strains displayed no observable differences in growth rate in media containing glucose or glycerol as a primary carbon source (Figure 2A, B), however we observed a growth defect for Δsdh1 when succinate was the sole available carbon source compared to the parent or Δsdh2 (Figure 2C). These results were consistent for virulent and attenuated strains. In addition, we observed a stationary phase exit defect in which Δsdh1 was unable to be rescued from two-month old cultures, and the sdh2 mutant grew poorly after a similar period (Figure 2D). The parental cultures or complemented strains exhibited no comparable decrease in growth rate or saturation even after eight months of stationary phase, indicating that these operons do not have perfectly redundant catalytic activities in vitro.
Succinate dehydrogenase catalyzes the two-electron oxidation of succinate to fumarate with a corresponding reduction of quinone to quinol, but physiologically, the succinate oxidation:fumarate reduction catalytic ratios are dependent on substrate concentrations and are critically dependent on the redox potential [17], [18]. Absolute pool sizes of metabolic intermediates are highly dynamic in living cells as a function of growth stage, pH, gas mixture, and temperature. As a result, the predominant direction of catalysis for each enzyme at any time cannot be inferred by annotation alone. In fact, the SDH reaction in mycobacteria should have an unfavorable free energy because the redox potential of menaquinone is lower than that of the succinate to fumarate reaction [19].
We evaluated gene function of the two sdh operons in a physiologically relevant context using a targeted metabolomic approach by analyzing differences in pool sizes of central carbon metabolites for cells in aerobic growth and in an anaerobic model [20]. Comparison of the mutant strains to the parental strain during aerobic growth revealed a significant 4-fold increase in intracellular succinate in Δsdh1 but no difference in Δsdh2. This was accompanied by a 0.5-fold decrease of malate concentration in Δsdh1 compared to the parental strain, whereas the Δsdh2 strain showed no difference; these data suggest a loss of succinate dehydrogenase activity in the Δsdh1 strain (Figure 3). Consistent with observations made by others [21], we detected an accumulation in the total intracellular succinate concentration of the parental strain of M. tuberculosis of 8-fold after 10 days of anaerobiosis, while the concentration in Δsdh1 increases only 1.5-fold during this span. Conversely, total malate concentration rises slightly in the wt strain (1.7-fold), while the Δsdh1 mutant shows a 7-fold increase. The accumulation of intracellular succinate is suggestive of an inability of this strain to perform succinate oxidation, but since total concentrations of α-ketoglutarate decrease, and glyoxylate, oxaloacetate and malate increase in hypoxia, a portion of this succinate is likely to be from the reported activity of isocitrate lyase [21]. Consistent with this, during hypoxia we observed significantly less accumulated succinate in the Δsdh1 mutant relative to the parent (whereas Δsdh2 had an intracellular succinate concentration higher than the parent) and malate concentrations were 2.2 (for Δsdh1) and 1.8-fold (for Δsdh2) increased, though these differences were not significant.
We next verified that the aerobic accumulation of succinate in the attenuated M. tuberculosis mutants was reflective of the condition in the virulent strain using the same method. During aerobic batch culture, H37RvΔsdh1 and H37RvΔsdh2 accumulated succinate in excess of the parental H37Rv strain, and this accumulation was corrected for in the complementing strain (mc27292) which constitutively expresses sdh1. This behavior is consistent with the complementation in the attenuated strains (see Figure S1, and Complementation in Text S1).
Based on these differences in metabolite pools, we analyzed the predominant direction of catalysis in the same aerobic and anaerobic models using stable isotope labeling (see Metabolomics in Text S1). Cells were grown in 7H9 medium supplemented with 10% OADC and labeled with [1,4-13C] aspartate in both four days of aerobic log phase growth and after twelve days in hypoxia using methods similar to those previously described [7]. We traced the fate of isotopically labeled carbon in TCA intermediates during aerobic growth (Figure S2A) and in hypoxia (Figure S2B) and determined the proportion of each labeled metabolite with respect to all isotopologues for each intermediate. The stable isotope labeling supported the classification of Sdh1 as an aerobic succinate dehydrogenase, but little difference in metabolite ratios was observed in strains lacking Sdh2 in these conditions.
We conclude from the metabolomic data that a functional reassignment should be considered for the operon encoded by Rv0247c-Rv0249c. We propose that Rv0247c-Rv0249c (Sdh1) encodes the primary succinate dehydrogenase of M. tuberculosis and the operon encoded by sdhCDAB (Sdh2) performs catalysis in an as yet undefined condition. To seek further support of this proposed classification, we analyzed gene expression of mutant strains in aerobic and hypoxic conditions (Table S2, and Methods in Text S1). Although no significant upregulation by the opposing sdh gene cluster was observed during aerobic growth, sdh1 is significantly upregulated in Δsdh2 during anaerobiosis. Genes in the sdh2 operon were not upregulated in Δsdh1 at either oxygen tension. This scheme is consistent with transcriptional data from oxygen-limited M. smegmatis that shows a 2-fold increase in sdh2 transcripts but a 30-fold decrease of sdh1 transcripts [14].
As preservation of a proton motive force (PMF) is an important component of anaerobic survival, we monitored CFUs of sdh mutant strains in aerobic and anaerobic conditions in the presence of sub-lethal concentrations of the protononophore carbonyl cyanide m-chlorophenyl hydrazone (CCCP). Whereas 10 µM CCCP had a bacteriostatic effect on normoxic cultures (Figure S3A), the same concentration of CCCP resulted in a loss of viability of greater than 3-logs at 35 days of treatment (Figure S3B). Both mutant strains were more susceptible to PMF inhibition than the parental strain, but we were unable to recover colonies from Δsdh2 cultures after 21 days. This data supports the conclusion that Sdh2 is the generator of the PMF in hypoxia, as we have previously observed in M. smegmatis [22]. Next, we assessed the contribution of each sdh mutant to aerobic respiration in bioreactors operating in batch mode and in continuous culture. Cells were inoculated into a bioreactor system in which DO concentration, optical density, midpoint redox potential and pH could be measured simultaneously and were monitored throughout the growth curve as oxygen was depleted by the organism. Surprisingly, the parental strain initiated down-modulation of its respiration rate at ∼40% DO, while Δsdh1 continued to respire unabated until the DO was entirely depleted (Figure 4A). Conversely, cells harboring a deletion of sdh2 consumed oxygen at a reduced rate and were able to modulate respiration as DO was depleted to ∼6%. These experiments revealed that constitutive overexpression of the complemented strain using the hsp60 promoter overcompensated the respiratory phenotype. In subsequent experiments, we found that the oxygen consumption curve could be complemented in a strain expressing sdh1 with a novel integrated tet-responsive promoter (Ptet2); at low levels of anhydrotetracycline inducer (25 ng/mL – see Figure S4A, Complementation in Text S1). We concluded that these induction levels reflect the concentration of active enzyme in each condition; therefore, perfect complementation would require levels of expression which closely match wt levels throughout the growth curve.
The increased oxygen consumption of Δsdh1 should result in an increased membrane potential and an increased growth rate. However, data collected during batch culture experiments revealed that the initial growth rate for Δsdh1 is actually slightly slower in aerobic conditions (Table 1); this rate decreases considerably once oxygen is depleted, yet the parental strain maintains faster population doubling times than either sdh mutant (Table 1, Figure S5). This apparent uncoupling of respiration from growth was further analyzed in a separate chemostat experiment. When cultures were grown with a 24 hr doubling time, Δsdh1 (but not the parental strain) was unable to maintain the growth rate at DO levels of 25%, 5%, or 1%, and consequently washed out of continuous culture (Figure 4B). Because the membrane potential (ΔΨ) is the major component of the PMF at neutral pH values, we assessed the membrane potential by measuring uptake of the lipophilic cation tetraphenylphosphonium (TPP+) [23]. In aerobiosis the ΔΨ was comparable between the wt and Δsdh mutants (53–66 mV) (Table 2). In hypoxia, the ΔΨ was considerably higher (30 vs. 18 mV) for the Δsdh1 mutant compared to the wt and Δsdh2 strains (Table 2).
Mycobacteria use menaquinone as their main electron carrier in the electron transport chain. Coupling succinate oxidation (E°′ ∼ +30 mV) to menaquinone reduction (E°′ ∼ −80 mV at pH 7) is an energetic challenge, because this reaction is endergonic [24]. A model to explain this conundrum posits that reversed electron transport across the cytoplasmic membrane can provide the energy required to drive the oxidation of succinate using the PMF [25]. This suggests that the increased respiration rate in the Δsdh1 strain is due to the absence of reverse electron flow and consequently an altered redox state of the quinone pool. To confirm the hypothesis that the respiratory rate is a function of the redox balance of the menaquinone/menaquinol pool we sought corroborating evidence using M. tuberculosis strains harboring deletions of the type II NADH dehydrogenases ndh and ndhA. These enzymes are thought to be the primary means of electron input in Mycobacteria [26] during aerobic growth. A down-modulation of oxygen consumption by these strains occurred at ∼50% and ∼10%, respectively (see Figure 4A). Complementation of ΔndhA was similar to that of the sdh enzymes with overcomplementation of oxygen consumption when ndhA was expressed episomally using a constitutive promoter (Figure S4B), further illustrating the necessity of “fine tuning” respiratory enzyme levels to achieve maximal growth. This finding supports the paradigm that enzyme activities that facilitate rapid reduction of the quinone pool serve to increase the respiratory rate in the wt strain (since their deletion reduces oxygen consumption), and fumarate reduction functions as a respiratory brake during aerobiosis by an opposing oxidation of the pool. The unexpected disparity in DO-sensitive modulation of respiration by the type II NADH dehydrogenases suggests a wider strategy to indirectly sense oxygen concentrations in the immediate environment and spend reducing equivalents accordingly before taking the drastic step of uncoupling biomass production from respiration.
The apparent diminution of succinate oxidation in Δsdh1 during aerobiosis, and its uncontrolled respiratory phenotype alluded to an imbalance in the redox state of the menaquinone pool. We sought to confirm this biochemically by extracting menaquinones (MK-9) from cells growing aerobically, and at 1% DO in bioreactors (see Methods). Ratios of menaquinol:menaquinone of the parental strain were balanced when grown aerobically, but heavily skewed toward the oxidized state at low DO, conversely Δsdh1 had higher concentrations of menaquinol (reduced form), which was sufficient to drive respiration even at low oxygen levels (Figure 5A). In aerobically growing cells, we found the quinone pool to be balanced (ratioMK-9red/MK-9oxid = 0.87), indicating equilibrium between respiratory rate and carbon flux. Because the balance of quinone reduction can shift rapidly, we sought further confirmation by monitoring data from a probe for midpoint redox potential. Cultures were grown in a bioreactor running in batch mode as described above but flowed compressed air into the bioreactor at 1L/hr. Using this measure, M. tuberculosis can be seen to utilize available oxygen then switch off respiration until oxygen builds up to a threshold concentration before switching on aerobic respiration again. Importantly, increases in the redox potential precede the onset of oxygen consumption by several minutes during which pH does not change (Figure S6 and S7); supporting the hypothesis that oxygen consumption is managed by quinone redox balance. Δsdh2 behaves in a manner similar to wt, but Δsdh1 appears to maintain a negative midpoint redox potential and respires all available dissolved oxygen without allowing it to build up in the vessel (Figure S8).
The above behavior is consistent with previous reports that respiratory rate can be directly controlled with first-order kinetics by the degree of reduction of the quinone pool in membrane vesicles and mitochondria [27], [28]. We took advantage of the relatively low midpoint redox potential of menaquinone [29], and sought evidence that the respiratory rate of intact mycobacterial cells could be stimulated using the membrane permeable reducing agent dithiothreitol (DTT). We hypothesized that cells which have entered the phase of respiratory downshift brought on by low oxygen concentrations should be stimulated to respire if menaquinol can be replenished by an exogenously applied reducing agent. To test this, M. tuberculosis strain mc26230 was grown to early stationary phase and oxygen consumption was monitored in a Clark-type oxygen electrode (see supplementary methods) with and without the addition of DTT (Figure 5B). Stimulation of oxygen consumption was observed up to concentrations of 40 mM reductant, after which little increase was observed. Importantly, no stimulation of oxygen consumption was noted in media alone or in preparations of heat-killed cells from the same culture. No synergistic increase in oxygen consumption was observed in log phase cells in similar experimental conditions when the starting DO of culture media was greater than 50% that of aerated media, i.e. cells that are already respiring at maximal rates are not induced to respire faster by the addition of reductant.
To assess the effect of disregulated respiratory activity on pathogenesis and persistence, we tested the ability of Mtb Δsdh1 and Δsdh2 to cause disease in several established murine models. Previous experiments utilizing a high-throughput genetic screen have revealed subunits of sdh1 (but not sdh2) to be underrepresented in the lungs of C57Bl/6J mice during chronic infection [10], [30]. It is not clear if the fitness defect observed in those screens is the result of reduced virulence or an inability of the mutants to maintain their numbers during chronic infection, but we were unable to recreate this phenotype with null deletion strains using the C57Bl/6J mice (Figure S9). To assess virulence, we infected immunodeficient Rag-1−/− mice via low-dose aerosol; these mice produce no mature T or B cells and are thus unable to control mycobacterial infection [31]. Whereas immunodeficient mice infected with H37Rv had a median survival time of 26 days, Δsdh1 infected mice had a slightly longer median survival time of 29 days. Interestingly, the Δsdh2 strain displayed a hypervirulent phenotype; these mice had a median survival time of only 22 days (Figure 6A). The overexpressing complemented strains for both of these deletions were less lethal than either the mutants or the parental strain.
Given the predictive constraints of the mouse model in TB infection, particularly the inability of the murine immune system to form fibrous caseous granulomas [32], we think that any impact of these mutations on survival (or strains harboring deletions in respiratory enzymes) could be lessened because oxygen levels are likely always sufficient for growth in the murine lung. A murine model was developed to address this limitation; the C3HeB/FeJ mouse is an inbred strain that develops fibrous encapsulated lung lesions post-aerosol infection which appear to contain hypoxic centers [33]. We reasoned that the respiratory mismanagement of Δsdh1 would lead to a survival deficit in the lesions of mice containing hypoxic lesions. To test this hypothesis, we infected C3HeB/FeJ mice via aerosol and monitored burden over time (Figure 6B). By twenty weeks of infection, Δsdh1 had tenfold fewer cells per lung than the H37Rv parent (5.79 log10 CFU vs. 6.67 log10 CFU) and Δsdh2 was similar to the wt. It is important to note that after nine weeks the Δsdh1 burden dropped slightly, while wt cells continued dividing until week twenty. This suggests that deletion of Sdh1 leads to an inability to maintain bacterial numbers in the host, however, the difference in bacterial burden between wt and Δsdh1 was not as dramatic as we would have expected based on our in vitro results. This might be explained by the fact that gross pathology of upper lungs at twenty weeks did not reveal encapsulated granulomatous) lesions (Figure S10), thus oxygen was likely not restricted in the lungs of these mice.
The bioenergetic program that sustains M. tuberculosis during latency and in models that recapitulate persistence is of great interest because this survival is likely due to inhibition of growth that stems from an idle metabolic state [3], [34]. A mechanistic understanding of quiescence is of crucial importance to the planning of new antitubercular compound screens, which can be designed to directly target this population. To this end, we sought to understand the function of the enzyme responsible for the direct coupling of anabolism via the TCA cycle and the electron transport chain - succinate dehydrogenase. Prior to this work, the individual roles of the two predicted succinate dehydrogenases of M. tuberculosis had not yet been experimentally determined, and no obvious phenotype was reported in M. tuberculosis H37Rv containing a null deletion of the hypoxia-upregulated fumarate reductase, frdABCD [7]. Genetic manipulation of M. tuberculosis followed by an intracellular metabolomic approach allowed us to probe the functions of the two annotated Sdh enzymes and their role in cell physiology. Importantly, these enzymes were found to strongly influence aerobic respiration, and deletion of sdh1 resulted in an increased rate of respiration, but did not result in faster cell growth. The work presented here validates the predicted role of sdh1 as the primary succinate dehydrogenase during aerobiosis
It has been almost eighty years since Loebel and colleagues formally noted the capacity of M. tuberculosis for curtailing its oxygen consumption under anaerobic or starvation conditions [35], but a mechanism for this phenomenon is absent from the literature. Two distinct phases of adaptation to decreasing oxygen tension have been described; NRP (non-replicating persistence) stage 1 - marked by the cessation of cell division at ∼1% oxygen, and NRP stage 2 – a quiescent state occurring below 0.06% oxygen in which biomass production ceases [36]. Our data imply that M. tuberculosis employs an orchestrated respiratory slowdown as oxygen levels fall; this program is initiated while oxygen is still plentiful. The respiratory rate is fine-tuned by the opposing activity of the succinate dehydrogenase and fumarate reductase activities to maintain an optimal growth rate. This suggests that this tuning is controlled by balancing substrate concentrations, as has been suggested in electrochemical analysis of isolated enzymes [37], post-translationally [38], and via catabolite repression [39]. Management of respiration has important consequences for the proclivity for survival of M. tuberculosis amid a range of pathological niches in which oxygen tension can vary significantly, because ATP generation is much more efficient when electrons are committed to oxidative phosphorylation than through substrate-level phosphorylation alone.
We favor a simple mechanistic explanation for the controlled respiratory slowdown that is consistent with structural studies of the terminal cytochrome c oxidase complex and the progression of the Q-cycle (Figure 7) [40]–[42]. Organisms will respire at optimal rates with a balanced quinone pool in which quinol (reduced) is present in sufficient concentration to immediately occupy the center P of the cytochrome oxidase complex; but when quinol is limiting - in an oxidatively skewed pool - respiration will progress at a less-than optimal rate. Figure 5A shows that whereas the wt strain has a largely oxidized quinone pool at 1% DO, the Δsdh1 mutant maintains a balanced pool, resulting in unchecked oxygen consumption. These data support a mechanism for respiratory downshift in wt M. tuberculosis that works as follows: as oxygen concentration drops below 40–30%, succinate oxidation also decreases leading to its buildup (hypoxic cells accumulate a sevenfold increase in intracellular concentration). This ‘unrespired’ succinate does not contribute to the reduction of membrane menaquinones, and as the ratio of menaquinol:menaquinone decreases from the activity of other electron donors, the cytochrome oxidoreductase is deprived of its substrate, thus decreasing the rate of oxygen consumption. However, the loss of the primary means of succinate oxidation in Δsdh1 results in a premature accumulation of succinate during aerobiosis that is partially relieved by Sdh2, which can function as a succinate dehydrogenase when cytosolic substrate concentrations favor this reaction. Published measurements of ubiquinone pools in E. coli show a similar trend [43] in quinone poise as oxygen decreases; but in that facultative anaerobe, reduced quinone increases as cells approach anoxia. The observation that a strong reductant can stimulate respiration of poorly respiring cells (Figure 5B) provides additional evidence of a menaquinol-limiting respiratory scheme. To our knowledge, this is the first time that any group has shown that oxygen consumption can be stimulated in living organisms that have shut off respiration by provision of exogenous electrons.
The oxygen consumption profiles of the two sdh mutants revealed another interesting aspect of mycobacterial physiology, a downshift of respiratory activity was initiated by the parental strain in the range of 40–30% DO. The decline in the rate of respiration indicates that the organism switches to a less thermodynamically efficient mechanism for ATP production as oxygen levels drop – but are still sufficient for growth. Current understanding of the mechanics of the decline in respiratory activity exhibited by M. tuberculosis upon adaptation to anaerobic conditions has been guided by analysis of the transcriptome of cells as they pass into hypoxia in various models [14], [44], [45]. Here we report that M. tuberculosis accomplishes gross control of aerobic respiration by depriving the cytochrome c oxidoreductase of menaquinol via a slow electron flux through Sdh1 and demonstrate how carbon passing through the TCA cycle is subject to this mechanism that couples growth and electron transport (Figure 7). Importantly, this modulation in the rate of oxygen consumption occurs long before oxygen becomes limiting for growth [26], and is absent any exogenously-provided inhibitor of respiration.
This respiratory management scheme should have direct in vivo relevance considering that physiological oxygen levels are only a fraction of those commonly used in in vitro culture models, and reflect the point at which we observe a downmodulation of oxygen consumption [46]. This might explain the pathological preference of M. tuberculosis for the upper lobes of the lung [47] where mycobacterial cellular respiration can function more efficiently. As cells are carried into tissues farther from the lung epithelia, oxygen becomes scarce and cells are forced into a less efficient bioenergetic program which could lead to decreasing ATP production and more reliance on glycolysis, β-oxidation, or storage compounds to meet energy demand. Aerosol infection of C3HeB/FeJ mice using the Δsdh1 strain led to a tenfold reduction in CFUs in the lungs (Figure 6B). Given the limitations of the murine TB model [48], [49], and the lack of encapsulated hypoxic lesions we observed in lung sections, we believe that the consequences of the phenotype reported here would be even more pronounced in models that more closely resemble human pathology, such as the rabbit or guinea pig.
The niche in which latent M. tuberculosis survives, avoiding immune surveillance and maintaining undetectable cell numbers, is presently unknown. Several hypotheses have been suggested including the necrotic centers of granulomas [50], [51], adipocytes [52], and recently mesenchymal stem cells [53]. This latter work is especially interesting in light of the conclusions presented here. It implies that the oxidative burst experienced when invading M. tuberculosis is engulfed by an alveolar macrophage would serve to inhibit respiration by shifting redox balance – toward an oxidized quinone pool in which quinol becomes limiting for oxygen reduction. Since cells in NRP-2 maintain an energized membrane, and are notably tolerant to single antibiotics but retain sensitivity to some combinations [54], we think it is plausible that persistence is a function of the oxidative state of the milieu, and is the result of reduced respiratory flux. The presence of adequate oxygen alone is not sufficient to stimulate respiration; quinone redox homeostasis must be restored before respiration can reach optimal levels and the cell can take advantage of the energetic benefit of oxygen as its terminal electron acceptor.
The necessity for members of the M. tuberculosis complex to maintain two possible frd enzymes (sdhCDAB & frdABCD) may be an indication of a metabolic plasticity which enables them to simultaneously utilize multiple carbon sources with different oxidation states and divert this carbon either into biomass production or storage molecules during growth, or into energy production for maintenance of PMF and repair during non-growth states [55], [56]. The previously observed rerouting of a portion of carbon flux into the reductive C4 arm of the TCA cycle [7] suggests the involvement of fumarate reductase activity in hypoxia, indicating that other pathways contribute to anaerobic survival to some extent. Redundancy in Frd catalysis remains a possible explanation, and further genetic analysis will need to be performed to establish this, but thus far we have been unable to delete both sdh1 and sdh2 (or sdh2 & frdABCD) sequentially to address this hypothesis. Interestingly, Baek and Sassetti found that transposon mutagenesis of sdh2 led to an inability to shut down growth in hypoxia, thus if the primary means of succinate oxidation is through Sdh1, oxygen (but not carbon) limitation does not result in cessation of growth, and succinate dehydrogenase activity continues to push carbon through the TCA cycle to continue biomass production [57].
There is now widespread acknowledgement of the fact that a reduction in the duration of TB chemotherapy could be achieved by finding ways to target non-replicating M. tuberculosis. The recent FDA approval of Bedaquiline lends credence to the idea that non-replicating cells still remain susceptible to inhibitors targeting maintenance bioenergetics, albeit at a reduced rate compared to current effective drugs [8], [58]–[60]. In this communication, we propose that the removal of a metabolic block on M. tuberculosis respiration imposed by the contending action of the aerobic succinate dehydrogenase and fumarate reductase activities would prevent the orderly metabolic shift to quiescence. Compounds that serve to reduce quinones in non-dividing organisms would exhibit the pleiotropic effects garnered by increasing respiration, including enhancing membrane potential-driven uptake and decreasing fitness. Thus, progress toward the goal of shortening chemotherapy might be better served by searching for enhancers of respiration, which may reduce the numbers of organisms which are shifted to a persistent state.
Attenuated strains of M. tuberculosis were constructed by allelic exchange via specialized transduction [15] from the parental strain H37Rv. Null mutants in M. tuberculosis strains H37Rv, mc27000/mc26230 (ΔpanCD, ΔRD-1) [61], show identical growth characteristics in standard atmosphere as the parental strain (unpublished results). T-Coffee [62] was used to assess homology between enzyme subunits (Figure 1) and scores are presented as alignments of individual subunits corresponding to sdh2. For a full list of strains used in this work, see (Table S1). For CFU experiments, mycobacteria were grown to OD600 0.5 and subcultured into media containing antibiotic and incubated at 37°C in a shaking incubator, or shifted to an anaerobic chamber (<1 ppm O2) in bottles with vented caps and incubated shaking at 37°C. For growth experiments using single carbon sources, 7H9 media was supplemented with NaCl and BSA and individual carbon sources (see Supplementary Methods for more detail).
Analysis was performed using an Acquity UPLC system (Waters, Manchester, UK) coupled with a Synapt G2 quadrupole–time of flight hybrid mass spectrometer. Column eluents were delivered via Electrospray Ionization. UPLC was performed in HILIC mode gradient elution using an Acquity amide column 1.7 µm (2.1×150 mm) using a method previously described [63]. The flow rate is 0.5 mL/min with mobile phase A (100% acetonitrile) and mobile phase B (100% water) both containing 0.1% formic acid. The gradient in both positive and negative mode is 0 min, 99% A; 1 min,99% A; 16 min, 30% A; 17 min, 30% A; 19 min 99% A; 20 min 99% A. The mass spectrometer was operated in V mode for high sensitivity using a capillary voltage of 2 kV and a cone voltage of 17 V. The desolvation gas flow rate is 500L/h, and the source and desolvation gas temperature are 120 and 325°C. MS spectra were acquired in centroid mode from m/z 50 to 1,000 using a scan time of 0.5 s. Leucine enkephalin (2 ng/µL) was used as lock mass (m/z 556.2771 and 554.2615 in positive and negative experiments, respectively). For further details, see Metabolomics in Text S1.
Measurement of oxygen consumption rate in M. tuberculosis was performed using a Clark-type oxygen electrode (Rank Brothers Cambridge, UK) with data collected using an ADC-24 data logger (Pico Technology, Cambridgeshire, UK). Cells were prepared in 490 cm2 roller bottles (HSR = 26), (Corning, NY). For culture densities below OD600 4.0, cultures were centrifuged for 5 minutes at 4,000 rpm and resuspended in fresh 7H9 media from which catalase was omitted. To detect induction of oxygen consumption by reductants, 5 mL early stationary phase cells (OD600 5.0) were added to the incubation chamber and basal oxygen consumption was monitored for 100–200 seconds, at which point compound was added. After 200 seconds, maximal uncoupled oxygen consumption rate was determined by the addition of 20 µM CCCP for 100 seconds.
We grew mycobacterial strains as described above in media containing OADC and the appropriate antibiotic for two passages before a single passage in media in which antibiotic was omitted immediately prior to animal infection. Female C57BL/6 mice, Rag-1−/−, and C3HeB/FeJ mice were obtained from Jackson Laboratory. Rag-1−/− mice were infected with ∼1×106 CFU of virulent mycobacteria via high volume tail vein injection. C57BL/6 mice and C3HeB/FeJ mice were infected via aerosol from a suspension of bacterial culture in PBS containing 0.05% Tween 80 and 0.004% antifoam, which yielded ≈100 or ≈50 cfu per lung. Four mice from each infection group were killed 24 h post-exposure, and lung homogenates were plated on 7H9-agar plates to determine the efficiency of aerosolization. We determined bacterial loads in lungs and spleen by plating for CFU at the indicated times from four mice per infection group. Five mice from each group were also used to determine survival times of infected mice. Pathological analysis and histological staining of organ sections were done on tissues fixed in buffered 10% formalin. Mouse protocols used in this work were approved by the Institutional Animal Care and Use Committee of Albert Einstein College of Medicine.
Mouse studies were performed in accordance to National Institutes of Health guidelines using recommendations in the Guide for the Care and Use of Laboratory Animals. The protocols used in this study were approved by the Institutional Animal Care and Use Committee of Albert Einstein College of Medicine (Protocol #20120114)
NP_214761 (Rv0247c), AFN48101 (Rv0248c), CCP42978 (Rv0249c), CCP46136 (SdhC), CCP46137 (SdhD), CCP46138 (SdhA), CCP46139 (SdhB), NP_216370 (Ndh), CCP43122 (NdhA)
|
10.1371/journal.pntd.0005253 | Genome-Wide Analyses of Individual Strongyloides stercoralis (Nematoda: Rhabditoidea) Provide Insights into Population Structure and Reproductive Life Cycles | The helminth Strongyloides stercoralis, which is transmitted through soil, infects 30–100 million people worldwide. S. stercoralis reproduces sexually outside the host as well as asexually within the host, which causes a life-long infection. To understand the population structure and transmission patterns of this parasite, we re-sequenced the genomes of 33 individual S. stercoralis nematodes collected in Myanmar (prevalent region) and Japan (non-prevalent region). We utilised a method combining whole genome amplification and next-generation sequencing techniques to detect 298,202 variant positions (0.6% of the genome) compared with the reference genome. Phylogenetic analyses of SNP data revealed an unambiguous geographical separation and sub-populations that correlated with the host geographical origin, particularly for the Myanmar samples. The relatively higher heterozygosity in the genomes of the Japanese samples can possibly be explained by the independent evolution of two haplotypes of diploid genomes through asexual reproduction during the auto-infection cycle, suggesting that analysing heterozygosity is useful and necessary to infer infection history and geographical prevalence.
| Strongyloides stercoralis, one of the most neglected helminths causes strongyloidiasis mainly in tropical and subtropical regions worldwide. The parasite’s complex lifecycle includes sexual and asexual reproduction outside and inside the host, respectively. The parasite can also asexually complete a life cycle within the host's body, which is called autoinfection causing life-long infection. In order to investigate the population structure and transmission patterns of this parasite we sequenced individual nematodes isolated from human faeces in Japan and Myanmar, where the parasite is present at low and high frequencies, respectively. Whole genome sequencing of small parasites is generally difficult because the amount of DNA is limiting. However, we overcame this problem by combining whole genome amplification with next-generation sequencing. Sequence comparisons revealed 0.6% of the genome is variable among samples, and the variants showed clear separation by the location of their origin. We found that heterozygosity within the genomes was higher in Japan, which is likely explained by the predominance of asexual reproduction through auto-infection, suggesting that analyses of heterozygosity are required to better understand the history of a population.
| The helminth Strongyloides stercoralis, which is one of the most common and globally distributed human pathogens of clinical importance, infects 30–100 million people worldwide [1,2]. This parasite most often resides in areas with tropical or subtropical climates and less frequently in areas with a temperate climate. It occurs infrequently in societies where faecal contamination of soil or water is rare, and therefore, new infections are very rare in countries with developed economies [3]. However, infection can persist for life unless effective treatment eliminates all adult parasites and migrating auto-infective larvae. Therefore, carriers are present in developed countries, representing a potential risk of horizontal transmission among humans [4]. Strongyloides stercoralis is also a natural parasite of dogs [5].
Strongyloides stercoralis is the only medically important nematode that can multiply in the host via an auto-infection cycle to reach critical levels and cause death [1,6,7]. The complex life cycle includes sexual and asexual reproduction. Infection with S. stercoralis begins when the infective third-stage larvae (iL3) in soil attach to and penetrate the human skin. After reaching the lung through the bloodstream, the parasites ascend to the trachea, and are swallowed to settle in the small intestine (their final destination) where the parasitic adults produce eggs through parthenogenesis. The larvae passed in the host faeces develop via either the homogonic route into iL3 forms or the heterogonic route into free-living adult stages that reproduce sexually outside the host. Although most eggs/larvae of the parasite are excreted from the host with faeces, homogonic larval development may occur inside the small intestine giving rise to auto-infective L3 which penetrate the intestinal wall and invade the tissues, ultimately entering the lung and returning to the small intestine to complete development to the parasitic female. In this circumstance, termed auto-infection, repeated generations of development may take place within a single host. [5]. Although strongyloidiasis is usually an indolent disease in immunocompetent hosts, it can cause a hyperinfective syndrome (disseminated strongyloidiasis) in immunocompromised hosts through the reproductive capacity of the parasite inside the host. Disseminated strongyloidiasis, if untreated, is associated with mortality rates of approximately 90% [8].
Despite its great medical importance, the threadworm S. stercoralis, is one of the most overlooked helminths [1]. The parasite's complex life cycle has long been considered a major impediment to attempts to control strongyloidiasis. Recently, the genome of S. stercoralis was sequenced and compared with other species of Strongyloides [9]. This comparative genomic study illuminates the use of genome-wide analysis to identify genes related to parasitism, to investigate diversity and population structures, and to determine the transmission route of S. stercoralis. Here, we aimed to determine the intra-species genomic variations of S. stercoralis present in Japan and Myanmar, which differ in socioeconomic status, history of infection and prevalence of this nematode.
The Ethics Committees of the University of the Ryukyus and the University of Medicine-1 Yangon approved this study. Participants, who were informed of the study's aims and procedures, provided written informed consent. All individuals infected with S. stercoralis were treated with ivermectin.
Faecal samples were collected in 2014 (Table 1) in Okinawa, Japan, representing an area where S. stercoralis is non-prevalent and where S. stercoralis has not been endemic for at least the last 50 years [10], and Htantabin, Myanmar as a prevalent area where new infections frequently occur. In Okinawa, Japan, faecal tests were performed for inpatients in one hospital and residents of two elderly nursing homes located in the southern part of Okinawa. For Myanmar samples, a community survey was conducted in three different villages of Htantabin area. Faeces were incubated on 2% (w/v) agar plates at 25°C for 2–4 days. This culture condition would allow a portion of parasites to undergo a complete free-living generation involving a sexual cross although worms may mate with their genetically identical siblings in the culture. Individual nematodes (iL3) that crawled out of the faeces were transferred to 0.2 ml tubes containing 10 μl of worm lysis solution (9 μl Direct PCR [Viagen], 0.5 μl of 20 mg/ml Proteinase K [Qiagen] and 0.5 μl of 1 M dithiothreitol [Wako]). The lysates were incubated at 60°C for 1 h and then at 95°C for 10 min. To identify nematodes, the 18S ribosomal RNA gene was amplified using 0.1 μl of worm lysate with the primers 988F and 1912R [11], and the amplicons were sequenced using an ABI 3130 sequencer (Applied Biosystems) with the BigDye Terminator v3.1 kit. Worm lysates were immediately used for further analysis or stored at −30°C.
Genomic DNA was amplified from 1 μl of worm lysate using an Illustra GenomiPhi V2 kit (GE Healthcare) according to the manufacturer’s protocol. Amplified products were quality-checked using 1% agarose gel electrophoresis, purified using a QIAamp DNA Mini Kit (Qiagen) and quantified using Qubit (Life Technologies).
Libraries were constructed using a Nextera DNA Sample Prep Kit (Illumina) with 100 ng of amplified DNA according to the manufacturer’s protocol. The libraries were sequenced using an Illumina MiSeq with a v3 Reagent kit (600 cycles) according to the manufacturer’s recommended protocol (https://icom.illumina.com/) to produce 300-bp paired-end reads to obtain ~3G base data. Non-WGA reads of the genome reference strain (SSTP) were obtained from NCBI SRA under accession number ERR066168, randomly sampled and used as a reference to evaluate WGA reads.
We used Trimmomatic [12] to eliminate adapter contamination from the reads and achieve a minimum quality score = 15 (SLIDINGWINDOW:4:15) before mapping against the S. stercoralis reference genome (ver. 2.0.4) [9] using SMALT v0.7.4 (https://www.sanger.ac.uk/resources/software/smalt/) with options–x (each mate is mapped independently) and–y 0.8 (mapping to the region of highest similarity in the reference genome at a similarity threshold > 80%). Duplicate reads were marked using the Picard tool (ver. 1.95), and indels were realigned with GATK (version ver. 3.3.0) [13] using the IndelRealigner. Variants were then called using GATK HaplotypeCaller. Variants were annotated using GATK and ANOVA (ver. 2014-11-12). Depth of coverage was calculated by counting mapped reads per site using GATK DepthOfCoverage [13]. Analysis of population genetics, including calculating nucleotide diversity (π) and inbreeding coefficient (FIN), were performed using vcftools (v0.1.12b) [14]. Mean of per-site nucleotide diversities between two genomes were reported as a pair-wise genome distance. Analysis of molecular variance (AMOVA) was conducted with R Poppr package [15]. Other statistical analyses were performed using R (ver 3.1.1) and in-house python scripts. In the previous study, using C. elegans as a model, we found WGA variant calls with low coverage data tends to call heterozygous loci homozygous [16]. To avoid this bias toward calling homozygous sites, we excluded relatively low coverage samples comprising < 70% of genomic regions with 15× depth (nematodes designated MyHTB122-6, Rk5-6, Rk6-4, Rk7-5, Rk8-3 and Rk8-8) from the heterozygosity-related analyses.
Principal component analysis (PCA) was performed using R (ver 3.1.1) implemented with SNPRelate package [17]. Bi-allelic SNPs were extracted from full variant information of all the samples and used for PCA analyses.
The mitochondrial genomes of Rk4-1 nematodes were reconstructed from the Illumina reads using MITObim ver 1.6 [18]. In the first step, Illumina reads were mapped to the S. stercoralis reference sequence (Genbank accession No. NC_028624) to generate a seed for the second step. In the second step, gaps and ambiguous regions in the seed were replaced by iterative mapping that was repeated until all gaps were closed, and the number of reads remained constant. Reconstructed mitochondrial sequences were refined by correcting bases using ICORN2 [19], and the assembly was used to represent the Japanese nematode reference mitochondrial genome.
Nucleotide sequences of SNP positions in scaffolds > 30 kb, which accounted for 96% of the total genome assembly, were extracted from the vcf files and were used to construct phylogenetic networks based on similarity/dissimilarity with the Neighbor Net method of SplitsTree4 [20]. Computational phasing of the diploid genotypic data was performed using SHAPEIT2 with its default parameters [21]. Phased sequence data from all samples were used to create a separate Maximum Likelihood tree using FastTree (ver 2.1.8) for each scaffold > 30 kb [22].
To generate a mitochondrial-based phylogeny, reads from each nematode sample were mapped to the Japanese parasite's reference sequence (see above) using SMALT v0.7.4, and SNPs were called using GATK [13]. The nucleotide sequences of the SNPs were extracted and used to generate Maximum Likelihood trees using FastTree (ver 2.1.8) [22].
All sequence data were submitted to the DDBJ Sequence Read Archive (DRA) under project accession number PRJDB5112.
We re-sequenced the genomes of 33 S. stercoralis nematodes collected in Myanmar (prevalent region, nine from three patients) and Japan (non-prevalent region, 24 from six patients) [10] (Table 1). We applied the WGA method [16] using the Illumina MiSeq to sequence the whole genome of a single nematode. We obtained 300-bp paired-end reads to > 20× coverage (> 3 Gb) for each nematode and mapped them to the S. stercoralis reference genome. The mapping ratios of each sample to the reference genome ranged from 77.46% to 96.96%, and the ratios for reads mapped in the correct orientation and distance (‘proper paired’ reads) ranged from 48.94% to 62.72% (S1 Table). In contrast, the mapping ratios of non-WGA reference reads were 90.95% with 71.79% proper pairs (S1 Table, S1 Fig). Although amplification bias depending on genome locations were observed in the WGA samples (S2 Fig), > 10× coverage was achieved for > 80% of the genomic locations, and the median coverage values ranged from 20 to 50 for most samples (S1 Table, S1 Fig).
We detected 298,202 variant positions, which accounted for 0.6% of the total genome, among the 33 samples when compared with the reference. Most variants were SNPs (231,583 positions), and small inserts or deletions (indels) were present at 67,655 positions (S2 Table). The number of variant positions in individual nematodes (including homozygous and heterozygous sites compared with the reference) ranged from 137,439–146,259 and 135,583–157,900 of the Myanmar and Japanese samples, respectively (S2 Table).
Comparisons with reference gene models revealed that 27.7% of the variants were located in intergenic regions, followed by 27.3%, 15.2%, 12.8% and 9.9% in exonic, upstream, downstream and intronic regions, respectively (Fig 1A). There were higher frequencies of variant positions in intergenic regions compared with those of the individual nucleotides in the total genome and lower frequencies of variant positions in exonic regions (Fig 1A).
In the exon variations, similar numbers of synonymous and non-synonymous SNPs were detected in 34,551 and 34,960 positions, respectively (Fig 1B). Frameshift indels and stop mutations were less frequent (5,932, 1,237, 1,158 and 119 for frameshift, non-frameshift, stop-gain and stop-loss, respectively) (Fig 1B).
The distribution of SNPs along the four longest scaffolds is shown in S3A Fig and the distribution of numbers of SNPs by 10-kb window for scaffolds bigger than 100 kb are shown in S3B Fig. Variants were unevenly distributed along the genome with numbers of variant positions in 10-kb window ranging from 3 to 922 (median = 31), suggesting that they represented ‘hotspots’. Further, the hotspot regions did not correspond to regions of high coverage mapping (S2 Fig and S3 Fig) (Pearson’s r = -0.01), suggesting that the variant call was not significantly influenced by WGA amplification bias. No significant differences in SNP distribution between the two countries were observed (high correlation coefficient between SNP numbers in 10-kb window of the two countries; Pearson’s r = 0.78, p < 2.2e-16).
Principal component analysis (PCA) of SNPs compared with the reference strain unambiguously separated the Japanese and Myanmar samples from the reference strain by the first PC, which account for 40.1% of the variance. Japanese and Myanmar samples were separated by the second PC (14.1% of variance) (Fig 2A). Fig 2B shows the PCA results without the reference. The Myanmar and Japanese samples were separated by PC1 (28.4%). PC2 (10.3%) grouped the Myanmar samples according to their host origins, although the separation in the Japanese samples was not unambiguous.
Pair-wise distances (π) of samples originated from different countries (Japan vs. Myanmar) were generally higher than those within populations (Fig 3). In the Myanmar samples, pair-wise distances between hosts were higher compared to those within hosts, although such differences were not observed in the Japanese samples (Fig 3). Because the parasitic adult stage of Strongyloides is mitotically parthenogenetic, multiple larval progeny of such adults will be, in theory, genetically identical. Although within-host samples showed high similarity to each other (π values < 7.5e-04) both in Japan and Myanmar, they still exhibited some differences from each other. Because of possibility of errors in WGA or sequencing process and difficulty in heterozygous SNP call, it is difficult to conclude that they are genetically different or identical progeny. Simulated experiments using proved progeny of single adults will be useful to answer this question.
Analysis of molecular variance (AMOVA) showed 23.7% of variance was associated with differences between populations and 6.3% with differences between hosts, whereas more than 100% of the variance was attributed to variation within samples (S3 Table). Although the negative phi-statistics and variance values observed in AMOVA (S3 Table) may reflect problems with sample size and analytical strategy, these results suggest a close relationship among the Japanese samples independent of host origin and high heterozygosity within the individual genomes.
Next, we constructed phylogenetic networks according to the SNPs, which support the PCA results (Fig 4). The tree contained two main clades, comprising Myanmar or Japanese samples. All samples in the Myanmar clade from the same host clustered together and were clearly distinct from those of other hosts. Most Japanese samples sub-clustered according to host origin, although the separations were not as clear as those of the Myanmar samples. Further, we found some Japanese samples (Rk5-6, Rk7-5, Rk8-3 and Rk8-8), which have lower coverage (S1 Table), were placed at positions distant from those of other worms of the same host origin. This is likely because of failure to call heterozygous SNP in low coverage samples [16]. We therefore removed these four samples (Rk5-6, Rk7-5, Rk8-3 and Rk8-8) and those having lower coverage than the four samples (based on % of genome regions with 15× coverage; MyHTB122-6 and Rk6-4) from further analyses. Two samples from host Rk9 (Rk9-3 and Rk9-11), which had higher coverage (S1 Table), occupied positions more distant from the other samples as well as a sample from host Rk9 (Rk9-6) (Fig 4A).
Next, we used the computationally-phased sequence dataset for the Japanese samples to construct phylogenetic trees for each scaffold (> 100 kb). The two haplotypes in a genome, shown as A and B haplotypes in S4A Fig, separated into distinct clusters for most of samples. This result suggests that the haplotypes in the diploid genomes of most samples evolved independently. Haplotypes of samples Rk8-7, Rk9-3 and Rk9-11 exhibited distinct haplotype organisations in each-scaffold tree (shown in black colour in S4A Fig). In Myanmar samples segregation of two haplotypes was not clear compared to the Japanese samples and individual scaffold trees showed various patterns (S4B Fig), suggesting past occurrences of chromosome exchange and/or recombination between Myanmar samples.
The mitochondrial tree exhibited a similar topology to the nuclear tree (Fig 4B). The Japanese samples were placed into one clade, clearly separated from Myanmar samples with a high support value. Within the Japanese samples, those from host Rk9 clustered with those from host Rk8 and occupied the basal position of the other Japanese samples. As observed in the phylogenies of nuclear genomes, the samples from hosts Rk4, Rk5, Rk6 and Rk7 were closely related, but were unambiguously sub-grouped according to host origin. Samples from host Rk9 (Rk9-3, Rk9-6 and Rk9-11), which clustered separately in the nuclear tree, grouped together in the mitochondrial tree (Fig 4B). Interestingly, mitochondrial genome sequences of worms from the same host origin were not perfectly identical (especially in worms from host Rk4) although differences were very small and this may be due to sequencing errors.
Strongyloides stercoralis employs distinct modes of reproduction as follows: asexual parthenogenetic reproduction by parasitic females inside the host and sexual reproduction by free-living adults outside the host. Asexual reproduction may promote increased heterozygosity because of the absence of recombination and segregation in diploids (known as Mullers’s ratchet or Meselson effect) [23,24]. We therefore compared the heterozygosities (πt) of samples from Japan, where the parasites likely persist longer in the host through asexual auto-infection because no new infections are suggested to be unlikely to have occurred in Japan in the last 50 years [10] and our Japanese samples were collected from elderly people (Table 1), and samples from Myanmar to represent frequent new infections by larvae that arose through sexual reproduction.
As expected, most Japanese samples (i.e. all except Rk9-11, Rk9-3 and Rk8-7) comprised higher heterozygosities (πt = 0.0015–0.0017 in scaffolds > 8 kb) compared with Myanmar samples (0.0011–0.0013) (S2 Table), and this difference was significant (P < 1.8e -5, Welch t-test, df = 23). Intra-genome heterozygosity does not seem to be highly associated with read depth (S5 Fig), and the excess of heterozygosity in the Japanese samples were consistently observed in the genome (S6 Fig). These results suggest that excess of heterozygosity in the Japanese samples is likely to be true, excluding the possibility of false calls due to contaminations or other uncertain factors. The negative inbreeding coefficients (FIN) observed in such Japanese samples (−0.36 to −0.22) may represent repeated parthenogenetic reproduction of the nematodes in their hosts (S2 Table). Exceptions were Rk9-11, Rk9-3 and Rk8-7, which comprised fewer heterozygosities (0.0009 to0.0013) and higher FIN values (−0.09 to 0.28). The Japanese samples deviated significantly from Hardy-Weinberg equilibrium at 34.4% of loci, with 99.3% in heterozygous excess, compared with 0.8% of the loci in Myanmar samples, none of which were in heterozygous excess (S4 Table), suggesting more frequent asexual reproduction (insufficient sexual reproduction) has been used by Japanese worms than Myanmar ones. This point was discussed in [25] with observation of deviation from Hardy-Weinberg equilibrium in some populations of rat Strongyloides (S. ratti) and also reviewed in [26].
Next, we compared the heterozygosities of the scaffolds assigned to autosomes and sex chromosome [9] of individual samples (Fig 5). Two main groups were observed as follows: 1. Myanmar samples with values ranging from 0.001–0.0015 in the sex and autosomal scaffolds, 2. The majority of Japanese samples had with higher heterozygosities compared with those of Myanmar samples in the sex and autosomal scaffolds. The exceptions Rk9-11, Rk9-3 and Rk8-7 were positioned separately from those shown in the plot. The autosomal heterozygosities of Rk9-3 were lower but had values similar to those of the sex chromosomes of the other Japanese samples, whereas the heterozygosities of the sex chromosomes of Rk8-7 were low and had a value consistent with that of the autosomes of the Japanese samples. The values of both the sex chromosomes and autosomes of Rk9-11 were low. In contrast, the numbers of homozygous SNP sites in these three samples (S7 Fig) were greater than other Japanese samples on the sex chromosome of Rk8-7, the autosomes of Rk9-3 and both types of chromosomes of Rk9-11 (with an increase of approximately 50% of autosomes compared with Rk9-3).
Together, these results suggest that samples Rk8-7, Rk9-3 and Rk9-11 arose through recent sexual crossing between very closely-related individuals and acquired more homozygous chromosome pairs in sex chromosomes and autosomes. These findings likely explain the positions of Rk9-3 and Rk9-11 in the network tree, which were distant from Rk9-6 (Fig 4).
A major weakness of research on parasitic helminth genomes is the inability to obtain sufficient quantities of DNA because at present, none of these parasites can be cultured through its entire life cycle outside of a living host. Nevertheless, the WGA technique may solve this problem by producing high yields of whole genomic DNA from a single parasite [16]. Here we used the WGA technique combined with the NGS technology to re-sequence the entire genomes of individual S. stercoralis to acquire a better understanding the population structure of this medically important human pathogen. To the best of our knowledge, this study represents the first genome-wide approach to estimate the genotypic variations in S. stercoralis populations.
We show here that WGA detects variants with sensitivity comparable with those of normal variant detection methods, although WGA requires more data (coverage) to correctly call heterozygous positions, likely because of amplification bias. Here, our analysis of nematodes collected in Japan and Myanmar detected approximately 0.3 million variant positions, representing 0.6% of the genome, by comparison to the reference strain isolated from a dog in the United States. Although the reference and samples in this study were originally isolated from different hosts, this level of diversity represents as low as the diversity of C. elegans (~0.05%) [27] compared with other nematodes such as Pristionchus pacificus (~2%) [28] and Bursaphelenchus xylophilus (~4%) [29]. This may be explained by the relatively recent divergence of S. stercoralis from a common ancestor of S. stercolaris and the sister species, stronger selective pressure on the obligate parasite compared with free-living organisms such as P. pacificus, or facultative parasites such as B. xylophilus or both. Additionally, the unique mode of reproduction of this species may have affected the diversity level. S. stercoralis is distributed worldwide in areas with warm climates, and it will be interesting to analyse the diversity of S. stercoralis isolated in Africa, South America and Australia to study their global diversity. The data from such an analysis may illuminate the origin and migration routes of S. stercoralis and allow comparison of these attributes in populations of the parasite in humans and dogs as gene flow of parasites are generally determined by host movement [30].
Besides the human strongyloidiasis situations in the two countries (Japan and Myanmar), situations of Strongyloides infection in dogs are also likely to differ between the two countries. Strongyloides infection rate in dogs was reported to be as low as 0.4% in Okinawa, Japan [31]. Although we can’t find any reports about Myanmar canine strongyloidiasis, infection rate in Myanmar is possibly very high as reported in other Southeast Asian countries [32,33]. Therefore, a genome-wide investigation of their population structures would be of interest to see if a similar intra-genome heterozygosity trend can be observed as in human Strongyloides and to identify if there are interspecies transmissions between dogs and humans.
The phylogenetic relationships inferred from nuclear and mitochondrial SNPs were basically similar to each other. However, the relationships of Japanese samples observed in the nuclear trees were more complicated and therefore difficult to interpret. We found this is likely not only because the Japanese samples originated from a small gene pool but is also potentially explained by independent evolution of two haplotypes of the diploid genomes through asexual reproduction. This suggests that analyses of heterozygosity (e.g. by phasing) are useful and necessary to gain a better understanding of the structures of populations of S. stercoralis.
Because S. stercoralis has not been endemic in Japan for decades [10], the Japanese samples collected from elderly hosts aged 58–104 years (Table 1) may have been maintained only by auto-infection cycles for a long time. The higher heterozygosity of Japanese compared with Myanmar samples is thus possibly explained by an accumulation of heterozygous positions during the auto-infection cycle [10]. The exceptions Rk9-3, Rk9-11 and Rk8-7, which have reduced heterozygosity in sex or autosomal scaffolds or both are likely explained by recent cross events between two very closely related individuals, possibly during their isolation from a faecal culture. This, in turn, provides robust evidence that parthenogenesis of the parasitic female is mitotic (non-meiotic) and that free-living adults exchange chromosomes outside the host. Further, positive FIN (inbreeding efficiency) values of the Myanmar samples suggest that new infections occur in the prevalent regions by infective larvae produced through sexual reproduction between closely related individuals.
It has been suggested that new infections are unlikely to have occurred in Japan in the last 50 years [10]. Assuming that the genomic mutation rate of S. stercoralis is the same as that of C. elegans (9*10−9/site/generation) [34] and the minimum S. stercoralis generation time is 8 days, 50 years of asexually cycling within a human host can cause approximately 1,900 heterozygous sites to accumulate in the 86-M base diploid genome. Although this value is high, it is only ~20% of the number of differences observed between samples isolated in Japan and Myanmar (S2 Table). These values suggest that the frequency of sexual reproduction, which can reduce heterozygosity, is also an important factor for determining the number of heterozygous sites in the nematode genome. The analysis of heterozygosity can therefore serve to help draw inferences about the history of infections and the prevalence of parasites in a specific area.
|
10.1371/journal.pntd.0005372 | Decline in infection-related morbidities following drug-mediated reductions in the intensity of Schistosoma infection: A systematic review and meta-analysis | Since 1984, WHO has endorsed drug treatment to reduce Schistosoma infection and its consequent morbidity. Cross-sectional studies suggest pre-treatment correlation between infection intensity and risk for Schistosoma-related pathology. However, evidence also suggests that post-treatment reduction in intensity may not reverse morbidity because some morbidities occur at all levels of infection, and some reflect permanent tissue damage. The aim of this project was to systematically review evidence on drug-based control of schistosomiasis and to develop a quantitative estimate of the impact of post-treatment reductions in infection intensity on prevalence of infection-associated morbidity.
This review was registered at inception with PROSPERO (CRD42015026080). Studies that evaluated morbidity before and after treatment were identified by online searches and searches of private archives. Post-treatment odds ratios or standardized mean differences were calculated for each outcome, and these were correlated to treatment-related egg count reduction ratios (ERRs) by meta-regression. A greater ERR correlated with greater reduction in odds of most morbidities. Random effects meta-analysis was used to derive summary estimates: after treatment of S. mansoni and S. japonicum, left-sided hepatomegaly was reduced by 54%, right-sided hepatomegaly by 47%, splenomegaly by 37%, periportal fibrosis by 52%, diarrhea by 53%, and blood in stools by 75%. For S. haematobium, hematuria was reduced by 92%, proteinuria by 90%, bladder lesions by 86%, and upper urinary tract lesions by 72%. There were no consistent changes in portal dilation or hemoglobin levels. In sub-group analysis, age, infection status, region, parasite species, and interval to follow-up were associated with meaningful differences in outcome.
While there are challenges to implementing therapy for schistosomiasis, and praziquantel therapy is not fully curative, reductions in egg output are significantly correlated with decreased morbidity and can be used to project diminution in disease burden when contemplating more aggressive strategies to minimize infection intensity.
| Schistosomiasis is the disease caused by infection with Schistosoma parasitic flukes. Depending on the infecting species, chronic Schistosoma infection can cause a variety of pathologies including liver and spleen enlargement, fibrosis and hypertension of the portal vein of the liver, or bladder ulceration and deformities and kidney blockage. Infection can also cause anemia, diarrhea, abdominal pain, and decreased physical fitness. In our study, we quantified the reductions in prevalence of infection-related morbidities among populations with Schistosoma infection, as achieved by giving one or more drug treatments. We systematically reviewed 71 available reports of Schistosoma-related morbidity reduction and determined, based on a meta-analysis of the primary data, that the odds of persisting morbidity progressively decrease when greater post-treatment reductions in parasite burden are achieved, as reflected by reduced egg counts in standard diagnostic testing. This suggests that repeated or more effective anti-parasite drug treatment will be a valuable tool for greater reduction of Schistosoma-related patient morbidities in affected areas.
| Schistosomiasis, caused by Schistosoma spp. blood flukes, is one of the most prevalent parasitic diseases in the world, with more than 240 million people infected and 800 million at risk of infection [1]. Chronic schistosomiasis is the form of infection that is predominant in endemic areas, which bear the greatest disease impact from long-lived Schistosoma infections [2]. Because of pathology caused by parasite eggs deposited into human tissues, schistosomiasis turns into a multi-year inflammatory disease of the intestine, liver, urinary tract, and other critical organs. Adult schistosome worms colonize the human body for years, excreting eggs every day. These eggs provoke granulomatous inflammation in order to achieve translocation from the venous circulation to either the bowel or bladder lumena. If eggs do not succeed in leaving the body in excreta, they remain trapped in nearby tissues, causing persistent chronic inflammation and scarring [3, 4].
For many years, clinical studies of the morbidity related to schistosomiasis have mainly focused on specific forms of advanced organ pathology and focal clinical signs. These include hepatosplenomegaly, periportal fibrosis, portal hypertension, bladder deformity, hydronephrosis, hematuria, abdominal pain and related organ scarring [5–7]. More recent research has also put emphasis on systemic morbidities associated with Schistosoma infection such as anemia, growth stunting, impaired cognition, undernutrition, diarrhea, and decreased physical fitness; however, this additional burden of schistosomiasis was not well studied in many older works, and until the 1990s, improvement in these outcomes was not generally appreciated as a potential benefit of morbidity control [8].
Schistosomiasis control is a constant challenge for endemic regions and their public health services, mainly due to difficulties in preventing early infection and frequent reinfection. Several strategies, such as environmental control of the intermediary host, provision of safe water, and medical treatment have been used, singly and in combination [9]. However, since the 1980s, especially with the advent of praziquantel, drug-based control of morbidity related to infection has been the primary WHO strategy for schistosomiasis control, with treatment given mainly through community- and school-based mass treatment [10]. The usual parameters employed to assess the effectiveness of treatment have been its effects on the intensity and prevalence of infection. Although there is an association between intensity of infection and the presence and severity of morbidity [11–14], the correlation is imperfect, and monitoring infection intensity may provide only an indirect means to gauge morbidity risk. Individuals with low intensity infections can express all forms of the disease, and thus we must consider that the morbidity caused by Schistosoma infection can also be triggered by just the presence of infection [8, 14–18].
In recent years, millions of people have been treated in different contexts and, in general, prevalence of morbidity has been reduced after treatment [7, 19–22]. Nevertheless, studies of morbidity reduction related to drug treatment have had some conflicting results [23–26], which may be a reflection of differences in follow-up after treatment, methods used to measure morbidities, the Schistosoma species, the presence of co-infections (especially malaria), the type of population and the region, the initial prevalence of infection, the incidence of reinfection, and other factors [7, 27]. Despite the potential benefits of treatment, many affected persons have not yet been reached by treatment programs [28].
Given this context, and that one of the main objectives of schistosomiasis control programs has been to achieve reductions in morbidity associated with Schistosoma infection [29], there is a need to accurately quantify the reduction of morbidity levels as a result of chemotherapy intervention, so that the specific benefits of more intensive interventions can be identified. To do this, we developed a meta-analysis to evaluate the impact of drug treatment and the reduction of infection intensity on levels of morbidity associated with schistosomiasis. In specific, because a quantitative link can be used in cost-effectiveness analysis comparing different treatments strategies, we aimed to determine the numerical relationship between egg reduction rates (ERR, observed in post-treatment diagnostic testing [30]) and the reduced risk of morbidity after treatment.
The data used in this project were aggregated, anonymized data from previously published studies; as such, this study does not constitute human subjects research according to U.S. Department of Health and Human Services guidelines (https://www.hhs.gov/ohrp/regulations-and-policy/guidance).
This research was developed by the authors and performed according to a protocol in which all the stages of the study were pre-defined. The protocol was recorded and published in the International Prospective Register of Systemic Reviews (PROSPERO) online database, number CRD42015026080, available at http://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42015026080. This study is reported in accordance with PRISMA guidelines (see attached checklist document, S1 File).
Studies that evaluated morbidities related to infection with Schistosoma species, before and after specific chemotherapy for schistosomiasis, were included in this review. In our quantitative meta-analysis, which focused on morbidity prevalence before and after chemotherapy, only morbidities reported by more than one study (from which the necessary data could be extracted) were included. No restrictions were placed in terms of location of the study, Schistosoma species, or publication date. Publications in English, Portuguese, Spanish, and French were included. We excluded animal studies, case studies, reviews, and studies with individuals selected only from clinics or hospitals. Regarding study design, any prospective, longitudinal studies of treatment impact on morbidity (with or without concurrent control group) were considered eligible for inclusion in the meta-analysis. Studies had to describe the study site, the species of Schistosoma parasite, the type of schistosomiasis morbidity evaluated before and after chemotherapy, the diagnostic method used to assess the morbidity, and the characteristics of participating study subject population. In addition, the numbers of subjects evaluated at baseline and at each follow-up were required, along with reporting of morbidity prevalence or mean laboratory values before and after treatment intervention.
The publications analyzed in this review were identified by searching public electronic databases including PubMed, and the Virtual Health Library VHL/BIREME (http://pesquisa.bvsalud.org/portal), which allows access to multiple databases (LILACS, MEDLINE and Cochrane Library), and Google Scholar (https://scholar.google.com/). The searches were conducted in August 2015. In addition, the bibliography reference lists of articles selected for review were evaluated for additional relevant studies, and additional articles were retrieved from personal collections at Case Western Reserve University.
Published studies were identified in the electronic databases using the PICO strategy (Patient, Intervention, Comparator, and Outcome) to develop the descriptors. The descriptors used to identify patients were ‘Schistosomiasis’ and ‘Schistosoma’; for interventions, ‘drug therapy’, ‘treatment outcome’, and ‘therapeutics’; for outcomes, ‘morbidity’, ‘anemia/anaemia’, ‘pain’, ‘diarrhea’, ‘attention’, ‘memory’, ‘underachievement’, ‘growth’, ‘nutritional status’, ‘physical fitness’, ‘hydronephrosis’ ‘hematuria/haematuria’, ‘knowledge’, ‘work capacity evaluation’, ‘body weight’, ‘hepatomegaly’, ‘splenomegaly’, ‘hypertension, portal’, ‘proteinuria’, ‘disability evaluation’, and ‘fibrosis’. These descriptors were taken from the terminology of classification systems for indexing each database, MeSH (Medical Subject Headings) and DeCS (Health Sciences Descriptors). In the VHL/BIREME database the descriptors were considered in three languages, English, Portuguese, and Spanish.
References obtained from each search were exported to reference manager software Mendeley (version 1.14). The selection of studies was carried out in two stages by two independent reviewers (GA and DJB), and in case of disagreement between them, a third reviewer (CHK) was asked to resolve differences. The first stage of selection analyzed the titles and abstracts of the publications. The selected studies had full texts recovered for the second stage of selection. For this step we designed a Microsoft Access database form, using the eligibility criteria as described, in order to assist in the archiving of eligible studies for the systematic review. Duplicate publications and papers reporting reanalysis of previously published data were excluded at this stage.
The data abstracted from selected publications were curated in the Microsoft Access study database. The reviewers extracted the following information from each text: full citation, year of publication, country and region where the study was conducted, characteristics of subjects (age, sex, and selection criteria), Schistosoma species studied, type of treatment offered and dose, follow-up time in months, number of individuals evaluated, type of morbidity evaluated in the study, method used to measure morbidity, prevalence or mean for each morbidity measure, egg counts or reduction rate of eggs, and prevalence of infection in the population. Wherever possible, all of the information listed above was recorded both before and after intervention (see supplemental information in Tables A-K in S1 Text).
Studies that evaluated more than one form of morbidity were included in the meta-analysis for each individual morbidity outcome. Moreover, when a paper reported more than one study with the same morbidity (performed among different subjects), this publication was listed twice. Thus, the number of identified studies was higher than the number of publications, but each study was included in the quantitative analysis no more than once per morbidity.
The quantitative analysis of the data included: first, an analysis of the impact of treatment per se on the odds of having morbidity after therapy; and second, a separate analysis by meta-regression of the specific impact of egg count reductions on the odds of post-treatment disease. The following morbidities were associated in common with infection with either S. mansoni, S. haematobium, or S. japonicum: splenomegaly, hepatomegaly, and mean hemoglobin. For intestinal schistosomiasis caused by S. mansoni or S. japonicum, we included periportal fibrosis, diarrhea, blood in the stool, and alteration in the main portal vein. For urogenital infection with S. haematobium, we included hematuria, proteinuria, abnormalities in the urinary bladder, and lesions of the upper urinary tract. Additional outcomes that could not be evaluated quantitatively due to differences in methods and classification included anthropometric measures, oxygen consumption, tolerance to physical activity, and abdominal pain. [N.B. The meta-analysis of outcomes of cognitive performance and school achievement will be published in a separate paper.]
For hematuria, we only included studies that evaluated the microhematuria detected by reagent strips, whereas studies of hematuria detected by visual inspection were excluded. For morbidity studies that presented ordinal rankings of severity, such as ‘periportal fibrosis grades 1–3’, we classified morbidity as ‘present’ for individuals with any degree of severity. Thus, the decrease in prevalence after chemotherapy in this meta-analysis represent the complete reversal of morbidity. Partial reversal of morbidity, such as a shift from grade 3 to grade 2 as reported in some studies, was not considered.
For studies that evaluated the morbidities more than once after the treatment, the first follow-up after the intervention was selected for inclusion in the pooled analysis and calculation of summary estimates. Other follow-up periods were analyzed later in subgroup analysis. Regardless of the number of segments in the study, the change in morbidity was always assessed against pre-treatment baseline values.
Quantitative pooled analysis of treatment effects catalogued from the eligible studies was performed using Comprehensive Meta-Analysis software, v.3.3 (CMA, Biostat, Englewood, NJ) which provided calculation of summary estimates of the impact of treatment, along with their confidence intervals. For morbidity reported as dichotomous outcomes, a pooled odds ratio was calculated with 95% confidence interval (CI95%) using Der Simonian and Laird random effects modeling. For continuous data, the measure of effect was the calculated standardized mean difference (SMD) and its CI95%. The Z-test was used to assess statistical significance at a P < 0.05 level. For each morbidity, summary data were presented visually by Forest plots showing the respective odds ratio or SMD and CI95% for the pooled analysis. (Additional data from subgroup analyses are shown in tables of effect size in this paper’s supplemental file, see Tables A-K in S3 Text).
For meta-regression of the impact of reduction in infection intensity after treatment, the egg reduction rate (ERR), for eggs detected on standard stool or urine diagnostic testing was calculated by the formula:
ERR=meaneggcountatbaseline−meaneggcountatfollowupmeaneggcountatbaseline*100
The ERR was initially calculated using either the geometric mean (ERRGM) or arithmetic mean (ERRAM) egg counts, depending on the data provided by the study. Twenty-six studies reported geometric mean data outcomes, while 12 reported arithmetic means. For consistency in our meta-regression of ERR vs. logarithmically transformed odds ratios, we converted the ERRAM values to estimated ERRGM values using correlations developed by Olliaro, et al., [30] in their systematic review of treatment effects on individual egg count values. The objective of this meta-regression was to assess the impact of treatment on morbidity according to the intensity reduction across the range of included studies. The percent reductions in log OR that we have projected for a 90% ERR are derived from the correlation coefficients and their CIs. Conversion of the estimated log10(OR) at 90% ERR to its corresponding OR by exponentiation yielded a fraction projected as the remaining odds of morbidity at that ERR value.
The perceived quality of individual studies was assessed, but not formally quantified in our analysis because of specific limiting features found in many NTD trials [31]. A summary of study design and quality factors for each included study is presented in supplemental information file S2 Text: ‘Study design and quality features for included studies’. We assessed study quality using the National Heart, Lung, and Blood Institute quality assessment tools for pre-post design studies (https://www.nhlbi.nih.gov/health-pro/guidelines) with one additional criterion about reporting of treatment coverage. Most studies worked with endemic populations living in small clusters and so did not select an entire population or a randomly-selected subsample to follow. In addition, many older studies did not detail their criteria for selection of the population. For our analysis, it was assumed that the included studies represented the best available information for the population and morbidity of interest at the time it was undertaken.
Assessment for potential publication bias was carried out by visual inspection of funnel plots, and statistically by calculating the Egger test [32].
Heterogeneity among studies in each meta-analysis was assessed using the Cochrane Q test (χ2 test) with significance assumed for P < 0.1, and Higgin's and Thompson's I2 statistic [33]. To explore heterogeneity and factors that could potentially modify the summary estimates of effect, we performed subgroup analyses stratified by parasite species, the study area, age of the subjects included in the studies, the time to follow-up after treatment, the type of diagnosis, the treatment performed, the number of treatments, and the initial prevalence of infection in the study population [34]. Not all morbidities had such stratifying data for all studies. For the sensitivity analysis, each meta-analysis was retested with the exclusion of one study at a time to assess the possibility of a disproportionate impact of any individual study on summary estimates.
Using the selected search terms, initial screening of the databases yielded 1852 study reports after removing duplicates. After titles and abstracts were assessed, 309 reports were selected for full review. Publications eliminated in the first stage were excluded because they were animal studies, review studies, case reports, immunological studies, studies of parasitological efficacy and safety only, diagnostic studies, reinfection studies, spatial distribution studies, or evaluations of mass treatment programs, surgical intervention, other diseases, prevalence of coinfection, or studies to estimate prevalence and intensity of infection. As outlined in Fig 1, 194 of these study reports were excluded after second stage screening, leaving a total of 115 reports for inclusion in the systematic review. However, 51/115 papers did not have sufficient quantitative data on morbidity or on subject characteristics, or had different data formats, such that a final total of 64 papers (see S1 Text) were ultimately included in the quantitative data synthesis (meta-analysis) presented in this report.
Seventy-one eligible studies were abstracted from sixty-four papers. Publication dates ranged from 1977 to 2013 (median year = 1996). Studies were conducted in twenty-one countries. Of the seventy-one studies, 51% were from East Africa (Kenya, Tanzania, Madagascar, Ethiopia, Burundi, Uganda), 18% were from West Africa (Ghana, Niger, Mali, Senegal, Burkina Faso, Côte d’Ivoire), 7% were from Southern Africa (Zambia and Zimbabwe), 8.5% were from South America or the Caribbean (Brazil, Venezuela, and St. Lucia), 5.5% were from China, 5.5% were from Sudan, and there was one study each from Central Africa (Congo), from Indonesia, and the Philippines. The greatest number of subjects evaluated for morbidity outcomes had S. haematobium infections (52%), followed by S. mansoni (38%), S. japonicum (8.5%) and mixed infections (1.5%). Most of the subjects were school-age individuals (45%), but some studies included subjects of all ages (34%), whereas 21% were in studies that selected their subjects according to sex, age, clinical status, or presence of comorbidities such as hookworm. Most studies (82%) used praziquantel as the specific treatment for schistosomiasis, and 70% used a PZQ dose of 40 milligrams per kilogram. 8.5% used metrifonate, 6% used oxamniquine, while 4% used hycanthone. Moreover, 14% of studies used some of combination therapy with mebendazole or albendazole for treatment of intestinal helminths. Overall, the studies enrolled a total of 24,214 subjects at baseline and 22,207 individuals were monitored for morbidity outcomes (considering the first follow-up of each study). We found no evidence of publication bias using unweighted, non-randomized values in the Egger test.
To examine the hypothesis that post-treatment intensity of Schistosoma infection remains a correlate of morbidity risk after therapy, we performed meta-regression of the odds of having infection-related morbidities post-treatment as a function of post-treatment ERR achieved in an individual study population. The ERR was measured as reductions in population mean intensity of infection from before to after treatment. The impact on morbidity was measured as the corresponding change in morbidity prevalence, comparing the study population’s odds of disease before and after treatment.
The meta-regression analysis suggested that there is a significantly greater reduction in the prevalence of some morbidities if greater egg reduction effects can be achieved. For hepatomegaly, a practical target of 90% egg reduction was projected to yield an estimated 84% (CI95% 52%, 95%) reduction in the odds of left lobe enlargement. The corresponding projected reduction in the odds of right lobe enlargement was 81% (CI95% 2%, 96%) while for unspecified lobar enlargement, the projected reduction in odds was 99% (CI95% 98%, 99.4%) (Fig 4). With respect to periportal fibrosis, a ninety-point ERR was predicted to reduce odds of this form of disease by 87% (CI95% 64%, 95%) (Fig 5). Greater ERR impact was projected for all morbidities related to urogenital schistosomiasis: a ninety-point egg reduction was predicted to yield reductions in the odds of hematuria by 99.8% (CI95% 99.7%, 99.9%), in the odds of proteinuria by 99.2% (CI95% 97%, 99.8%), and in the odds of ultrasound abnormalities in the urinary tract by 99.2% (CI95% 96%, 99,9%) (Fig 6).
Nevertheless, it was observed that even near-total reduction in egg counts by drug treatment was unlikely to lead to complete reduction of all morbidity risk. Not shown, the meta-regression showed only non-significant correlation of ERRs with reductions in the study cohort prevalence of splenomegaly or with post-treatment increases in hemoglobin levels.
Quantification of the net changes in Schistosoma infection-associated morbidity prevalence, from before to after treatment, is one way to critically value the impact of drug-based control of schistosomiasis, which is the strategy currently recommended by WHO and other agencies [10]. Our systematic review and meta-analysis sought to summarize many decades of research on disease-related benefits of treatment for schistosomiasis. To do this, we catalogued treatment impact on eleven key morbidities linked to infection by any of the three major Schistosoma parasites of humans, S. haematobium, S. mansoni, and S. japonicum. Overall, our results suggest that drug treatment significantly reduces but does not eliminate these common pathologic consequences of Schistosoma infection, and that the odds of improvement are linked to the magnitude of treatment-related reductions in adult worm burden of parasitic infection.
Chances of post-treatment morbidity reductions were higher for morbidities related to urogenital schistosomiasis than for morbidities caused by intestinal schistosomiasis. For the included urogenital morbidities associated with infection by S. haematobium, the greatest reduction after treatment was for odds of having hematuria; the lowest reduction was for the odds of having upper urinary tract lesions detected by ultrasound, primarily characterized by hydronephrosis. The presence of blood in the urine is a well-accepted marker of S. haematobium infection and its presence is used as a mapping and screening tool for urogenital schistosomiasis in Africa [40, 41]. Although hematuria, proteinuria, and bladder abnormalities appear to respond quickly to anti-schistosomal therapy, in step with the ERRs achieved, the relatively smaller improvements in prevalence of hydronephrosis suggest that this form of morbidity is a more slowly resolving and sometimes irreversible form of urinary tract schistosomiasis. This phenomenon thus limits the overall impact of drug treatment in communities at high risk for S. haematobium infection [42–44]. Post-treatment reductions in the odds of morbidities related to intestinal infections (S. mansoni and S. japonicum) ranged from 37–74%. The least impact was for splenomegaly, whereas the largest observed decrease was for blood in the stool. Regarding measured impact on splenomegaly, it was not uncommon that the studies selected for meta-analysis involved subjects who were co-infected with other chronic pathogens, especially malaria, which could explain a lesser effect of anti-schistosomal therapy on splenomegaly after treatment [22, 36]. Other reviews of Schistosoma-related ultrasound morbidities have suggested that the regression of splenomegaly, while sometimes observed after anti-schistosomal therapy, is not specific enough to be used as an indicator for the regression of Schistosoma-associated disease. In their analysis, malaria was the main co-factor contributing to this effect [45]. In addition, like hydronephrosis in urogenital schistosomiasis, splenic enlargement in intestinal schistosomiasis is likely a marker of more severe and more prolonged chronic intestinal schistosomiasis, and it may be more difficult to achieve regression with late treatment [46, 47].
In the pooled analysis of treatment effects, the 74% reduction in odds of blood in the stool was the best result among morbidities associated with intestinal infection. Large -scale questionnaire surveys of blood in the stool, trialed as rapid assessment tools for identifying high-risk communities in sub-Saharan Africa, have shown that this symptom can be a valuable indicator for the diagnosis of S. mansoni in endemic areas, having low to moderate sensitivity and medium to high specificity [41]. For S. japonicum in China, a separate study has estimated that the highest risk indicator of infection-associated morbidity is a history of bloody stools [5]. Although regression in the odds of bloody stools was a quick indicator of anti-schistosomal treatment effect, the reductions were lower than for S. haematobium-associated hematuria, suggesting that this manifestation is less likely to be specific for intestinal schistosomiasis in the context of many other circulating enteropathogens.
In our subgroup analyses, some study features were clearly linked to either better or more limited reductions in morbidity prevalence after treatment. In many cases, more significant treatment effects were observed when studies were performed on school age children or on subpopulations selected for existing pathology at baseline. This was seen for the outcomes of hepatomegaly, diarrhea, periportal fibrosis, and abnormalities of the urinary bladder. In endemic regions, it is believed that age is an important proxy of cumulative exposure to the parasite and the related tissue damage that it causes. As the process of infection progresses from acute injury to a more chronic forms of fibrotic scarring, it becomes proportionately more difficult to reverse Schistosoma-associated pathology [44, 48]. In addition, the meta-analysis also suggests differences between S. mansoni and S. japonicum infections in their likelihood of morbidity reduction in response to therapy. Post-treatment odds of splenomegaly and periportal fibrosis were not significantly reduced for infection with S. japonicum, although studies of S. mansoni treatment effects were able to demonstrate significant impact for these two morbidity markers. These findings were consonant with two earlier reviews that have highlighted the persistence of abnormalities caused by S. japonicum [7, 45].
The chances of observing reductions in hepatomegaly, diarrhea, proteinuria, and bladder abnormalities were higher when the studies were performed on subjects who were definitely infected, i.e., all egg-positive. However, this was not the case for splenomegaly reduction or for periportal fibrosis. Eggs are most consistently detected in stool or urine with heavier infections, and persons with light intensity infection may have morbidity but have egg-negative status on the day of survey testing. Those studies that included these egg-negative infections may have shown a greater impact on morbidities because of there being a proportionately greater impact of treatment on resident worm burden (with possible complete parasitological cure) in light infection.
Also in our analysis, follow-up interval was an important factor in gauging the impact of therapy. For those morbidities related to intestinal schistosomiasis, i.e. hepatomegaly, splenomegaly, and periportal fibrosis, a longer follow-up period, especially > 24 months, was associated with greater reductions after treatment. The exception was left hepatic lobe enlargement, which had the best reductions in the first year after treatment, but decreased benefit over longer time periods. Of note, the reductions in morbidities associated with urogenital schistosomiasis, with the exception of injuries to the upper urinary tract, were more likely to be significant if evaluated in the first six months after treatment.
In our summary estimates, only two morbidities showed no consistent or significant change between pre- and post-treatment surveys. These were the prevalence of portal vein dilation and change in mean hemoglobin level. Only studies delivering two chemotherapeutic interventions and those having a follow-up time greater than 24 months were associated with significant reductions in the diameter of the portal vein. In clinical studies, portal vein diameter is an indicator that correlates with portal vein pressure and risk for hemorrhage [47]. This finding likely reflects a more advanced stage of disease with a smaller chance of a beneficial chemotherapy effect from a single dose. With respect to hemoglobin levels, it was only possible to identify statistically significant changes when the follow-up was performed at an interval greater than twelve months after treatment. Prior analysis has indicated that the benefit in terms of gains in hemoglobin levels is greatest among those who have anemia at baseline, or those who have greater levels of microhematuria or infection intensity [15, 20]. Studies of S. japonicum have found that the peak elevation of post-treatment hemoglobin levels occurs at 15 months [16]. Of importance to public health, it appears that monitoring of schistosomiasis-associated anemia impact should be planned for a period at one year or more after treatment.
The relative intensity of infection is an important correlate of morbidity, because the formation of the disease is related to the daily deposition of parasite eggs into host tissues [17, 48–50]. While immediate granulomatous inflammation is the cause of some of the morbidities included in our review (hematuria, proteinuria, bladder irregularities for S. haematobium; bloody stool, diarrhea, and hepatic enlargement for S. mansoni and S. japonicum; and anemia of inflammation for all three species), cumulative damage over decades of infection is linked to advanced fibrotic complications of infection such as hydronephrosis, portal fibrosis, and portal dilation. Our meta-regression profiles indicate that acute reductions in worm burden, as reflected by the ERRs achieved after drug therapy, are associated with reversal of most of the acute pathologies of infection. However, the more advanced chronic forms of disease were less responsive to single rounds of treatment, even with adequate ERRs, and our stratified analysis suggests that multiple rounds of treatment are necessary to improve (or hopefully prevent) these outcomes.
As study limitations, there is moderate risk of bias in this study’s estimates. The data analyzed in this study may have been influenced by confounders such as uneven sex distributions, the presence of co-infections, and variation in local reinfection rates that could not be controlled for in the meta-analysis. Moreover, the evidence may be limited in terms of generalizability because of the limitations in the design of included studies, and because the diverse populations selected for analysis yielded a high degree of heterogeneity across studies. To help minimize these effects, we have used random effects modeling in the meta-analysis and have performed sensitivity analysis to look for possible skewing of estimates by results from single influential studies [34].
Our meta-analysis identified that significant gaps exist in the available literature on post-treatment reduction of morbidities. In our study’s quality assessment, the study factors that most frequently could not be evaluated were: subject inclusion/exclusion criteria, the power analysis of the selected study sample size, and the use of blinding for assessment of study outcomes. Loss to follow-up was > 20% from baseline in many studies, and the potential biasing effect of this phenomenon was often not considered. Meta-analysis and meta-regression are observational research that depends on the quality of the studies that are included. As previously noted by others [31], the level of evidence for many NTD clinical studies has to be categorized as only “very low, low, or moderate quality”. That said, is has been the chronic underfunding of clinical trials (performed in resource limited settings) that has been an important part of the ‘neglect’ of NTDs. In order to strengthen the evidence base for Schistosoma morbidity control, there is a clear need to perform additional cohort trials that are both well-designed and well-reported.
The main findings of this meta-analysis are: i) post-treatment reduction in morbidity varies according to Schistosoma species; ii) for most pathologies, the odds of persisting morbidity progressively decrease with greater reductions in post-treatment egg counts (ERR); iii) however, not all morbidities respond in parallel with egg reduction. The population studied, their ages and infection status, and the interval for follow-up all influenced the magnitude of morbidity reductions noted in a given study cohort. Our findings illuminate and help to quantify the magnitude of improvements after treatment of Schistosoma-associated morbidities. These new estimates may prove useful in cost-effectiveness estimations for program planning, and can provide direction for future operational research on treatment implementation strategies.
|
10.1371/journal.pgen.1002536 | The miR-35-41 Family of MicroRNAs Regulates RNAi Sensitivity in Caenorhabditis elegans | RNA interference (RNAi) utilizes small interfering RNAs (siRNAs) to direct silencing of specific genes through transcriptional and post-transcriptional mechanisms. The siRNA guides can originate from exogenous (exo–RNAi) or natural endogenous (endo–RNAi) sources of double-stranded RNA (dsRNA). In Caenorhabditis elegans, inactivation of genes that function in the endo–RNAi pathway can result in enhanced silencing of genes targeted by siRNAs from exogenous sources, indicating cross-regulation between the pathways. Here we show that members of another small RNA pathway, the mir-35-41 cluster of microRNAs (miRNAs) can regulate RNAi. In worms lacking miR-35-41, there is reduced expression of lin-35/Rb, the C. elegans homolog of the tumor suppressor Retinoblastoma gene, previously shown to regulate RNAi responsiveness. Genome-wide microarray analyses show that targets of endo–siRNAs are up-regulated in mir-35-41 mutants, a phenotype also displayed by lin-35/Rb mutants. Furthermore, overexpression of lin-35/Rb specifically rescues the RNAi hypersensitivity of mir-35-41 mutants. Although the mir-35-41 miRNAs appear to be exclusively expressed in germline and embryos, their effect on RNAi sensitivity is transmitted to multiple tissues and stages of development. Additionally, we demonstrate that maternal contribution of miR-35-41 or lin-35/Rb is sufficient to reduce RNAi effectiveness in progeny worms. Our results reveal that miRNAs can broadly regulate other small RNA pathways and, thus, have far reaching effects on gene expression beyond directly targeting specific mRNAs.
| RNA interference (RNAi) has become a widely used approach for silencing genes of interest. This tool is possible because endogenous RNA silencing pathways exist broadly across organisms, including humans, worms, and plants. The general RNAi pathway utilizes small ∼21-nucleotide RNAs to target specific protein-coding genes through base-pairing interactions. Since RNAs from exogenous sources require some of the same factors as endogenous small RNAs to silence gene expression, there can be competition between the pathways. Thus, perturbations in the endogenous RNAi pathway can result in enhanced silencing efficiency by exogenous small RNAs. MicroRNAs (miRNAs) comprise another endogenous small RNA pathway, but their biogenesis and mechanism of gene silencing are distinct in many ways from RNAi pathways. Here we show that a family of miRNAs regulates the effectiveness of RNAi in Caenorhabditis elegans. Loss of mir-35-41 results in enhanced RNAi by exogenous RNAs and reduced silencing of endogenous RNAi targets. The embryonic miR-35-41 miRNAs regulate the sensitivity to RNAi through lin-35/Rb, a homolog of the human Retinoblastoma tumor suppressor gene previously shown to regulate RNAi effectiveness in C. elegans. Additionally, we show that this sensitivity can be passed on to the next generation of worms, demonstrating a far-reaching effect of the miR-35-41 miRNAs on gene regulation by other small RNA pathways.
| The ability of double stranded RNA (dsRNA) to induce silencing of specific genes was discovered in the nematode Caenorhabditis elegans and dubbed RNA interference (RNAi) [1]. RNAi was subsequently identified in many organisms, including mammals, providing a powerful experimental tool for inactivating specific genes [2], [3]. In worms, RNAi can be initiated by long dsRNAs from exogenous sources, such as injected or ingested bacterially produced dsRNA [1], [4]. Cellular factors process the dsRNAs into small interfering RNAs (exo–siRNAs) of ∼21 nucleotides long, which target complementary mRNAs for degradation [5]. Exo–siRNAs form complexes with Argonaute proteins that mediate target mRNA cleavage or degradation through other mechanisms [6]. Recently, a nuclear RNAi pathway was discovered in worms where the Argonaute NRDE-3 silences target genes through a co-transcriptional mechanism of silencing [7], [8].
Comparable to exo–RNAi, endogenous dsRNAs can enter processing pathways that produce endo–siRNAs that silence target genes through base-pairing interactions [6], [9]. Endo–siRNAs have been identified by small RNA cloning in C. elegans, Drosophila and mouse and are predicted to target thousands of endogenous genes, particularly mRNAs present in germline and embryos [10], [11], [12], [13], [14], [15]. In C. elegans, both the exo- and endo–siRNA pathways require the endoribonuclease enzyme Dicer (DCR-1) to process the initiating dsRNAs into ∼21 nt siRNAs, but distinct Argonaute proteins usually bind the different types of siRNAs to mediate gene silencing [16], [17]. For example the Argonaute RDE-1 is required for RNAi initiated by exogenous but not endogenous dsRNAs [17].
In nematodes and plants, RNAi-mediated silencing can be amplified through the production of secondary siRNAs by RNA dependent RNA polymerases (RdRPs). In C. elegans, loss of the RdRP RRF-3 results in greatly reduced levels of endo–siRNAs and an enhanced response to exogenous dsRNA 18,19. Likewise, the exonuclease ERI-1 is required for the accumulation of some endo–siRNAs and worms with mutations in eri-1 display hypersensitivity to exo–siRNAs [18], [20]. Another class of mutants with enhanced RNAi is represented by lin-35/Rb, which encodes the worm homolog of the Retinoblastoma tumor suppressor, and includes other genes in the lin-35/Rb pathway such as lin-15, dpl-1 (mammalian DP), and hpl-2 (mammalian HP1) [21], [22], [23]. The molecular mechanism by which lin-35/Rb negatively regulates the RNAi pathway remains to be fully understood. Worms with mutations in lin-35/Rb have up-regulated levels of mRNAs corresponding to cloned endo–siRNAs, suggesting decreased levels or function of endo–siRNAs in these mutants [24]. Reduced endo–RNAi activity may free limiting factors for the exo–RNAi pathway. Furthermore, some of the up-regulated genes in lin-35/Rb mutants encode Argonaute proteins that might also contribute to enhanced exo–RNAi [24]. Additionally, lin-35/Rb mutants have increased expression of germline-specific genes in somatic tissues [23]. This expression pattern is predicted to enhance RNAi sensitivity by providing factors normally utilized in the germline for silencing pathways [23].
MicroRNAs (miRNAs) represent another class of small RNAs that derive from endogenously expressed transcripts [25]. Typically, miRNAs are transcribed as long primary transcripts containing hairpin structures that are released by Drosha cleavage. The resulting precursor undergoes processing by Dicer to produce the mature ∼22 nt miRNA, which forms a complex with Argonaute proteins. While mammalian miRNAs seem to distribute evenly among the four Argonaute proteins, worm and fly miRNAs mostly associate with specific Argonautes [6]. For example, Argonaute Like Genes 1 and 2 (ALG-1/2) bind worm miRNAs, but not exo- or endo–siRNAs [26]. In contrast to exo and endo–siRNAs, animal miRNAs pair with imperfect sequence complementarity to their target mRNAs, causing target degradation and translational repression [27]. The common factor for most small RNA pathways is Dicer. Loss of Dicer activity in worm or mammalian cells results in defective RNAi and severely reduced levels of most miRNAs and endo–siRNAs [16], [18], [28], [29], [30]. However, mutations in other genes required for endo- or exo–RNAi seem to have little effect on the miRNA pathway in C. elegans [16], [18].
In an effort to understand the function of specific miRNAs, a large collection of deletion mutants was generated in C. elegans [31]. While most individual mutants displayed no obvious phenotypes, simultaneous deletion of multiple family members (miRNAs with identical sequences at positions 2–7) often resulted in developmental defects and lethality [32], [33]. For example, the miR-35-41 miRNAs are clustered within ∼800 nt of a common transcript and share a high degree of sequence homology [34]. Deletion of this miRNA cluster results in defective germ and intestinal cell proliferation and temperature sensitive embryonic lethality [32], [35]. While the genes regulated by miR-35-41 that are responsible for these phenotypes are yet to be determined, several direct targets of these miRNAs were recently validated in embryonic extracts [36]. Although the miR-35-41 miRNAs have only been found in worms and planaria [34], [37], poorly conserved miRNA clusters have also been observed to be highly expressed and function in early development in other species, including Drosophila, mouse and humans [38], [39], [40].
Here we show that the miR-35-41 miRNAs regulate other small RNA pathways in C. elegans. Deletion of the miRNA cluster results in worms with enhanced RNAi sensitivity in multiple developmental stages and tissues. This effect is likely related to decreased endo–RNAi activity in mir-35-41 mutants, as they also exhibit up-regulation of many endo–siRNA targets. We found that the miR-35-41 miRNAs negatively regulate the exo–RNAi pathway through lin-35/Rb and that this function can be supplied maternally to the progeny. These results point to a new level of cross-regulation between small RNA pathways, whereby the expression of specific miRNAs can impact the efficiency of RNAi mediated by exo– and endo–siRNAs.
In C. elegans the mir-35-41 miRNAs are required for embryonic viability and proliferation of certain cell types, but the targets relevant for these phenotypes have not yet been established [31], [32], [35]. In the mir-35-41(gk262) and mir-35-41(nDf50) strains all seven members of the miRNA cluster are deleted, resulting in temperature sensitive embryonic lethality (Figure S1) [31], [32]. While using RNAi to identify factors that genetically interact with mir-35-41(gk262), we discovered that this strain exhibits enhanced sensitivity to RNAi. In agreement with previous studies [1], [4], wild type worms fed bacteria expressing double stranded RNA targeting the muscle gene unc-22 resulted in a high penetrance of worms with the twitching phenotype, while only 2% resembled the null unc-22 loss-of-function phenotype of paralysis (Figure 1A and Table 1) [41]. In contrast, the majority (83%) of mir-35-41(gk262) worms were paralyzed by unc-22(RNAi) (Figure 1A and Table 1). This same treatment produced paralysis in about 20% of the established RNAi hypersensitive strain, rrf-3(pk1426) (Table 1), consistent with previous studies [19], [20]. We confirmed that the enhanced RNAi phenotype of mir-35-41(gk262) was due to loss of the miRNA gene instead of the overlapping anti-sense Y62F5A.9 protein-coding gene. A transgene encoding only the mir-35-41 locus rescued the hypersensitivity of this strain to unc-22(RNAi) (Figure 1A) and supported miRNA expression in mir-35-41(gk262) worms (Figure 1B).
Loss of mir-35-41 results in enhanced RNAi in multiple tissues and stages of development. We observed that RNAi knockdown of lin-1 (hypodermal), sqt-1 (hypodermal), unc-86 (neuronal), pos-1 (germline) and sex-1 (embryonic) resulted in enhanced phenotypes in mir-35-41(gk262) relative to wild type worms (Table 1). Furthermore, the effects were similar or stronger in mir-35-41(gk262) compared to the RNAi hypersensitive rrf-3 mutants. These results indicate that expression of the miR-35-41 miRNAs negatively impacts the efficiency of exo–RNAi in C. elegans.
To determine if the enhanced RNAi phenotype of mir-35-41(gk262) was dependent on core exo–RNAi factors, we crossed this strain to mutants in the RNAi pathway. Upon unc-22(RNAi) treatment, the single mutants defective in RNAi (rde-1, rde-4 and rrf-1) showed no phenotypic response, as expected (Table 2) [42], [43], [44]. Likewise, addition of mutations in rde-1, rde-4 or rrf-1 rendered mir-35-41(gk262) mutants completely RNAi defective (Table 2). These data show that mir-35-41(gk262) mutants require rde-1, rde-4 and rrf-1 to exhibit an RNAi response. While this might be expected, there is precedence for an RNAi hypersensitive strain (lin-15B(n744)) being independent of rrf-1 activity [23].
The response of mir-35-41(gk262) to unc-22(RNAi) is comparable to that of the exceptionally RNAi hypersensitive strain lin-35(n745), which carries a putative null mutation [21], [22], [23], [45]. About 20–30% of the population of the enhanced RNAi strains rrf-3(pk1426), eri-1(mg366) and ergo-1(tm1860) exhibited paralysis in response to unc-22(RNAi) (Table 3). In comparison, over 80% of the mir-35-41(gk262) and lin-35(n745) strains were paralyzed by this RNAi treatment (Table 3). The rrf-3, eri-1 and ergo-1 genes are part of the endogenous RNAi pathway in C. elegans and loss of these genes results in decreased levels of endo–siRNAs [16], [17], [18], [46]. Diminished activity of the endo–RNAi pathway is one explanation for enhanced RNAi sensitivity to dsRNA from exogenous sources. The Argonaute protein NRDE-3 serves as a sensor for endo–siRNAs, as nuclear localization of this protein is dependent on the presence of siRNAs [8]. In contrast to wild type worms that express endo–siRNAs, GFP::NRDE-3 was mostly cytoplasmic in the seam cells of eri-1(mg366) worms, which are defective in endo–siRNA accumulation (Figure 2A) [8], [16], [18]. As in wild type, localization of GFP::NRDE-3 was nuclear in the mir-35-41(gk262) and lin-35(n745) mutant strains (Figure 2A). Additionally, the hypersensitivity of mir-35-41 mutants continued in the absence of nrde-3 activity (Figure 2B). These results show that in mir-35-41 mutants sufficient endo–siRNAs are present to target NRDE-3 to the nucleus and the enhanced RNAi phenotype of these mutants does not require NRDE-3 activity.
To gain insight into the role of the mir-35-41 miRNAs in regulating the exo–RNAi pathway, we performed microarray analysis of gene expression in WT versus mir-35-41(gk262) embryos. We chose statistically significant changes in gene expression of at least 1.5-fold differences between the triplicate sample averages (p value≤0.005 (two-sample t-test)) (Table S1). Of the 550 up-regulated genes in mir-35-41(gk262), only a small fraction (4%) were predicted miR-35 targets based on a list of 606 genes from four different algorithms (Targetscan, PicTar, RNA22 and MiRWIP). Furthermore, obvious genes that might explain the heightened RNAi sensitivity of mir-35-41(gk262) were not mis-regulated at the mRNA level. Comparing the microarray results to a list of 6,469 endo–siRNA targets obtained from four independent studies revealed significant enrichment of endo–siRNA targets in the up-regulated genes (37%, p value 1.3e-05 two tailed exact Fisher's significance test) (Table S1) [11], [13], [18], [47]. In contrast, endo–siRNA targets were under-represented in the list of genes down-regulated in mir-35-41(gk262) (13% p value 2.2e-16 exact Fisher's significance test) (Table S1). Differential expression of the established endo–siRNA target E01G4.5 was confirmed by RT-qPCR (Figure 3). Similar to the previously reported effect of eri-1 on this gene, both the spliced and unspliced forms of E01G4.5 were up-regulated in mir-35-41 mutants (Figure 3) [8].
The mir-35-41 and lin-35/Rb mutant strains both show strongly enhanced RNAi and up-regulation of endo–siRNA targets. Also both strains resemble WT for NRDE-3 localization, which is dependent on siRNAs for nuclear residence (Table 3, Table S1, Figure 2) [21], . These similarities led us to investigate if the mir-35-41 and lin-35/Rb genes might regulate each other. Although lin-35/Rb mRNA levels were not significantly different, protein levels of this gene were reduced to about 20% in mir-35-41(gk262) relative to wild type worm embryos (Figure 4A and 4B), suggesting that mir-35-41 positively regulates the accumulation of LIN-35 protein. In contrast, mir-35 miRNA levels were unaltered in lin-35(n745) mutant embryos (Figure 4C). The decreased levels of LIN-35 protein could be responsible for the RNAi hypersensitivity of mir-35-41 mutants. To test this possibility, we crossed an extrachromosomal array expressing lin-35 into mir-35-41(gk262). Worms carrying the array, Ex[lin-35(+); sur-5::GFP], are distinguished by GFP expression [48]. In the mir-35-41(gk262) worms with extra copies of lin-35 provided by the array (GFP+), the hypersensitivity to unc-22(RNAi) was rescued with the majority of worms showing the twitching instead of paralysis phenotype (Figure 4D). The non-GFP (GFP−) siblings were comparable to the original mir-35-41(gk262) strain for sensitivity to unc-22(RNAi) (Figure 4D). Other GFP transgenes that lack lin-35 did not affect the RNAi hypersensitivity of mir-35 mutants, consistent with the idea that lin-35 is required for the rescue (Figure 4D). The extra copies of lin-35 also reduced the RNAi hypersensitivity of lin-35(n745) genetic mutants (Figure 4D). Rescue of the RNAi phenotype by extra copies of lin-35 was stronger in mir-35-41(gk262) than in lin-35(n745), possibly due to the reduction versus complete absence of LIN-35 protein in these mutants, respectively. To test if extra copies of lin-35 specifically rescues the RNAi hypersensitivity of mir-35-41(gk262), we introduced the transgene into other enhanced RNAi mutants. The response to unc-22(RNAi) was unaffected by the addition of the lin-35(+) transgene to eri-1 or rrf-3 mutants, supporting the conclusion that the RNAi hypersensitivity of the mir-35-41 mutants can be attributed to reduced lin-35/Rb. A further prediction of this model is that a mir-35-41;lin-35 double mutant would exhibit the same RNAi response as the lin-35 single mutant. However, the double mutant proved to be inviable and could not be tested for RNAi sensitivity.
The lin-35/Rb gene is a member of the synthetic multivulva B (synMuv B) family, which includes transcriptional repressor and chromatin modifying genes [45]. Worms with combined mutations in synMuv B and synMuv A genes produce multiple vulva structures because of cell lineage defects [49]. In addition to lin-35, some of the other synMuv B genes, such as lin-53, lin-9 and dpl-1, also negatively regulate the exo–RNAi pathway and cause enhanced RNAi when mutated [22], [23]. We tested if the decreased levels of lin-35/Rb in mir-35-41(gk262) are sufficient to produce the multiple vulva (Muv) phenotype when combined with the synMuv A mutants lin-15a(n767) or lin-8(n111). Double mutants consisting of mir-35-41(gk262) and either of the synMuv A genes did not display the Muv phenotype (n = 230 adult worms/strain). In comparison, 100% of the lin-35(n745); lin-15a(n767) and the lin-35(n745); lin-8(n111) double mutants were Muv (n = 240 adult worms/strain). Taken together, these results suggest that RNAi efficiency is more sensitive than the vulva formation pathway to reduction in lin-35/Rb levels or that tissues dependent on lin-35/Rb for vulva formation produce sufficient protein in the absence of the miR-35-41 miRNAs.
Previous work has shown that the miR-35-41 cluster of miRNAs is predominantly expressed in embryos with comparatively little expression in larval and adult somatic tissues [34]. In fact, miRNAs from this cluster make up about 75% of all mature miRNAs present in early embryos [50]. Since the miR-35-41 miRNAs are present in one-cell stage embryos before zygotic transcription by RNA Polymerase II has initiated [50], [51], the miRNAs or their precursors are likely supplied by maternal germ cells. Thus, we tested if maternal contribution of mir-35-41 activity would be sufficient to regulate RNAi sensitivity. Since our results suggest that the RNAi hypersensitivity of mir-35-41 mutants is through lin-35/Rb, we also analyzed maternal rescue of lin-35(n745). Crossing of mir-35-41(gk262) or lin-35(n745) hermaphrodites to wild type males rescued the unc-22(RNAi) paralyzed phenotype to ∼20% in the F1 progeny compared to ∼80% in the parental strains (Figure 5). These heterozygous F1's were allowed to self-fertilize and the resulting F2 progeny were scored for the paralysis phenotype and then genotyped. Regardless of the zygotic genotype, the worms that had come from mothers with one wild type allele for mir-35-41 or lin-35 were less sensitive to unc-22(RNAi) than the original mutant strains (Figure 5). Thus, maternal contribution of mir-35-41 or lin-35/Rb is sufficient to regulate RNAi sensitivity.
We have shown that the mir-35-41 miRNA gene inhibits the exogenous RNAi pathway by positively regulating the expression of LIN-35/Rb protein. The effect of the miR-35-41 miRNAs on LIN-35/Rb levels is likely indirect as miRNAs typically repress gene expression and obvious binding sites for the miRNAs are not present in the lin-35 sequence. Regulation of LIN-35/Rb appears to be through a post-transcriptional mechanism since mRNA levels were unaffected in mir-35-41 mutants. We, and others, have been unable to identify direct targets of mir-35-41 that could explain the mis-regulation of LIN-35 levels, RNAi hypersensitivity, or embryonic lethal phenotypes of mir-35-41 mutant strains [32]. Positive regulation of the endo–siRNA pathway by mir-35-41 is likely indirect through targets of this miRNA family. Although few predicted targets of miR-35-41 were up-regulated upon loss of these miRNAs, this result is consistent with a recent study showing that embryonic targets of miR-35-41 undergo deadenylation while the rest of the mRNA remains stable in many cases [36]. Thus, microarrays lack the sensitivity to detect genes mis-regulated in mir-35-41 mutant embryos. The seven miRNAs in the miR-35-41 cluster share a common seed sequence and may regulate many genes that affect the viability and RNAi sensitivity phenotypes of mir-35-41 mutants, which could prevent individual targets from being discovered through genetic approaches.
The reason for enhanced exo–RNAi in lin-35 mutant strains is yet to be fully understood. Consistent with its role as a transcriptional repressor, loss of lin-35/Rb activity results in mis-expression of germline specific genes in somatic cells [23], [24], [52]. However, this effect alone cannot explain the enhanced RNAi, as lin-35 mutants exhibit hypersensitivity in both germline and somatic cells [22], [52]. Several genes for Argonaute proteins that function in the RNAi pathway are also up-regulated in lin-35/Rb mutants [24]. Additionally, targets of endo–siRNAs are overexpressed in the absence of lin-35/Rb, indicating that the endo–RNAi pathway is defective in lin-35/Rb mutants [24]. Thus, the increased expression of RNAi factors and reduced competition with the endo–RNAi pathway may underlie the improved efficiency of RNAi in lin-35 mutants. We found that sufficient endo–siRNAs are produced to target NRDE-3 to the nucleus in lin-35/Rb and mir-35-41 mutants, suggesting that these genes regulate the function, but not accumulation, of endo–siRNAs. Furthermore, both the spliced and unspliced forms of the endo–siRNA target E01G4.1 were up-regulated in mir-35-41 mutants. Since RNAi can silence targets at the transcriptional level in worms, mir-35-41 may also regulate RNAi effectiveness through this pathway [7], [8], [53].
RNAi initiated from exogenous dsRNA can propagate across generations in worms and other organisms [54], [55], [56], [57]. While the Argonaute RDE-1 and dsRNA binding protein RDE-4 are required to activate the RNAi response, they are dispensable for maintenance of RNAi [55], [57]. Instead, chromatin remodeling factors seem to mediate the inheritance of RNAi-induced phenotypes [53], [57]. Thus, the original RNA signal that directs post-transcriptional silencing of complementary targets may also stimulate chromatin remodeling events that repress gene expression across generations. Our results indicate that mir-35-41, through regulation of lin-35/Rb, decreases exo–RNAi potency and that this effect can be maternally contributed. The RNAi hypersensitivity of lin-35/Rb mutants can also be maternally rescued when the gene is expressed from the chromosomal locus but not from a transgene (Figure 4D and Figure 5). The inability of Ex[lin-35(+); sur-5::GFP] transgenes to provide maternal rescue has previously been observed and is likely due to the generally poor expression of transgenes in the germline [48], [58].
The mir-35-41 miRNAs are expressed in oocytes and present in early embryos prior to the onset of zygotic transcription [32], [50]. This expression pattern is consistent with our demonstration that maternal contribution is sufficient to regulate exo–RNAi sensitivity. However, this activity can also be provided by the wave of zygotic expression of mir-35-41 during embryogenesis [50], since mating of mir-35-41(gk262) to wild type worms rescues the RNAi hypersensitivity phenotype. Although the sequences of the mir-35-41 miRNAs are not conserved across species, many animals express high levels of poorly conserved miRNAs from clusters in embryonic stem cells and during early embryogenesis [38], [59], [60], [61]. One role for these miRNAs is to target maternal mRNAs for degradation [36], [38], [59], [62]. Our results show that miRNAs can also affect the activity of other small RNA pathways, demonstrating a broad function in regulating gene expression through both direct and indirect silencing mechanisms.
C. elegans worm strains were maintained on NGM plates seeded with OP50 bacteria, under standard conditions [63]. Worms were synchronized by hypochlorite treatment of gravid hermaphrodites followed by overnight hatching of embryos at 20°C. Strains used in this study include the following: wild type (WT) Bristol N2 strain, NL2098 rrf-1(pk1417)I, MT10430 lin-35(n745)I, VC514 mir-35-41(gk262)II, NL2099 rrf-3(pk1426)II, MT111 lin-8(n111)II, WM49 rde-4(ne301)III, GR1373 eri-1(mg366)IV, WM158 ergo-1(tm1860)V, WM27 rde-1(ne219)V, MT1806 lin-15A(n767)X, YY158 nrde-3(gg66)X. Double mutants: PQ300 [mir-35-41(gk262);rde-1(ne219)], PQ301 [mir-35-41(gk262);rde-4(ne301)], PQ302 [mir-35-41(gk262); eri-1(mg366)], PQ303 [mir-35-41(gk262);rrf-1(pk1417)], PQ304 [mir-35-41(gk262); ergo-1(tm1860)], PQ421 [mir-35(gk262);lin-8(n111)], PQ422 [mir-35-41(gk262);lin-15A(n767)], PQ423 [lin-35(n745);lin-8(n111)], PQ424 [lin-35(n745); lin-15A(n767)], PQ459 [mir-35-41(gk262);nrde-3(gg66)]. Non-integrated transgenic strains: PQ20 [mir-35-41(gk262); apEX160 [mir-35-41]], MH1461 [lin-35(n745); kuEx119 [lin-35(+); sur-5::GFP]], PQ416 [mir-35-41(gk262); kuEx119 [lin-35(+), sur-5::GFP]], PQ439 [mir-35-41(gk262); Ex [myo-2::GFP)], PQ456 [eri-1(mg366); kuEx119 [lin-35(+), sur-5::GFP]], PQ458 [rrf-3 (pk1426); kuEx119 [lin-35(+); sur-5::GFP]]. Integrated transgenic strains: YY174 [ggIS1 [nrde-3p::3xflag::gfp::nrde-3]], YY178 eri-1(mg366); [ggIS1 [nrde-3p::3xflag::gfp::nrde-3]], PQ427 [mir-35-41 (gk262); yy1774 [ggIS1 [nrde-3p::3xflag::gfp::nrde-3]], PQ428 [mir-35-41(gk262);nrIs20 [sur-5::nls-gfp]], PD4251 ccIs4251I [myo-3::Ngfp-lacZ].
Worm viability was assayed by synchronizing worms and plating L1 hatchlings at 20°C. Worms were transferred to individual plates at the L4 stage and allowed to lay embryos for 24 h at 20°C or 25°C. Parents were then removed from the plates and embryos were counted. Viable worms were counted 40 hours later. Percent viable progeny represents the number of viable worms/total number of embryos laid.
RNAi plates were prepared using 25 ug/mL carbenicillin and 6 mM IPTG (Isopropyl β-D-1-thiogalactopyranoside, Apex). RNAi clones were obtained from the Ahringer feeding RNAi library [64]. For most experiments synchronized L1 worms were cultured on OP50 until the L4 stage (40 hours) and then transferred to RNAi food. After 28 hours adults were scored for unc-22, lin-1 and sqt-1 phenotypes. For pos-1 and sex-1 RNAi, L1 worms were directly plated on RNAi food and allowed to grow to adults. Parents were then transferred to individual RNAi plates and allowed to lay about 50 embryos. Percent embryonic lethality was calculated as the number of embryos that did not hatch by 40 hours later. For paternal/zygotic rescue experiments, wild-type male worms containing a body muscle GFP marker (PD4251) were crossed to mir-35(gk262) (mir-35−/−) mutant hermaphrodites. Parents were removed from the mating plate after 24 hours. Young GFP positive worms represented heterozygous mir-35−/+ F1 cross progeny. F1s were allowed to grow to the L4 stage and 50 GFP positive worms were transferred to unc-22(RNAi). F1s were scored for phenotypes 28 hours later. For maternal rescue, GFP positive F1's were transferred to individual plates and allowed to lay self-fertilized F2 embryos. Fifty F2 worms at the L4 stage were transferred to unc-22(RNAi). Each worm was scored for phenotypes 28 hours later, lysed and genotyped.
Trizol reagent (Invitrogen) was used to extract total RNA from frozen embryo pellets of the wild type (N2) and mir-35(gk262) strains. RNA samples were reverse-transcribed and hybridized to arrays following the Affymetrix manufacturer's protocol. Samples from three independent replicates of each strain were hybridized to the Affymetrix GeneChip C. elegans Genome Arrays representing 22,500 transcripts. Microarray samples were processed by the UCSD GeneChip Microarray Core. The raw data was normalized and t-statics were computed using R and Bioconductor (www.bioconductor.org) with the “affy” package and Benjamini-Hochberg (BH) correction method for multiple comparisons [65]. RNA levels that changed at least 1.5-fold with a probability of p<0.005 after BH correction were considered significantly different in mir-35(gk262) mutants relative to wild-type. Quantitative real-time PCR of reverse transcribed RNA (RT-qPCR) was performed with DNase treated RNA and the primers listed in Table S2. Northern blots to detect mRNAs and miRNAs and Western blots to detect proteins were performed as previously described [66]. The antibody for LIN-35 was provided by the Horvitz lab [45].
|
10.1371/journal.ppat.1005487 | The Myeloid LSECtin Is a DAP12-Coupled Receptor That Is Crucial for Inflammatory Response Induced by Ebola Virus Glycoprotein | Fatal Ebola virus infection is characterized by a systemic inflammatory response similar to septic shock. Ebola glycoprotein (GP) is involved in this process through activating dendritic cells (DCs) and macrophages. However, the mechanism is unclear. Here, we showed that LSECtin (also known as CLEC4G) plays an important role in GP-mediated inflammatory responses in human DCs. Anti-LSECtin mAb engagement induced TNF-α and IL-6 production in DCs, whereas silencing of LSECtin abrogated this effect. Intriguingly, as a pathogen-derived ligand, Ebola GP could trigger TNF-α and IL-6 release by DCs through LSECtin. Mechanistic investigations revealed that LSECtin initiated signaling via association with a 12-kDa DNAX-activating protein (DAP12) and induced Syk activation. Mutation of key tyrosines in the DAP12 immunoreceptor tyrosine-based activation motif abrogated LSECtin-mediated signaling. Furthermore, Syk inhibitors significantly reduced the GP-triggered cytokine production in DCs. Therefore, our results demonstrate that LSECtin is required for the GP-induced inflammatory response, providing new insights into the EBOV-mediated inflammatory response.
| Ebola virus (EBOV), a highly virulent pathogen, causes a severe hemorrhagic fever syndrome. The fatal infection is characterized by a systemic inflammatory response similar to septic shock. Ebola glycoprotein (GP) is thought to contribute to disease pathogenesis, as high amounts of shed GP from virus-infected cells are detected in patients, and activate macrophages and dendritic cells (DCs) to produce proinflammatory cytokines. Here, we show that LSECtin plays an important role in GP-mediated inflammatory responses in human DCs. LSECtin is a DAP12-coupled receptor able to initiate specific signaling events in human DCs. LSECtin interacts with Ebola GP and results in DAP12 phosphorylation. LSECtin knockdown impairs the production of proinflammatory cytokines induced by Ebola GP. Thus, this study suggests that LSECtin may contribute to Ebola GP-mediated pathogenicity.
| Ebola virus (EBOV), a member of the family Filoviridae, is the causative agent of severe hemorrhagic fever in humans, which is responsible for the outbreak in West African countries in 2014 [1]. Following EBOV infection, dendritic cells (DCs) and macrophages are the early and preferred replication sites of this virus, after which other cell types, including endothelial cells, epithelial cells and hepatocytes, are rapidly infected [2,3]. In experimental animal models, excessive production of proinflammatory cytokines and chemokines occurs during lethal EBOV infection, leading to endothelial cell permeability, multiorgan failure, and severe clotting disorders and culminating in a final septic shock-like syndrome [4–6]. More importantly, a fatal outcome in infected patients is also associated with aberrant innate immunity characterized by a “cytokine storm”, with hypersecretion of numerous proinflammatory mediators [3,7], suggesting that the inflammatory response plays an important role in EBOV pathogenesis. Consequently, identification of the molecular mechanisms of the inflammatory response is very important to our understanding of EBOV diseases.
EBOV genome consists of seven genes that encode seven structural proteins. Glycoprotein (GP) gene is the fourth of seven genes and encodes type I transmembrane GP termed pre-GP via transcriptional RNA editing [8,9]. The pre-GP is cleaved by furin into two subunits, GP1 and GP2, which remain linked by a disulfide bond [10,11]. This heterodimer (GP1,2) is known to form a trimer on the viral surface. The cleavage of surface GP by tumor necrosis factor-α-converting enzyme (TACE) releases a trimeric GP, termed shed GP [12]. During EBOV infection, significant amounts of shed GP can be detected [12]. Recently, it has been demonstrated that shed GP can induce the production of proinflammatory cytokines by activating non-infected DCs and macrophages, which can explain the dysregulated inflammatory host reactions to Ebola infection [13]. In addition, Ebola virus-like particles (eVLPs) consisting of virus protein (VP40) and GP are able to induce the activation of DCs [14]. GP is required for eVLPs to induce DCs cytokine production [15,16]. All of these results support that GP can induce inflammatory response. However, the molecular mechanism underlying GP-mediated inflammatory responses is unclear.
Inflammatory responses are rapidly elicited in response to infection by pathogens [17,18]. Innate immune cells including macrophages and DCs play important roles in this process. DCs and macrophages express diverse pattern recognition receptors (PRRs) that recognize conserved pathogen-associated molecular patterns (PAMPs) to elicit inflammatory immune responses via upregulation of proinflammatory cytokines such as tumor necrosis factor (TNF) and IL-6. C-type lectin receptors (CLRs) have been identified as PRRs and play important roles in initiating an innate immune response [19]. The functions of these receptors in immunity as PRRs for carbohydrates present on fungi and some bacterial have been well defined, such as via dectin-1 and DC-SIGN, which signal through Syk [20] and Raf-1 [21], respectively. However, the role of CLRs in inflammation mediated by virus components is less documented. The lectin LSECtin is encoded in the same chromosomal locus as DC-SIGN and also expressed by human peripheral blood DCs as well as DCs and macrophages generated in vitro [22,23]. It has been reported that LSECtin binds exogenous Ebola GP [24,25] and mediates its internalization as a PRR [23]. However, it is unclear whether LSECtin initiates specific signaling events and is involved in GP-mediated inflammatory responses.
In this study, we report that LSECtin is a DAP12-coupled activating receptor that recognizes Ebola GP. We show that triggering of endogenous LSECtin in DCs by either its mAb or GP activates Syk and ERK and leads to CARD9- and Syk-dependent cytokine production. Collectively, these findings suggest that LSECtin functions as a DAP12-coupled receptor and acts as a functional PRR for Ebola GP.
LSECtin is a C-type lectin receptor and binds Ebola GP as a pattern recognition receptor. To verify whether LSECtin interacts GP, recombinant protein GP1-Fc was prepared and subjected to Coomassie blue staining and Western blotting (S1A and S1B Fig). In addition, we found that under nonreducing conditions, recombinant protein GP1-Fc is in monomeric form (S1C Fig). Using an enzyme-linked immunosorbent assay, our data demonstrated that GP1-Fc binds LSECtin in a dose-dependent way (S1D Fig). LSECtin has a typical carbohydrate recognition domain (CRD) and binds Ebola GP in a Ca2+-dependent manner [25]. Amino acid sequence alignment of the CRD of LSECtin with those of other C-type lectins indicates that 2 amino acids, Asn256 and Asn274, interact with Ca2+ through their carbonyl groups. Thus, we mutated the residues to aspartic acid. The mutant LSECtin (N256D or N274D) did not bind Ebola GP, which suggests that these residues are critical for recognition of EBOV GP (S1D Fig). Furthermore, we also performed a cell surface staining assay and demonstrated that Jurkat cells lentivirally transfected with LSECtin rather than mutant LSECtin bind GP (S1E Fig).
Next, we explore the expression profile of LSECtin in human blood leukocytes. As shown in S2A Fig, the anti-LSECtin mAb CCB059 did not stain granulocytes, monocytes or lymphocytes. We further investigated the expression of LSECtin on monocyte-derived DCs (MDDCs) by culturing monocytes in the presence of GM-CSF and IL-4 (S2B Fig). This treatment resulted in a strong up-regulation of LSECtin. In addition, this result was confirmed by PCR and Western blotting (S2C and S2D Fig). Ebola GP interacts with LSECtin stably expressed on Jurkat cell line. It was thus of interest to investigate whether GP could also bind LSECtin on human MDDCs. To address the issue, MDDCs were transfected with siRNA specific for LSECtin or with control siRNA for 48h (S3 Fig) and stained with GP1-Fc. First, we found that Ebola GP can bind MDDCs. More importantly, the activity is partially dependent on LSECtin. These results suggest that LSECtin involves GP binding to MDDCs (Fig 1A).
To investigate whether GP/LSECtin interaction can lead to the production of proinflammatory cytokines within the human immune system, MDDCs transfected with siRNA specific for LSECtin or with control siRNA were stimulated with eVLPs and eVP40 which were produced in insect cells. Compared with the stimulation of VP40, eVLPs significantly enhanced the production of cytokines and chemokines, suggesting that GP is required for eVLPs to activate DCs. Furthermore, we found that after LSECtin “knockdown”, MDDCs stimulated with eVLPs produced less TNF-α, IL-6, IL-8, IL-10 and MIP-1α (Fig 1B). Although eVLPs produced in insect cells or mammalian 293T cells exhibit similar DC-stimulating activities [26], eVLPs were also produced in 293T mammalian cells to determine whether the data are influenced by the insect cell expression system for GP. Similar with the results shown in Fig 1B, eVLPs produced in 293T mammalian cells induced less production of TNF-α, IL-6, IL-8, IL-10 and MIP-1α in LSECtin “knockdown” MDDCs (Fig 1C). These results suggest that LSECtin involves the Ebola GP-induced cytokine production whether it was produced in insect cells or human 293T cells.
Soluble GP1-Fc did not induce cytokine production (S4 Fig), which is consistent with the previous report [15]. To simulate the configuration and multivalency of GP on eVLPs or shed GP, GP1-Fc was coated on a culture well for stimulation of MDDCs. To directly investigate the role of LSECtin in GP1-mediated proinflammatory cytokine production, control or LSECtin siRNA-transfected MDDCs were stimulated with plate-bound GP1-Fc. Similar to the results as described above, GP1-Fc-induced cytokine production was impaired in LSECtin “knockdown” DCs (Fig 1D). In addition, our results demonstrated that the production of cytokines induced by eVLPs and plate-bound GP1-Fc was inhibited by Pepinh-MYD, a MyD88 inhibitor peptide, which is consistent with the previous reports that eVLP and shed GP induced cytokine production through TLR4/MyD88 signaling (S5A and S5B Fig) [13,27]. However, the cytokine production induced by LPS is not impacted in LSECtin “knockdown” DCs, suggesting that MyD88 signaling pathway is intact (Fig 1D). Given that the production of cytokines is partially inhibited by LSECtin silencing or MyD88 inhibitory peptide, we next determined whether there is a synergistic efficacy of LSECtin silencing in combination with MyD88 inhibitory peptide. We found that MDDCs treated by double silencing produced less TNF-α and IL-6 than LSECtin silencing or MyD88 inhibitory peptide alone treated cells (Fig 1E). In addition, we also demonstrated that there is a synergistic efficacy of LSECtin and TLR4-induced cytokines after LSECtin/TLR4 double “knockdown” DCs (S6 Fig). These results suggesting that LSECtin and TLR4/MyD88 signaling collaborate to mediate inflammatory response induced Ebola GP.
GlcNAcβ1-2Man disaccharide has been demonstrated to be a specific inhibitor of interaction between LSECtin and Ebola GP [25]. Our result also demonstrated that GlcNAcβ1-2Man inhibits the GP binding to MDDCs (Fig 1A). More importantly, the production of TNF-α and IL-6 can be inhibited by the addition of GlcNAcβ1-2Man and the effect was specific for Ebola GP as LPS-induced TNF-α and IL-6 production was unaffected by the presence of the GlcNAcβ1-2Man (Fig 1F). Collectively, these results suggest that LSECtin is selectively expressed in MDDCs and involved in GP-mediated proinflammatory cytokine production.
The above results suggest that Ebola GP induced cytokine production by MDDCs through both LSECtin and MyD88 signaling. To specially and clearly explore the LSECtin signaling, we use anti-LSECtin mAbs, including CCA023, CFD051 and CCB059, to stimulate LSECtin signaling upon crosslinking in MDDCs. We treated MDDCs with immobilized anti-LSECtin mAbs. The production of TNF-α and IL-6 was significantly increased in MDDCs after 24h of treatment with CFD051, compared with CCA023, CCB059 and the control mIgG1, suggesting that only CFD051-LSECtin engagement promoted cytokine production (Fig 2A). We also observed similar changes in mRNA levels. We found that LSECtin engagement induced rapid but transient mRNA expression for the cytokines IL-6 and TNF-α (Fig 2B). The IL-6 and TNF-α mRNA amounts peaked approximately 3h after LSECtin ligation and subsequently declined to close to baseline 6h after LSECtin crosslinking. In addition, LSECtin engagement increased the maturation of MDDCs, as characterized by increased surface expression of HLA-DR, CD83 and CD86 (Fig 2C). Interestingly, GP/LSECtin interaction also triggers the maturation of DCs as the surface expression of CD40, CD80 and CD86 decreased in LSECtin “knockdown” DCs (S7 Fig).
To determine whether TLR signaling is involved in LSECtin-mediated cytokine production, MDDCs were pretreated with Pepinh-MYD (a MyD88 inhibitor) and stimulated the cells with plate-bound CFD051. Our results show that MyD88, a crucial adaptor of TLR signaling, was dispensable for LSECtin-mediated cytokine production (S8 Fig). To prove that the cellular effects mediated by LSECtin engagement were specific, we treated the siRNA-transfected MDDCs with CFD051 overnight. We found that after LSECtin “knockdown”, MDDCs stimulated with CFD051 produced less TNF-α and IL-6 (Fig 2D). To exclude the possibility that the TNF-α and IL-6 production was simply due to Fc receptor engagement, we prepared F(ab′)2 fragments of anti-LSECtin mAb and used them to stimulate MDDCs. As shown in Fig 2E, plate-bound F(ab′)2 fragments from anti-LSECtin mAb induced the production of TNF-α and IL-6 and markedly increased their production in the presence of LPS (a TLR4 agonist), suggesting cooperation between TLR4 and LSECtin signaling in the MDDCs.
The NF-κB factors are held in the cytoplasm in an inactive state complexed with the inhibitory IκBα proteins. Upon stimulation by LPS, IκBα is phosphorylated and subsequently degraded resulting in NF-κB activation. S9 Fig shows that LPS induced a significant reduction in IκBα that last up to 1h. The basal IκBα levels were restored by 2h. However, LPS and LSECtin mAb combined treatment induces IκBα degradation that last up to 2h. The results indicated that LSECtin and TLR4 signaling crosstalks at the level of NF-κB activation. Taken together, our results demonstrated that LSECtin engagement can specially promote TNF-α and IL-6 production and enhance the maturation of MDDCs.
The above results showed that LSECtin mediates positive signaling in MDDCs. However, there is no signal transduction motif in this protein’s cytoplasmic tail. Therefore, it is likely that LSECtin is associated with an adaptor molecule to transduce signals. Co-immunoprecipitation and immunoblot analysis showed that LSECtin selectively associated with DAP12 but not with FceRIγ (Fig 3A). We also used reverse IP to show that LSECtin co-precipitated DAP12 (Fig 3B). Importantly, the interaction between endogenous LSECtin and DAP12 was also obvious in MDDCs (Fig 3C). Thus, these results suggested that LSECtin is associated with DAP12.
LSECtin does not possess any positively charged residues in the transmembrane domain that is required for the interaction with DAP12 in many other receptors. The negatively charged amino acid D50 in the transmembrane region of DAP12 is dispensable for the interaction (S10 Fig). However, the interaction of LSECtin and DAP12 was mediated through the transmembrane region of LSECtin (Fig 3D), as deficiency of the transmembrane region abolished its association with DAP12. Next, we further showed that a short stretch of transmembrane region proximal to the intracellular domain of LSECtin (amino acids 32–43) was required for association with DAP12 (Fig 3E). In addition, the interaction of LSECtin and DAP12 was independent of the only two hydrophilic threonines (T41 and T42) within the transmembrane region of LSECtin (S10 Fig).
The previous results in Fig 2 show that immobilized antibody to LSECtin can induce the production of TNF-α and IL-6. To determine whether LSECtin-mediated signaling is dependent on DAP12, MDDCs were transfected with siRNA specific for DAP12 or with control siRNA for 72 h (Fig 3F). We found that LSECtin failed to induce the production of TNF-α and IL-6 in DAP12 “knockdown” MDDCs after treatment with CFD051 antibody (Fig 3G), suggesting that LSECtin transduces signaling in a DAP12-dependent manner.
LSECtin binds Ebola GP and is required for eVLP-induced cytokine production. To determine whether eVLPs were able to induce tyrosine kinase-based intracellular signals through LSECtin, Jurkat cells were transfected with LSECtin and DAP12. Jurkat cells do not express LSECtin and DAP12 and refractory to Ebola GP-mediated infection [28]. The LSECtin- and DAP12-transfected cells were either left unstimulated or stimulated with eVLP. Whole-cell extracts were subjected to Western blotting using an anti-phosphotyrosine Ab (4G10) to detect tyrosine-phosphorylated proteins. Compared with correspondingly stimulated LSECtin or DAP12 transfectants, eVLP-treated LSECtin-DAP12 cells yielded increased amounts of tyrosine-phosphorylated proteins (Fig 4A). LSECtin bearing the two amino acid mutants (N256D andN274D) does not bind Ebola GP1-Fc. Consistent with this observation, LSECtin mutants (N256D and N274D) did not deliver an activation signal in response to eVLP stimulation, suggesting that LSECtin recognizes its ligand dependently of Ca2+-binding sites (Fig 4B).
Signaling through DAP12 is mediated by its ITAM, which relies on phosphorylation of the two tyrosines within the ITAM for propagation of a signal [29]. To determine whether the two tyrosines of DAP12 are required for LSECtin/DAP12-mediated phosphorylation of protein tyrosines, we transduced a lentivirus encoding wild-type (WT) or mutant DAP12 in which the ITAM tyrosines at positions 91 and 102 were mutated to phenylalanine (2YF) into Jurkat-LSECtin stable cells. Our results show that eVLP stimulation enhanced the phosphorylation of protein tyrosines in Jurkat-LSECtin cells expressing WT but not mutant DAP12 (Fig 4C), suggesting that the two tyrosines within the ITAM are required for LSECtin/DAP12-mediated phosphorylation of protein tyrosines. We next determine whether ligation of LSECtin results in tyrosine phosphorylation of DAP12. Jurkat-LSECtin/DAP12 and Jurkat-LSECtin/DAP12(Y2F) stable cells were stimulated with eVLP. To detect DAP12 phosphorylation, DAP12 protein was immunoprecipitated and tyrosine phosphorylation of DAP12 was examined with 4G10 antibody. Our results show that eVLP induced DAP12 phosphorylation in Jurkat-LSECtin cells expressing WT but not mutant DAP12, although mutant DAP12 can also co-precipitate LSECtin (Fig 4D). More importantly, MDDCs stimulated with eVLP were found to induce DAP12 phosphorylation, but knockdown of LSECtin resulted in a substantial decrease in DAP12 phosphorylation (Fig 4E). In addition, the phosphorylation of DAP12 is independent on TLR4 activation as the effect is not affected after TLR4 “knockdown” in MDDCs (S11 Fig). These results suggest that eVLP-triggered DAP12 phosphorylation is mediated through LSECtin.
In addition, we also used plate-coated anti-LSECtin CFD051 mAb to stimulate LSECtin- and DAP12-transfected cells. This treatment also increased the phosphorylation of protein tyrosines in Jurkat cells expressing LSECtin and DAP12 (Fig 4F). And the enhanced phosphorylation of protein tyrosines is dependent on the two tyrosines within the ITAM of DAP12 (Fig 4G). These results indicate that ligation of LSECtin can induce tyrosine kinase-based intracellular signals in the presence of DAP12.
The ITAM in intracellular domain of DAP12 can be phosphorylated and transduce signaling via inducing the phosphorylation of Syk. To determine whether endogenous LSECtin could activate the Syk tyrosine kinase, we stimulated MDDCs with plate-bound anti-LSECtin mAb. Consistent with the results shown in Fig 2A, only plate-bound CFD051 Ab induced phosphorylation of the kinases Syk and ERK in MDDCs (Fig 5A). In addition, the Syk inhibitor piceatannol abrogated the expression of cytokines in MDDCs induced by CFD051 in mRNA and protein levels (Fig 5B and 5C). Identical results were achieved with interfering RNAs (siRNAs) against Syk (S12 Fig), confirming specificity of the Syk inhibitor. Syk inhibition by piceatannol also abrogates the enhanced expression of TNF-α and IL-6 by LSECtin-TLR4 cross-talk (S13 Fig). Considering that another C-type lectin, DC-SIGN, has been shown to mediate signal transduction through Raf-1 [21], we investigated whether Raf-1 is also involved in LSECtin-mediated signaling. The Raf inhibitor GW5074 did not inhibit the expression of cytokines in MDDCs induced by CFD051 (S14 Fig). We next determined whether ligation of LSECtin by eVLP leads to the activation of Syk and ERK. We found that the activation of Syk and ERK induced by eVLP was significantly impaired in LSECtin “knockdown” DCs (Fig 5D). In addition, the phosphorylation of ERK induced by eVLP was also significantly impaired in TLR4 “knockdown” DCs, but the activation of Syk is independent on TLR4 signaling (S15 Fig). These results suggest that eVLP activates Syk through a LSECtin-dependent way. We then examined whether Syk is involved in cytokine expression induced by eVLP. Syk inhibition by piceatannol reduced the production of TNF-α and IL-6 by DCs stimulated with eVLP (Fig 5E), plate-bound GP1-Fc (Fig 5F) or eVLPm (Fig 5G). R406 is another specific Syk inhibitor and in clinical trials for human inflammatory diseases [30]. In the presence of R406, the production of TNF-α and IL-6 by DCs is also significantly suppressed after the stimulation by eVLPs (Fig 5H), plate-bound GP1-Fc (Fig 5I) or eVLPm (Fig 5G). As a control, LPS does not activate Syk kinase and the cytokine production induced by LPS is not suppressed in the presence of either piceatannol or R406 (S16 Fig), which suggests that the inhibition of Syk signaling by piceatannol or R406 is specific to the eVLP and GP1-Fc treatment.
The adaptor molecule CARD9 is required for Syk-mediated inflammatory responses [31]. We therefore investigated the role of CARD9 in LSECtin-mediated responses in MDDCs transfected with siRNA specific for CARD9 or with control siRNA (S17 Fig). We found that LSECtin failed to induce the production of TNF-α and IL-6 in CARD9 “knockdown” MDDCs after treatment with CFD051 antibody (Fig 5J). More importantly, the production of TNF-α and IL-6 by DCs is significantly reduced in CARD9 “knockdown” MDDCs after the stimulation by eVLPs (Fig 5K), plate-bound GP1-Fc (Fig 5L) or eVLPm (Fig 5M). Taken together, these results indicate LSECtin engagement is capable of activating Syk and downstream signaling pathways in DCs, leading to the production of cytokines.
Here, we show that the myeloid C-type lectin receptor LSECtin is a DAP12-coupled activating receptor that induces inflammatory responses by recognizing EBOV GP. LSECtin crosslinked by mAb or ligated with EBOV GP induces the phosphorylation of protein tyrosines and up-regulates the expression of proinflammatory cytokines via Syk and CARD9. Transduction of these events is dependent on the interaction between LSECtin and DAP12, which bears an ITAM in its cytoplasmic domain.
Signaling transduced by some members of the CLRs is crucial for tailoring immune responses to pathogens [32]. To investigate the role of CLRs in regulation of myeloid cell function, mAbs to selectively trigger surface receptors has been widely used, which provides important insight into the signaling and function of different CLRs [33]. We used mAbs to selectively crosslink LSECtin, inducing Syk- and CARD9-dependent inflammatory cytokine production in DCs. It is noteworthy that the anti-LSECtin mAbs failed to induce cytokine production by DCs after transfection with LSECtin siRNA, which further confirms the specificity of the anti-LSECtin mAbs. Thus, LSECtin signaling by itself is sufficient to induce activation of the Syk/CARD9 pathway and gene expression. We also observed that anti-LSECtin mAb treatment combined with LPS enhanced the production of TNF-α and IL-6 by DCs, which indicates that LSECtin might regulate TLR signaling.
C-type lectins comprise a heterogeneous group of transmembrane proteins that recognize various self- and non-self-ligands [19]. These characteristics of CLRs increase the host’s flexibility in recognizing various molecular patterns, including those in exogenous pathogens and endogenous ligands. Our previous studies showed that LSECtin binds activated T cells and inhibits their function through an unidentified endogenous ligand [34]. However, the function of LSECtin as a PRR is still undefined. We have shown here that LSECtin recognizes Ebola GP and transduces an activating signal in DCs. This is contrary to DC-SIGN-mediated immunomodulatory function. For example, DC-SIGN was employed by measles virus to suppress antiviral type I IFN responses and then escape antiviral immunity [35]. It is noteworthy that we used eVLPs or plate-coated GP1-Fc to induce the production of proinflammatory cytokines by MDDCs. Previous studies showed that soluble GP1 alone does not induce cytokine production in human macrophages [15], and we confirmed this with MDDCs. The data is different from that soluble shed GP which can induce the secretion of cytokines. Shed GP is a trimer, but GP1-Fc is a monomer in our study (S1 Fig). Therefore, the different structures of soluble shed GP and GP1 maybe cause their varied ability of activating DCs. In addition, sera Lectins especially MBL in FBS used in our stimulation systems might also interfere GP binding DCs since MBL present in human sera is capable of affecting the binding of shed GP to cells [13].
DAP12 contains a cytoplasmic ITAM that recruits Syk and promotes activation of ERK [36, 37]. Piceatannol and R406, two Syk inhibitors, both significantly inhibit the cytokine production induced by eVLPs or plate-coated GP1-Fc. eVLPs trigger protein tyrosine phosphorylation in LSECtin- and DAP12-co-expressing Jurkat cells, and this effect is dependent on the ITAM of DAP12. Alignment of the LSECtin amino acid sequence indicates that 2 amino acids within CRD, Asn256 and Asn274, interact with Ca2+ through their carbonyl groups. Recognition of Ebola GP by LSECtin appears to be dependent on carbohydrates, as eVLPs do not trigger protein tyrosine phosphorylation in mutant LSECtinN256D or N274D- and DAP12-co-expressing Jurkat cells. These results show that LSECtin is a novel DAP12-coupled myeloid CLR that acts as a PRR for Ebola GP.
Fatal EBOV infection in humans is associated with severe immune dysregulation and the hypersecretion of numerous proinflammatory cytokines. Recently, it has been demonstrated that trimeric shed GP released from virus-infected cells could activate non-infected DCs and macrophages, causing massive release of pro- and anti-inflammatory cytokines [13]. In addition, Qiu et al. reported that ZMapp, a blend of three EBOV GP-specific mAbs, protected EBOV-infected nonhuman primates [38]. This protection occurred even when ZMapp administered 5 days after infection, a time at which the clinical signs of disease are apparent. However, the mechanisms by which protection is achieved are unclear [39]. Given that GP participates in the production of numerous proinflammatory cytokines, it is reasonable to speculate that ZMapp not only neutralizes EBOV infection but also inhibits the excessive cytokine storm by blocking the interaction between GP and its PRRs, such as TLR4 and LSECtin. Therefore, therapeutic strategies to inhibit the cytokine storm should be considered during treatment for Ebola infection, especially for the patients with obvious clinical symptoms. In this regard, treatment with anti-GP, anti-TLR4 and anti-LSECtin Abs could be used to reduce the inflammatory responses caused by shed GP and may be helpful to alleviate the septic shock-like syndrome observed with EBOV infection.
mAbs to human LSECtin were established by immunization of Balb/C mice with recombinant LSECtin extracellular domain protein. Three independent clones, CCA023 (IgG2a), CFD051 (IgG1) and CCB059 (IgG2b), were established [34]. The anti-human LSECtin mAb CCB059 (IgG2b) was selected for staining by flow cytometry. The mouse IgG1 isotype control was from R&D Systems (Minneapolis, MN, USA). mAbs against human HLA-DR, CD83 and CD86 were from eBioscience (San Diego, CA, USA); mAbs against human CD40 and CD80 were from Biolegend (San Diego, CA, USA); anti-phosphotyrosine Ab (4G10) was from Millipore; anti-DAP12 and the other phospho-specific Abs were from Cell Signaling Technology (Danvers, MA). The Syk inhibitors piceatannol and R406 were purchased from Calbiochem (San Diego, CA, USA) and Selleckchem (Houston, TX, USA) respectively. Raf-1 inhibitor GW5074 was purchased from Calbiochem (San Diego, CA, USA); The MyD88 inhibitory peptide Pepinh-MYD was from InvivoGen (San Diego, CA, USA). The GlcNAc β1-2Man disaccharide was purchased from Dextra Laboratories (Reading, UK).
The Ebola GP1 coding sequence used is from the GP gene of the Zaire EBOV strain Mayinga (GenBank accession no. AF272001), which contains eight adenosine (A) residues at the editing site. The coding sequence was synthesized by TSINGKE Biological Technology. The GP1 cDNA was cloned by PCR and inserted into pIRES2-EGFP-Fc vectors such that the recombinant protein contained the Fc portion of human IgG. The pIRES2-EGFP-GP1-Fc plasmid was transfected into 293T cells, and the supernatants (free of FBS) were collected for protein purification using protein A/G agarose (GE Healthcare). To determine the content, purified GP1-Fc was subjected to Coomassie blue staining and Western blotting.
The generation of Ebola VLPs in insect cells (eVLP) has been described previously [26]. Briefly, recombinant baculoviruses co-expressing Ebola VP40 and GP (rBV-GP-VP40) proteins or only expressing Ebola VP40 (rBVVP40) proteins infect Spodoptera frugiperda Sf9 insect cells at an MOI of 1. After 48h, the supernants were collected and VP40 and eVLPs proteins were purified in a discontinuous sucrose gradient (10–50%). A visible band between the 30% and 50% sucrose layers was harvested, concentrated by ultracentrifugation and then resuspended in PBS.
Ebola VP40 and GP genes were cloned into pIRES2-EGFP. Mammalian 293T cells were transfected with pIRES2-EGFP-VP40 alone or in combined with pIRES2-EGFP-GP expression vectors at equal DNA concentrations. 48h post-transfection, the supernatants (free of FBS) were collected and clarified with a cell spin. VLPs were purified by centrifugation through a sucrose cushion at 26000 rpm in a Beckman SW-28 rotor for 2 h at 4°C. eVLPs were resuspended in PBS. VP40 and eVLPs containing VP40 and GP proteins produced in mammalian 293T cells was designated VP40m and eVLPm respectively. The final concentration of eVLP protein was quantitated using the DC protein assay (Bio-Rad, Hercules, CA).
Human peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats from healthy donors using a Ficoll-Paque Plus (GE Healthcare, Piscataway, NJ) gradient. Monocytes were purified from the PBMCs by adherence for 1h at 37°C in complete medium and were differentiated into MDDCs in the presence of 800U/ml GM-CSF and 400U/ml IL-4 (PeproTech). The DCs were stimulated with plate-bound anti-LSECtin mAb, eVLPs eVLPm or plate-bound GP-Fc (10μg/ml) for the indicated times and then lysed and subjected to Western blotting to detect the phosphorylation of Syk and ERK.
RNA was isolated with RNAeasy Mini Kit (Qiagen, Valencia, CA) and cDNA was synthesized with First Strand cDNA Synthesis Kit (Fermentas). Quantitative PCR was performed with a SYBR Green PCR kit (Roche, Laval, Canada) in an iQ5 (Bio-Rad) detection system. The sequences of the primer pairs of TNF-α, IL-6, CARD9 and TLR4 were described before [40–43]. LSECtin primer pairs were purchased from Qiagen.
MDDCs were transfected with 20 nM siRNA using the transfection reagent INTERFERin (Polyplus Transfection) as described [44]. Briefly, 5×105 cells were seeded into 6-well plates and then transfected with corresponding siRNAs. After 6 hours, culture medium was replaced with fresh growth medium to reduce cellular toxicity of the transfection reagent. The siRNA sequence was as follows: LSECtin-specific siRNA, 5′-GCGCGAGAACTGTGTCATGAT-3′; DAP12-specific siRNA, 5′- ACAGCGTATCACTGAGACC-3′ [45]; and negative control siRNA, 5′-TTCTCCGAACGTGTCACGTTT-3′. At 48h after transfection, the cells were stimulated. Syk and TLR4 siRNA was purchased from Dharmacon. CARD9 siRNA was purchased from OriGene.
The sequence of the gene encoding human LSECtin was obtained from the National Center for Biotechnology Information’s server (GenBank accession no. Q9NY25). LSECtin cDNA was cloned by PCR and introduced into the pcDNA3.1/Myc-His A vector, which has a Myc tag at the N terminus, as did the different LSECtin mutants. Human FceRIγ and DAP12 were inserted into the pCMV-Flag-Mat-1 vector with a Flag tag at the N terminus. To determine how LSECtin associates with DAP12, we constructed different LSECtin mutants. Myc-LSECtin ΔICD lacks the entire intracellular domain (1-31aa). Myc-LSECtin ΔICD&TM lacks the entire intracellular and transmembrane domains (1-55aa). TMΔ1(deletion of 32-43aa), TMΔ2 (deletion of 44-49aa) and TMΔ3 (deletion of 50-55aa) of Myc-LSECtin were confirmed to lack different transmembrane regions, as indicated.
A Student t test was used for statistical analysis. Results with a P value of less than 0.05 were considered as statistically significant.
Peripheral blood mononuclear cells (PBMC) are collected from healthy human volunteer donors under approval of Institutional Review Board of Academy of Military Medical Science. The study did not involve any direct contact with human subjects and all samples were anonymized.
|
10.1371/journal.ppat.1005019 | Herpesvirus Genome Recognition Induced Acetylation of Nuclear IFI16 Is Essential for Its Cytoplasmic Translocation, Inflammasome and IFN-β Responses | The IL-1β and type I interferon-β (IFN-β) molecules are important inflammatory cytokines elicited by the eukaryotic host as innate immune responses against invading pathogens and danger signals. Recently, a predominantly nuclear gamma-interferon-inducible protein 16 (IFI16) involved in transcriptional regulation has emerged as an innate DNA sensor which induced IL-1β and IFN-β production through inflammasome and STING activation, respectively. Herpesvirus (KSHV, EBV, and HSV-1) episomal dsDNA genome recognition by IFI16 leads to IFI16-ASC-procaspase-1 inflammasome association, cytoplasmic translocation and IL-1β production. Independent of ASC, HSV-1 genome recognition results in IFI16 interaction with STING in the cytoplasm to induce interferon-β production. However, the mechanisms of IFI16-inflammasome formation, cytoplasmic redistribution and STING activation are not known. Our studies here demonstrate that recognition of herpesvirus genomes in the nucleus by IFI16 leads into its interaction with histone acetyltransferase p300 and IFI16 acetylation resulting in IFI16-ASC interaction, inflammasome assembly, increased interaction with Ran-GTPase, cytoplasmic redistribution, caspase-1 activation, IL-1β production, and interaction with STING which results in IRF-3 phosphorylation, nuclear pIRF-3 localization and interferon-β production. ASC and STING knockdowns did not affect IFI16 acetylation indicating that this modification is upstream of inflammasome-assembly and STING-activation. Vaccinia virus replicating in the cytoplasm did not induce nuclear IFI16 acetylation and cytoplasmic translocation. IFI16 physically associates with KSHV and HSV-1 genomes as revealed by proximity ligation microscopy and chromatin-immunoprecipitation studies which is not hampered by the inhibition of acetylation, thus suggesting that acetylation of IFI16 is not required for its innate sensing of nuclear viral genomes. Collectively, these studies identify the increased nuclear acetylation of IFI16 as a dynamic essential post-genome recognition event in the nucleus that is common to the IFI16-mediated innate responses of inflammasome induction and IFN-β production during herpesvirus (KSHV, EBV, HSV-1) infections.
| Herpesviruses establish a latent infection in the nucleus of specific cells and reactivation results in the nuclear viral dsDNA replication and infectious virus production. Host innate responses are initiated by the presence of viral genomes and their products, and nucleus associated IFI16 protein has recently emerged as an innate DNA sensor regulating inflammatory cytokines and type I interferon (IFN) production. IFI16 recognizes the herpesvirus genomes (KSHV, EBV, and HSV-1) in the nucleus resulting in the formation of the IFI16-ASC-Caspase-1 inflammasome complex and IL-1β production. HSV-1 genome recognition by IFI16 in the nucleus also leads to STING activation in the cytoplasm and IFN-β production. However, how IFI16 initiates inflammasome assembly and activates STING in the cytoplasm after nuclear recognition of viral genome are not known. We show that herpesvirus genome recognition in the nucleus by IFI16 leads to interaction with histone acetyltransferase-p300 and IFI16 acetylation which is essential for inflammasome assembly in the nucleus and cytoplasmic translocation, activation of STING in the cytoplasm and IFN-β production. These studies provide insight into a common molecular mechanism for the innate inflammasome assembly and STING activation response pathways that result in IL-1β and IFN-β production, respectively.
| Kaposi’s sarcoma associated herpes virus (KSHV), a γ-2 herpesvirus, is etiologically associated with Kaposi’s sarcoma (KS) and primary effusion lymphoma (PEL) [1]. The hallmark of KSHV infection is the establishment of latent infection, reactivation and reinfection, and KS and PEL lesion endothelial and B cells, respectively, carry episomal KSHV latent dsDNA genome [1]. Human PEL (B) cell lines BCBL-1 and BC-3 carry >80 copies of the episomal latent KSHV genome/cell and the lytic cycle can be induced by chemicals. Purified virions from the supernatants are used for in vitro infection of human dermal microvascular endothelial cells (HMVEC-d) and foreskin fibroblast cells (HFF) [2].
During infection of its target cells, KSHV must be coming in contact with the host innate immune system’s pattern recognition receptors (PRR), such as Toll-like receptors (TLRs), RIG-I-like receptors (RLRs), NOD-like receptors (NLRs) and absent in melanoma 2 (AIM2)-like receptors (ALRs). TLRs on the plasma membranes and endosomes as well as the RLRs, NLRs and AIM2 in the cytoplasm recognize pathogen or danger-associated molecular patterns (PAMP/DAMP) [3, 4, 5]. KSHV infection of HMVEC-d cells induces inflammatory cytokines including the secretion of IL-1β into the supernatants which are similar to the microenvironments of KS and PEL lesions [6]. IL-1β, IL-18 and IL-33 are synthesized as inactive proforms, undergo proteolytic processing by activated caspase-1 generated by the cleavage of procaspase-1 via inflammasomes. Most of these molecular platforms are formed by homotypic interactions of a sensor protein recognizing the danger trigger, adaptor molecule ASC (apoptosis-associated speck-like protein containing CARD), and the effector procaspase-1. NLRs are cytoplasmic inflammasome sensors of foreign molecules, including ROS, K++, alum, bacterial products, RNA and RNA viruses replicating in the cytoplasm, while AIM2 recognizes cytoplasmic DNA including transfected DNA and DNA of pox viruses replicating in the cytoplasm [4, 7, 8, 9]. They initiate the host defenses by regulating the production of IL-1β, IL-18, IL-33 or type I interferons (IFN) α/β [7,8,9,10].
Whether innate responses recognize and respond to the presence of foreign episomal genomes of herpesviruses as well as other DNA viruses in the infected cell nuclei leading into the induction of inflammatory responses was not known initially. Our studies revealed that in vitro KSHV infection of endothelial cells induces caspase-1 activation via the nuclear resident gamma-interferon-inducible protein-16 (IFI16) also known as interferon-inducible myeloid differentiation transcriptional activator. Colocalization of IFI16 with viral genome in the infected endothelial cell nucleus, induction of IFI16-ASC inflammasomes by UV-inactivated KSHV and the absence of induction by lentivirus vectors expressing KSHV genes demonstrated that a) KSHV genes individually do not play a role in IFI16-inflammasome activation, b) the IFI16-inflammasome is not induced against linear integrated foreign DNA, and c) episomal KSHV genome is required for IFI16-inflammasome activation [11]. When we analyzed the gene expression in uninfected and infected HMVEC-d cells, a significant increase in caspase-1 gene expression from 2 to 24 h post-infection (p.i.), significant induction of the ASC gene only at 24 h p.i., a slight but not significant increase in IFI16 gene expression, and no increase in NLRP-1, NLRP3 and AIM2 genes were observed [11].
We have subsequently demonstrated that only the IFI16-inflammasome is constitutively induced in KSHV latently infected endothelial and PEL cells [12], as well as in B-lymphoma, epithelial and lymphoblastoid cells latently infected with γ-1 Epstein-Barr virus (EBV) [13]. Colocalization of IFI16 with the latent KSHV and EBV genome in the nuclei suggested that continuous sensing of latent genome results in the constitutive induction of IFI16-ASC inflammasomes. In addition, our studies showed that IFI16 recognizes the α-herpes simplex virus type-1 (HSV-1) genome soon after its entry into the nucleus resulting in the formation of IFI16-inflammasomes [14].
The 730 aa (1–2190 bp) IFI16 protein consists of an n-terminal ASC interacting PYRIN domain (41–261 bp), 200-amino-acid HIN I (401–895 bp) and HIN II (1043–1541 bp) domains involved in the sequence independent DNA recognition, and 2 nuclear localizing signals (NLS; 296–311 and 387–407 bp) which attribute to its nuclear entry after synthesis in the cytoplasm [15]. Though IFI16 is a predominately nuclear protein, after recognizing KSHV and HSV-1 DNA during de novo infection, the IFI16-ASC complex initially colocalized in the infected cell nucleus and subsequently localized in the perinuclear areas [11, 14]. Similarly, we observed the colocalization of IFI16 and ASC both in the nucleus and cytoplasm of cells latently infected with KSHV and EBV [12, 13]. Western blot analysis of de novo KSHV infected HMVEC-d cells showed steady levels of ASC and procaspase-1 in the nuclear fractions. Infected cells also showed higher levels of both ASC and procaspase-1 in the cytoplasmic fractions which demonstrated that ASC and procaspase-1 undergo subcellular redistribution upon infection. Active caspase-1 (p20) was detected in the nucleus of infected HMVEC-d cells at 2 and 8 h post-infection demonstrating that the inflammasome is activated upon sensing KSHV in the nucleus, and the majority of activated caspase-1 was subsequently detected in the cytoplasmic fractions at later times of infection probably to prevent caspase-1 mediated adverse activities in the nucleus. Detection of caspase-1 in the cytoplasm during de novo KSHV and HSV-1 infection as well as in latently infected cells demonstrated that after recognizing viral DNA in the nucleus, the newly formed IFI16-ASC inflammasome complex is transported to the cytoplasm [11, 12, 13, 14]. However, the mechanism behind the redistribution of this complex is not known.
HSV-1 infection also induced IRF-3 phosphorylation through the IFI16-STING interaction in the cytoplasm. Even though the recognition of HSV-1 genome in the nucleus via IFI16 is suggested to be the factor behind the cytoplasmic STING-IRF-3 activation and IFN-β production early during infection [16], the mechanism of post-genome detection signaling from nucleus to cytoplasm resulting in STING activation is not known. KSHV infection induces only a moderate IFN-β response early during de novo infection which was inhibited by a variety of early lytic and latent gene products at later times of infection [17]. The role of IFI16 in IFN-β production during KSHV infection is not known.
Using IFI16-EGFP constructs transfection in human osteosarcoma U2OS cells, Li et al., [15] studies showed that the two NLS motifs of IFI16 (aa 96–100 and aa 128–131) are essential for the entry of newly synthesized IFI16 in the cytoplasm to the normal cell nucleus. Using a FISH assay, they demonstrated that during HSV-1 (strain 17+) infection of U2OS cells (5 PFU/cell) containing transfected IFI16-EGFP construct, virion DNA colocalized only with full length IFI16-EGFP with intact NLS and not with mutated NLS-IFI16-EGFP that were localized in the cytoplasm. They also observed that as reported by us for KSHV [11, 12], EBV [13] and HSV-1 [14], a subset of wild type IFI16 translocated to the cytoplasm. In addition, co-IP of HSV-1 DNA-protein complexes followed by qPCR with four HSV-1 primer sets (UL30, US6,RL1 and RS1) demonstrated the nuclear IFI16 interaction with viral DNA in the nucleus. Using uninfected U2OS transfected with DNA, Li et al., [15] concluded that acetylation at the NLF motifs of IFI16 results in the cytoplasmic retention of newly synthesized IFI16 by prohibiting nuclear import, and the histone acetyltransferase p300 regulated the cytoplasmic IFI16 acetylation during transfection of DNA. However, the fate of nuclear IFI16 during HSV-1 infection, whether IFI16 undergo acetylation during HSV-1 infection, the role of p300 during viral DNA recognition in the nucleus, and the mechanism behind the IFI16 redistribution into the cytoplasm during infection was not studied [15].
Here, we demonstrate that the presence of KSHV genome in the nucleus induces the p300 mediated acetylation of IFI16 and this modification is the driving force behind the nuclear to cytoplasmic redistribution of the IFI16-inflammasome which was facilitated by Ran-GTPase. IFI16 acetylation is required for its interaction with ASC, inflammasome assembly and function. In addition, cytoplasmic redistribution of acetylated IFI16 is also essential for STING-IRF-3 mediated IFN-β production in KSHV and HSV-1 infected cells. These studies for the first time demonstrate that IFI16 acetylation is a dynamic post-herpes viral genome recognition event required for the IFI16-mediated innate responses of inflammasome induction (KSHV, EBV and HSV-1) and IFN-β production (KSHV and HSV-1).
KSHV enters HMVEC-d and HFF cells by a rapid endocytic process which is followed by the transport of genome-containing capsid to the nuclear pore vicinity, capsid disassembly and entry of the linear dsDNA into the nucleus within 15–30 min p.i., followed by the establishment of a latent infection [18]. Our studies have shown that IFI16 colocalized with the KSHV genome at 2 h p.i. in the nucleus of HMVEC-d cells [11]. To determine the earliest time of interaction of IFI16 with KSHV genome, HMVEC-d cells were infected with KSHV containing BrdU-labeled genome (BrdU-KSHV) and immunostained with anti-BrdU antibodies (Fig 1A; Table 1). IFI16 was predominantly localized in the uninfected cell nucleus (Fig 1A, top panel). By 15 min p.i., viral particles were seen in the cytoplasm and near the nuclear periphery (Fig 1A, red arrows, middle panel). In contrast, significant accumulation of viral DNA was observed at 30 min p.i. in the infected cell nuclei, and most of them colocalized with IFI16 (Fig 1A, white arrows). In addition, a few IFI16 signal spots were also detected in the cytoplasm at 30 min p.i. (Fig 1A, yellow arrow). These results suggested that IFI16 senses the KSHV genome soon after its entry into the nucleus during de novo infection with a concomitant redistribution to the cytoplasm.
To determine the kinetics of IFI16 redistribution to the cytoplasm, the cytoplasmic and nuclear fractions from uninfected cells and cells infected with KSHV for various times were analyzed by western blots (WB). Consistent with the IFA results, a very faint IFI16 band was detected at 30 min p.i. in the cytoplasm which steadily increased during the observed period of 24 h p.i. (Fig 1B, lanes 9–12) with a corresponding decrease in the nuclear IFI16 levels (Fig 1B, lanes 4–6). TBP and tubulin proteins were used as markers of nuclear and cytoplasmic preparation purity and as controls for equal loading (Fig 1B, lanes 1–12). When IFA was performed to validate the biochemical data, IFI16 was predominantly in the nucleus of uninfected cells (S1A Fig, top panel). In contrast, at 30 min p.i., few IFI16 signal spots were visible in the cytoplasm which increased steadily during the observed period of 24 h p.i. (S1A Fig, red arrows). These results demonstrated that KSHV infection induces IFI16 redistribution from the nucleus to the cytoplasm as early as 30 min p.i. with steady increase thereafter.
IFI16 has been shown to function as a transcriptional modulator via unknown mechanisms [19]. We theorized that acetylation of IFI16 could be one of the reasons for cytoplasmic transport since acetylation of HMGB-1 (high-mobility group protein B1) protein involved in transcription/ chromatin bending has been shown to result in HMGB-1’s translocation into the cytoplasm [20]. Furthermore, IFI16 acetylation within the NLS motifs during transfection of DNA in U20S cells promoted cytoplasmic retention by blocking nuclear import of newly synthesized IFI16 [15]. However, the fate of IFI16 during nuclear DNA sensing was not studied.
To investigate the acetylation status of IFI16 during KSHV infection, uninfected and infected cell lysates were immunoprecipitated (IP-ed) with anti-acetylated lysine antibody and western blotted for IFI16. Compared to the uninfected cells, we observed a robust increase in the acetylation of IFI16 only in the infected cells (Fig 1C, lanes 1 and 2). In contrast, equal levels of acetylated tubulin were observed in both uninfected and KSHV infected cells (Fig 1C, lanes 1 and 2). The input IFI16 and loading control tubulin were of similar levels. These results suggested that the acetylation machinery was functional in both uninfected and infected cells and KSHV infection induced increased acetylation of IFI16.
When we next investigated the kinetics of IFI16 acetylation in the nuclear and cytoplasmic fractions by co-IP experiments, as early as 30 min p.i. an appreciable level of nuclear IFI16 acetylation was observed which steadily increased during the observed 24 h p.i. (Fig 1D, lanes 2–6). Correspondingly, we detected a faint band of acetylated IFI16 in the cytoplasm at 30 min p.i., with steady increase from 2 to 24 h p.i. (Fig 1D, lanes 9–12), which corroborated the results in Fig 1B, lanes 9–12. The faint acetylated IFI16 band detected in the nucleus of uninfected cells probably represents the basal level (Fig 1D, lane 1). These detections were not due to nuclear contamination as shown by the absence of TBP and presence of tubulin in these fractions (Fig 1D). As positive control for nuclear and cytoplasmic acetylation, the proteins were IP-ed with acetylated lysine antibody and western blotted for H3 and tubulin, respectively (Fig 1D, lanes 1–12). Total H3 level was also analyzed by western blot as input control.
These results were also validated by IFA using anti-IFI16 and anti-acetylated lysine antibodies (S1B Fig). In the uninfected cells, IFI16 was detected in the nucleus and acetylated lysine signals were observed both in the nucleus and in the cytoplasm (S1B Fig, top panel). We also observed some basal level of IFI16 and acetylated lysine colocalization in the nucleus of uninfected cells (S1B Fig, UI, red arrow). In contrast, KSHV infection significantly increased the colocalization of acetylated lysine and IFI16 in the nucleus as well as in the cytoplasm in a time dependent manner (S1B Fig). Taken together, these results demonstrated that during de novo KSHV infection, IFI16 recognizes the viral genome with a concomitant increase in its acetylation in the nucleus and redistribution of acetylated IFI16 to the cytoplasm of the infected cells.
The cellular transcriptional coactivator protein p300 functions as a histone acetyltransferase and has been shown to be involved in the cytoplasmic acetylation of IFI16’s NLS domains [15]. To investigate the significance of nuclear acetylation of IFI16 and its redistribution, we utilized the p300 competitive inhibitor C-646. Based on the results in BCBL-1 and HMVEC-d cells incubated with various concentrations of C-646 for 4 and 24 h (S2A and S2B Fig) we selected the least toxic 1 μM concentration (5–6% cell death) for all further experiments. C-646 treatment did not interfere with viral entry or nuclear delivery of viral genome, and equal levels of the characteristic KSHV latent LANA-1 protein dots were detected in the treated and untreated cells (S2C, S2D, and S2E Fig). Significant increase in acetylation was observed in the KSHV infected cells which was reduced by C-646 treatment (S2F Fig, lanes 1–4). The specificity of C-646 was examined by the acetylation level of H2B, one of the target proteins of p300. IP with acetylated lysine antibody and WB for H2B showed six fold reduction in H2B acetylation by C-646 compared to the untreated KSHV (24 h) infected cells (S2G Fig, lanes 1 and 2). These results demonstrated that de novo KSHV infection induced acetylation, which is in part due to p300, can be inhibited by C-646.
To determine the effect of C-646 on IFI16 acetylation, HMVEC-d cells were either uninfected or infected with KSHV in the presence or absence of C-646, whole cell lysates IP-ed with anti-acetylated lysine antibody and western blotted for IFI16. Compared to untreated infected cells, C-646 treatment completely abolished the infection induced IFI16 acetylation (Fig 1E, lanes 1–6). Immunoprecipitation of IFI16 followed by WB for IFI16 demonstrated equal pull down; in addition, β-actin levels did not change due to treatment and showed equal loading (Fig 1E, lanes 1–6). IP of IFI16 and WB with anti-acetylation antibody also validated these results which showed decreased levels of acetylated IFI16 by C-646 treatment in infected cells (Fig 1F, lanes 1–3). To investigate the effect of C-646 on KSHV infection induced acetylation mediated cytoplasmic redistribution of IFI16, HMVEC-d cells were infected in the absence or presence of C-646, cytoplasmic and nuclear fractions isolated and western blotted for total IFI16. KSHV infection induced redistribution of IFI16 into the cytoplasm was abolished in C-646 treated cells (Fig 1G, lanes 7–12). Interestingly, we also observed that the nuclear IFI16 levels decreased at later time points by C-646 (Fig 1G, lanes 4–6) which suggested that acetylation may have a role in the stabilization of IFI16.
These results demonstrated that IFI16 acetylation during KSHV infection is dependent on p300 and acetylation is required for the redistribution of IFI16 from the nucleus to the cytoplasm after recognition of the KSHV genome in the nucleus.
To validate these results, we performed in situ-PLA which detects endogenous levels of proteins and gives the spatial distribution and localization of a single or multiple proteins (Fig 2A). PLA uses oligonucleotide-linked secondary antibodies and a fluorescence-based assay to detect closely associated proteins. If epitopes of a single protein or two protein epitopes are within 40 nm proximity, the antibody-linked oligonucleotides will ligate with adaptor oligonucleotides to form complete circles that are amplified via DNA replication and detected with fluorescent sequence-specific probes which will appear as distinct dots visible under fluorescent microscopy.
HMVEC-d cells were uninfected or infected in the presence or absence of C-646 and subjected to PLA using rabbit and mouse anti-IFI16 antibodies detecting different epitopes, and the detected red dots depict IFI16 (Fig 2A). IFI16 was predominantly nuclear in both untreated and C-646 treated uninfected cells (Fig 2A, top 2 panels, yellow arrows). In the absence of C-646, we observed abundant cytoplasmic IFI16 localization in KSHV infected cells at 24 h p.i. (Fig 2A, lower panels, white arrows), and an uninfected cell in the same field showed predominantly nuclear IFI16 (Fig 2A, blue arrow). In contrast, while IFI16 was detected in the nucleus of C-646 treated infected cells, we did not observe IFI16 redistribution in the cytoplasm (Fig 2A, lower panels). These results demonstrated that inhibition of acetylation compromised the cytoplasmic redistribution of IFI16.
To further elucidate the effect of C-646 on acetylation of IFI16, PLA was performed using anti-IFI16 and anti-acetylated lysine antibodies and the observed red dots represent the acetylated IFI16 (Fig 2B). Low levels of nuclear acetylated IFI16 PLA dots were detected both in the treated and untreated uninfected cells (Fig 2B, top panel, yellow arrows). In contrast, at 30 min p.i., acetylated IFI16 dots were appreciably increased in the nucleus with few dots visible in the cytoplasm, which increased to numerous acetylated IFI16 spots in a time dependent manner (Fig 2B, left panels, white arrows). In contrast, with C-646 treatment the acetylated IFI16 dots did not increase either in the cytoplasm or in the nucleus of infected cells (Fig 2B, lower three right panels). These studies demonstrating the reduction in cytoplasmic IFI16 redistribution by C-646 treatment validated our findings, and confirmed that IFI16 acetylation in the nucleus during KSHV infection is required for its redistribution to the cytoplasm.
We have previously shown that replication incompetent UV treated KSHV (UV-KSHV) enters the cells, delivers the viral DNA into the nucleus and induces the IFI16-inflammasome [11], which demonstrated that the presence of KSHV genome is the requirement for IFI16 recognition and further consequences. When lysates from HMVEC-d cells infected with KSHV or UV-KSHV for 24 h were IP-ed with anti-acetylated lysine antibody and western blotted for IFI16, similar to live-KSHV infected cells, acetylation of IFI16 increased in a time dependent manner by infection with UV-KSHV (Fig 2C, lanes 1–7). These results suggested that the presence of viral genome is enough to induce the IFI16 acetylation process and viral gene expression is not required.
We next determined whether acetylation of IFI16 and its cytoplasmic redistribution also occur in other cell types. Compared to uninfected cells, as in HMVEC-d cells, KSHV infected HFF cells (24 h p.i.) showed increased acetylation of IFI16 which was significantly inhibited by C-646 (S3A Fig, lanes 1–4), and WB for total IFI16 showed a slight reduction in C-646 treated cells (S3A Fig, lanes 1–4). In PLA analysis, infected HFF cells in the absence of the inhibitor showed robust acetylation of IFI16 and its redistribution to the cytoplasm, which was significantly abrogated by C-646 (S3B Fig). Uninfected cells showed only a basal level of acetylated IFI16 in the nucleus (S3B Fig). Evaluation of the total IFI16 levels by PLA using mouse and rabbit anti-IFI16 antibodies revealed that IFI16 was solely nuclear in the uninfected cells (S3C Fig), while the KSHV infected cells showed IFI16 both in the nucleus and in the cytoplasm (S3C Fig). However, when the cells were treated with C-646, IFI16 was only detected in the nucleus (S3C Fig). These results demonstrated that acetylation of IFI16 is essential for its redistribution to the cytoplasm of KSHV infected HFF cells.
We have shown that IFI16 recognizes the latent KSHV genome and only the IFI16-inflammasome is constitutively induced in endothelial and PEL cells carrying latent genome. Hence, we determined the acetylation status of IFI16 in these cells. Whole cell lysates from control BJAB and KSHV (+) BCBL-1 cells were IP-ed with anti-acetylated lysine antibody and western blotted for IFI16. Compared to BJAB cells, we detected increased IFI16 acetylation in BCBL-1 cells which was significantly reduced by C-646; however, total IFI16 was pulled down equally in each group (S4A Fig, lanes 1–4). Examination of total IFI16 in the cytoplasmic and nuclear fractions from untreated or C-646 treated BCBL-1 cells revealed ~6–11 fold less cytoplasmic IFI16 protein levels at 4 and 24 h of drug treatment, respectively, compared to the untreated controls (S4B Fig, lanes 4–6). These results demonstrated the acetylation dependent cytoplasmic redistribution of IFI16 in the latently infected cells. As in de novo infected cells, nuclear IFI16 protein levels also decreased in the presence of C-646 indicating that IFI16 stability in KSHV infected cells may be dependent upon its acetylation.
To validate these results, PLA was performed in BJAB and BCBL-1 cells using anti-IFI16 and anti-acetylated lysine antibodies (S4C Fig). Compared to the few nuclear acetylated IFI16 PLA dots in the BJAB cells (S4C Fig, upper left panel), we observed a significant increase in the acetylated IFI16 in the nucleus as well as in the cytoplasm of KSHV+ BCBL-1 cells (S4C Fig, lower left panel, yellow and white arrows, respectively). C-646 treatment resulted in a drastic reduction in acetylated IFI16 (S4C Fig, right panels). When PLA was done to examine total IFI16 and its redistribution in the absence or presence of C-646, we did not observe any cytoplasmic IFI16 in the BJAB cells (S4D Fig, upper panels). Corroborating the biochemical data in S4B Fig, increased nuclear and cytoplasmic IFI16 were observed in untreated BCBL-1 cells whereas IFI16 was mostly nuclear in the C-646 treated cells (S4D Fig, lower panels, yellow arrows). The KSHV latently infected endothelial (TIVE-LTC) and B (BJAB-KSHV) cells were also analyzed for IFI16 acetylation. IP of the whole cell lysates from control endothelial TIVE and BJAB, KSHV (+) TIVE-LTC and BJAB-KSHV cells with anti-acetylated antibody followed by IFI16 WB revealed significantly higher levels of acetylated IFI16 in both TIVE-LTC and BJAB-KSHV cells than in the KSHV negative control cells (S4E Fig, lanes 1–4). Equal amounts of IFI16 were detected in IP and in WB reactions (S4E Fig, lanes 1–4).
By PLA for IFI16 acetylation in the presence or absence of C-646, TIVE cells showed a minimal amount of acetylated IFI16 in both treated and untreated cells (S4F Fig, upper panels). In contrast, the TIVE-LTC cells showed increased levels of acetylated IFI16 both in the nucleus and in the cytoplasm (S4F Fig, lower left panel). This cytoplasmic redistribution of acetylated IFI16 was abolished by C-646 (S4F Fig, lower right panel). Total IFI16 levels in C-646 treated or untreated TIVE and TIVE-LTC cells were also analyzed by PLA using mouse and rabbit anti-IFI16 antibodies. In untreated and C-646 treated TIVE cells, IFI16 was solely nuclear (S4G Fig, upper panels). In contrast, TIVE-LTC cells showed robust IFI16 cytoplasmic redistribution (S4G Fig, lower left panel) which was significantly reduced by C-646 (S4G Fig, lower right panel).
Taken together, these results demonstrated that similar to de novo infected HMVEC-d cells, p300 mediated acetylation plays an important role in the cytoplasmic redistribution of IFI16 in cells latently infected with KSHV.
As an IFI16-ASC inflammasome is formed during EBV infection of B cells and in latently infected cells, we performed PLA for IFI16 and acetylated lysine in primary human B cells infected with KSHV or EBV as well as in cells latently infected with EBV (S5 Fig). Compared to uninfected cells, both KSHV and EBV infected primary B cells showed acetylation as well as cytoplasmic redistribution of acetylated IFI16 (S5A Fig). Compared to EBV negative Ramos cells, EBV latently infected Raji (latency I) and LCL (latency III) cells showed both nuclear and cytoplasmic acetylated IFI16 (S5B Fig). These results demonstrated that acetylation of IFI16 and its cytoplasmic redistribution also occur in EBV infected cells.
To determine the specificities of nuclear herpesvirus genome activation of IFI16 acetylation and its cytoplasmic distribution, we next used vaccinia virus replicating its dsDNA exclusively in the cytoplasm. The acetylation of IFI16 was not induced by vaccinia virus infection of HMVEC-d cells (S6A Fig). Only similar levels of a few dots representing basal level of acetylation were detected in the nucleus of both uninfected and vaccinia virus infected cells (S6A Fig). When mouse and rabbit antibodies were used to perform the PLA, IFI16 was predominantly detected in the nucleus of both uninfected as well as vaccinia infected HMVEC-d cells (S6B Fig). These results demonstrated that vaccinia viral DNA in the cytoplasm was not recognized by nuclear IFI16, and hence acetylation of the nuclear IFI16 and cytoplasmic translocation were not observed. These findings clearly supported our observations that the presence of nuclear KSHV, EBV and HSV-1 genomes induced the acetylation of IFI16 in the nucleus which then relocated into the cytoplasm of infected cells.
The dynamic process of exporting molecules of >50-kDa from the nucleus is initiated by exportins binding to cargo and Ran-GTP protein. The guanine-nucleotide exchange factor (GEF) of Ran that converts Ran-GDP to GTP form is in the nucleus and GTPase-activating proteins (GAPs) for Ran-GTPase are present in the cytoplasm as well as on the cytoplasmic face of the nuclear pore.
To determine whether Ran is responsible for IFI16 transport from the nucleus to the cytoplasm, the lysates from uninfected or KSHV infected HMVEC-d cells (4 h p.i.) in the presence or absence of C-646 were IP-ed with anti-Ran-GTPase antibodies and WB for IFI16. Compared to the uninfected cells that showed a basal level of IFI16-RAN association (Fig 3A, lanes 1 and 2), KSHV infected cells showed robust association of IFI16 with Ran-GTPase which was inhibited by C-646 (Fig 3A, lanes 3 and 4). Comparable levels of IFI16 and Ran proteins were pulled down with their corresponding antibodies (Fig 3A, lanes 3 and 4). Higher IFI16-RanGTP association in untreated KSHV infected cells corroborated the higher cytoplasmic redistribution of IFI16 shown in the earlier figures. When PLA was performed using anti-Ran and IFI16 antibodies, consistent with the IP results, the association between these two molecules increased during KSHV infection, which was abolished by C-646 (Fig 3B). These results demonstrated that acetylation enhances the association of IFI16 with Ran-GTP during infection facilitating its transport to the cytoplasm and this association is dependent upon acetylation.
The nuclear resident IFI16 translocates to the nucleus after its translation in the cytoplasm via its two NLS domains and acetylation of NLS has been shown to retain IFI16 in the cytoplasm [15]. To determine whether the cytoplasmic IFI16 detected during KSHV de novo infection and latency represents newly synthesized IFI16 or redistributed from the nucleus, we used 50 nM Leptomycin B (LPT) to block nuclear export to the cytoplasm. This concentration of LPT was not overly toxic (6–8%) to HMVEC-d cells nor did it significantly affect the establishment of KSHV infection (S7A, S7C, and S7E Fig). When HMVEC-d cells infected with KSHV in the presence or absence of LPT were analyzed, infected cells showed enhanced cytoplasmic redistribution of IFI16 which was abolished by LPT treatment (Fig 3C, top panel, lanes 5–8). Compared to untreated cells, nuclear IFI16 increased in LPT treated cells probably due to blocked cytoplasmic redistribution (Fig 3C, top panel, lanes 1–4). Reduced cytoplasmic and increased nuclear cyclin-B1 in LPT treated cells confirmed the hampered nuclear to cytoplasmic protein transport (Fig 3C, second panel, lanes 1–8).
Since IFI16-ASC-procaspase-1 assembly was initiated in the nucleus, we next examined the effect of LPT on the transport of the other components of IFI16-inflammasomes. Procaspase-1 was detected in the nucleus of untreated uninfected and infected cells (Fig 3C, third panel, lanes 1 and 3). The increased cytoplasmic procaspase-1 in untreated infected cells was significantly decreased by LPT with a corresponding increase in the nucleus (Fig 3C, third panel, lanes 7 and 8, and 3 and 4). We have previously observed the presence of cleaved caspase-1 in the nucleus of infected HMVEC-d cells at 2 h and 8 h p.i. and only in the cytoplasm at 24 h p.i. [11]. Similarly, cleaved caspase-1 was detected in the infected cell cytoplasm at 24 h p.i. which was abolished by LPT treatment with a concomitant increase in the nucleus (Fig 3C, lanes 3, 4, 7 and 8).
When cell lysates of KSHV infected HMVEC-d cells in the presence or absence of LPT were analyzed by IP with anti-acetylated antibody and IFI16 WB, IFI16 was acetylated minimally in uninfected cells and to the same extent in untreated and LPT treated infected cells; however, tubulin was acetylated in both uninfected and infected samples (Fig 3D, top 2 panel, lanes 1–4). Similarly, the IFI16 and ASC association was equal in untreated and LPT treated infected cells (Fig 3D, third panel). Equal amounts of ASC and IFI16 were pulled down with their corresponding antibodies and their total protein levels demonstrated that these proteins were available in sufficient and equal amounts in each of the experimental groups (Fig 3D, lower panels lanes 1–4).
To rule out the possibility that the detection of acetylated IFI16 is not due to the accumulation of newly synthesized IFI16 in the cytoplasm of KSHV infected cells, we blocked protein synthesis by using cycloheximide (CHX) at 200 μg/ml which was neither toxic nor affected the KSHV infection of HMVEC-d cells (S7B, S7D, and S7F Fig). Cytoplasmic and nuclear proteins from CHX treated or untreated cells left uninfected or infected with KSHV for 4 h were isolated and subjected to western blot analysis. In the presence or absence of cycloheximide, we did not detect cytoplasmic IFI16 in the uninfected cells. In contrast, we observed similar levels of cytoplasmic IFI16 in the infected cells in both the presence and absence of cyloheximide (Fig 3E, lanes 1 to 8). These results coupled with the LPT results suggested that the increased level of IFI16 in the cytoplasm of infected cells during KSHV infection is due to the translocation of acetylated IFI16 from the nucleus into the cytoplasm.
In PLA, nuclear IFI16 was detected in the untreated and treated uninfected cells (Fig 3F, left panels, yellow arrows). In contrast, in agreement with the biochemical findings, increased cytoplasmic redistribution of IFI16 in KSHV infected HMVEC-d cells was detected (Fig 3F, top right panel, white arrows) which was abrogated in LPT treated cells (Fig 3F, lower right panel). In addition, the IFI16-ASC complex was observed in both the cytoplasm and nucleus of infected cells which was constrained to the nucleus of LPT treated cells (Fig 3G, right most panels). This redistribution of IFI16-ASC complex PLA spots corroborated with earlier IFA and WB findings which demonstrated that IFI16-ASC inflammasome activation leads to redistribution of IFI16-ASC to the cytoplasm [11].
Taken together, these results demonstrated that a) blocking nuclear export by LPT did not interfere in the acetylation of IFI16, formation of IFI16-ASC complex or activation of caspase-1, b) blocking protein synthesis by CHX did not affect the cytoplasmic distribution of IFI16 from the nucleus, and c) the increased level of IFI16 in the cytoplasm in the infected cells was due to its redistribution from the nucleus and not due to newly translated cytoplasmic IFI16.
Since the redistribution of acetylated IFI16 and inflammasome activation showed a similar pattern in the infected cells, we sought to determine whether the acetylation of IFI16 and IFI16-inflammasome activation are linked or independent of each other. As the association of IFI16 with the adaptor ASC is the first step in inflammasome activation, we examined these interactions by PLA. As shown in Fig 4A and 4B, in the untreated and uninfected HMVEC-d cells, few IFI16-ASC interacting PLA dots were visible in the nucleus representing the basal level of association which was reduced by C-646 treatment (Fig 4A, top panels). In contrast, in untreated KSHV infected HMVEC-d cells, we observed a robust interaction between IFI16 and ASC both in the nucleus and in the cytoplasm (Fig 4A, yellow and white arrows, respectively, lower left panel). When the C-646 treated HMVEC-d cells were infected with KSHV, the PLA dots representing IFI16-ASC interactions in the nucleus were greatly reduced with little redistribution to the cytoplasm (Fig 4A, lower right panels).
Examination of IFI16 and ASC by IFA (S8A Fig) also revealed that IFI16 was predominantly in the nucleus of the uninfected cells, while de novo KSHV infected HMVEC-d cells showed strong IFI16-ASC colocalization in the nucleus and redistribution to the cytoplasm (S8A Fig). When the cells were treated with C-646, only minimal IFI16-ASC interaction and cytoplasmic redistribution was detected (S8A Fig, third panel). Similarly, when BCBL-1 cells were examined by PLA and IFA, strong interactions between IFI16 and ASC were detected both in the nucleus and cytoplasm which were compromised by C-646 (Fig 4B and S8B Fig). Control BJAB cells did not show considerable IFI16 and ASC interaction in either untreated or C-646 treated cells (Fig 4B).
To confirm the IFI16-ASC interactions detected by PLA, cell lysates from uninfected and 4 and 24 h de novo KSHV infected HMVEC-d cells in the presence or absence of C-646 were IP-ed with ASC and western blotted for IFI16. ASC was associated with IFI16 at 4 and 24 h p.i. but no such strong association was seen in the uninfected cells, (Fig 4C, lanes 1–3). In contrast, C-646 treatment disrupted the association between IFI16 and ASC (Fig 4C, lanes 4–6). A similar amount of IFI16 was pulled down in each group either treated or not treated with C-646 (Fig 4C, lanes 1–6). A similar to primary infection, we observed increased interaction of IFI16 with ASC in the latently infected BCBL-1 cells which was greatly reduced in C-646 treated cells (Fig 4D, lanes 3 and 4). The inputs of IFI16 and ASC were similar in all groups. These results demonstrated that the presence of KSHV genome in the nucleus induced the IFI16-ASC interaction and inflammasome formation, which are dependent upon the acetylation of IFI16 in both de novo and latent KSHV infected cells.
The IFI16-inflammasome complex is formed by the homotypic interactions between PYD domains of IFI16 and ASC and CARD domains of ASC and procaspase-1, leading into the aggregation of IFI16 molecules [11]. To confirm that the IFI16-inflammasome complex is dependent upon the acetylation of IFI16, proteins in the cell lysates cross-linked with glutaraldehyde for 10 min were used for WB. We observed high molecular weight IFI16 aggregates in de novo KSHV infected HMVEC-d cells (24 h) and in BCBL-1 cells (Figs 4E, lane 2, and 6F, lane 3) and these were severely compromised by C-646 treatment (Figs 4E, lane 3, and 6F, lane 4). No such aggregation was detected in the uninfected cells (Fig 4E, lane 1, and 4F, lanes 1 and 2). These results further confirmed that acetylation of IFI16 is critical for IFI16-inflammasome formation.
Formation of the IFI16-ASC-procaspase-1 inflammasome leads to the generation of functional caspase-1 via auto-cleavage which results in the cleavage of the pro-forms of IL-1β, IL-18 and IL-33 cytokines. Hence, we investigated the effect of C-646 on activation of caspase-1 and its downstream cytokines production. In untreated KSHV infected HMVEC-d cells, caspase-1 activation was detected at 4 and 24 h p.i., whereas, the C-646 treated counterparts did not show considerable cleavage of caspase-1 (Fig 4G, top panel, lanes 1–6). Activation of IL-1β and IL-33 was also inhibited by C-646 treatment (Fig 4G, panels 2, 3 and 4, lanes 1–6). We also observed the inhibition of procaspase-1 and pro-IL-1β cleavages by C-646 treatment in BCBL-1 cells (Fig 4H, lanes 3 and 4), and cleavage of procaspase-1 and pro-IL-1β was not detected in BJAB cells (Fig 4H, lanes 1 and 2). The C-646 treatment did not significantly affect the viability of BJAB and BCBL-1 cells (S8C Fig). Compared to uninfected cells, increased secretion of IL-1β was observed in KSHV infected HMVEC-d culture supernatants (18.5 pg/ml) which was significantly reduced (>5-fold) by C-646 treatment (3.8 pg/ml) (Fig 4I).
We next determined the levels of active caspase-1 in BCBL-1 cells with or without C-646 by FACS using fluorescent caspase-1 detection 660-YVAD-FMK probe (S8D and S8E Fig), and the percent active caspase-1 cell populations are shown in S8F Fig. Control BJAB cells unstained or stained with FLICA-660 did not show significant caspase-1 active cells (S8D and S8F Fig). In contrast, nearly 50% of the untreated BCBL-1 cells contained active caspase-1 which was reduced to ~18–19% in C-646 treated cells (S8E and S8F Fig).
These results confirmed that acetylation of IFI16 promotes formation of functional IFI16-ASC-procaspase-1 inflammasomes leading into active caspase-1 generation and downstream cytokine production in KSHV infected cells.
The recognition of viral genome by IFI16 leads into its increased interaction with ASC and inflammasome formation (Fig 4A–4D). Since reduction in IFI16 acetylation hampered IFI16-ASC association (Fig 4A–4D), we determined whether ASC plays roles in the acetylation of IFI16 and whether ASC associates with IFI16 after acetylation of IFI16. We knocked down the HMVEC-d cell ASC by Si-RNA electroporation and infected with KSHV. Knockdown efficiency confirmation by WB showed ~90–95% ASC reduction with no effect on IFI16 protein (Fig 5A, top two panels, lanes 1–8). The lysates from control and ASC knocked down cells were IP-ed with anti-acetylated lysine antibody and WB for IFI16. We observed the acetylation of IFI16 in both control and ASC knocked down cells (Fig 5A, third panel, lanes 1–8). As expected, in the absence of ASC formation of the IFI16-inflammasome complex was abrogated as shown by the absence of IFI16 in IP-reactions with anti-caspase-1 antibody and by the absence of caspase-1 activation in comparison to the Si-control KSHV infected cells. (Fig 5A, fourth, sixth and seventh panels, lanes 1–8). Caspase-1 was pulled down in all groups including ASC knocked down cells (Fig 5A, fifth panel, lanes 1–8). These results suggested that IFI16 acetylation occur independent of ASC.
We next determined whether IFI16 relocates to the cytoplasm in the absence of IFI16-ASC inflammasome formation. Western blot analysis of the cytoplasmic and nuclear fractions from Si-ASC uninfected and KSHV infected HMVEC-d cells showed efficient knockdown of ASC (Fig 5B, top panel, lanes 1–8). At 24 h p.i. in the Si-control cells, we observed the presence of IFI16 in the cytoplasm which was reduced by >2-fold in the ASC knockdown cells (Fig5B, second panel, lanes 7 and 8). No IFI16 was detected in the uninfected cell cytoplasm (Fig 5B, second panel, lanes 5 and 6). Interestingly, the nuclear IFI16 level was higher in KSHV infected ASC knockdown cells compared to the uninfected cells (Fig 5B, second panel, lanes 1–4). Since KSHV infection does not increase the IFI16 mRNA and protein levels [11], this moderate increase may be due to reduced, or lack of cytoplasmic redistribution of IFI16.
When the cytoplasmic and nuclear fractions were IP-ed with anti-acetylated lysine antibody and western blotted for IFI16, there was no change in the nuclear acetylated IFI16 levels in control and ASC knockdown cells (Fig 5B, third panel, lanes 3 and 4). However, similar to the total IFI16 redistribution, >3-fold reduction in the acetylated IFI16 level was observed in the cytoplasm of ASC knockdown cells (Fig 5B, third panel, lanes 7 and 8).
These results clearly demonstrated that in the absence of ASC, acetylation of IFI16 still takes place which is prior to inflammasome formation. The cytoplasmic redistribution of IFI16 in ASC knockdown cells must be inflammasome independent which might be attributed to cytoplasmic export of acetylated IFI16 either alone or in complex with other proteins. However, the reduced amount of IFI16 in the cytoplasm in comparison to Si-control suggested that the IFI16-ASC inflammasome contributes to the majority of the IFI16 detected in the cytoplasm of infected cells.
As a follow up to C-646 inhibition of KSHV induced p300 catalyzed acetylation of IFI16, we determined the interaction of p300 with IFI16. When HMVEC-d cells infected with KSHV for 24 h in the presence or absence of C-646 were IP-ed for p300 and western blotted for IFI16, we observed increased interaction of IFI16 with p300 which was reduced to basal levels with C-646 treatment (Fig 6A, lanes 1–8).
After detection of a physical association between IFI16 and p300 during KSHV infection, we evaluated the enzymatic activity of p300 and its counterpart HDAC in the cytoplasmic and nuclear fractions of HMVEC-d cells infected with KSHV in the presence or absence of their corresponding inhibitor (1 μM C-646 for p300 and 20 μM TSA for HDAC). KSHV infection (24 h) significantly induced p300 activity in the nucleus but not in the cytoplasm of infected cells compared to uninfected cells, which was inhibited by C-646 (Fig 6B). Similarly, HDAC activity was also induced significantly in the nucleus of infected cells which was inhibited by TSA (Fig 6C). These results suggested that IFI16 acetylation is probably due to increased activity of p300. Increased nuclear p300 activation during infection further supports that acetylation of IFI16 is probably mediated by increased p300 activity in the nucleus and not in the cytoplasm. Decreased activity of enzymes by the inhibitors further verified the specificities of these assays and the functionality of C-646 and TSA (Fig 6B and 6C).
Next, we knocked down p300 to validate our inhibitor studies. Efficient p300 knockdown by Si-p300 with no effect on IFI16 and ASC protein levels was observed (Fig 6D, top three panels, lanes 1–8). The co-IP studies of anti-acetylated lysine antibodies and IFI16 demonstrated the abrogation of IFI16 acetylation in Si-p300 KSHV infected cells while Si-control infected cells showed robust IFI16 acetylation (Fig 6D, fourth panel, lanes 1–8). Similarly, the caspase-1 and IFI16 association was detected in the control group but was abrogated in p300 knockdown infected cells, and caspase-1 was pulled down in all the tested groups (Fig 6D, fourth and fifth panels, lanes 1–8). These results further validated our findings with C-646.
In PLA studies with anti-IFI16 and anti-acetylated antibodies, very few acetylated IFI16 PLA dots were observed in the nucleus of control or p300 knockdown infected cells (Fig 6E, top panels, yellow arrows). In Si-control infected cells (24 h p.i.), a high number of acetylated IFI16 dots were visible in the nucleus and in the cytoplasm (Fig 6E, lower left panel), while only a few dots, as in uninfected cells, were detectable in the p300 knockdown KSHV infected cells (Fig 6E, lower right panel). IFI16 was solely in the nucleus of uninfected cells by total IFI16 PLA (Fig 6F, top panels, yellow arrows). Similar to the acetylated IFI16, total IFI16 was found in both the nucleus and cytoplasm of Si-control infected cells, while p300 knocked down infected cells showed only nuclear IFI16 (Fig 6F, lower panels).
When PLA was performed using anti-IFI16 and anti-ASC antibodies, the red dots representing the IFI16 and ASC association were in both the nucleus and the cytoplasm of Si-control KSHV infected cells (Fig 6G, lower left panel, white and yellow arrows). In contrast, the IFI16 and ASC association was completely abrogated in p300 knockdown infected cells (Fig 6G, lower right panel).
These results further strengthened the finding that acetylation is required for the cytoplasmic redistribution of IFI16 and p300 is responsible for the acetylation of IFI16.
Besides inflammasome induction in KSHV, EBV and HSV-1 infected cells, IFI16 has also been shown to be involved in the induction of IFN-β gene through its cytoplasmic activation of the STING molecule leading into phosphorylation of the transcription factor IRF-3 which subsequently translocates into the nucleus to stimulate the IFN-β gene promoter [21]. KSHV infection induces only a moderate IFN-β response early during de novo infection and early lytic and latent gene products inhibit this response at later times of infection [17], and the role of IFI16 in IFN-β production during KSHV infection is not defined.
When we analyzed the role of IFI16 and its acetylation in IFN-β production, we detected IFN-β in the supernatants of KSHV infected HMVEC-d cells at 6 h p.i., which was significantly reduced by >4 fold by C-646 treatment (Fig 7A). A significant level of phosphorylated IRF-3 detected in the nucleus at 6 h p.i. was reduced in C-646 treated cells (Fig 7B and 7C). Immunoprecipitation with anti-acetylated lysine antibody followed by IFI16 WB revealed the presence of acetylated IFI16 from 30 min to 24 h p.i. in KSHV infected cells, which was abolished by C-646 treatment (Fig 7D, top panel, lanes 1–8). IP reactions with anti-STING antibodies demonstrated the increased IFI16-STING interaction from 30 min to 6 h p.i. and its decrease at 24 h p.i., which was abolished by C-646 (Fig 7D, second panel, lanes 1–8). Similarly, the levels of pIRF-3 increased in untreated KSHV infected cells which were abolished in C-646 treated infected cells (Fig 7D, fourth panel, lanes 1–6).
Next, we knocked down STING in HMVEC-d cells to determine whether IFI16 acetylation is upstream or downstream to STING activation. Efficient knockdown was achieved by electroporation using STING specific Si-RNA (Fig 7E, top panel). KSHV infection was not affected under these conditions as shown by the increased IFI16 acetylation which was not affected by STING knockdown (Fig 7E, second panel) which suggested that IFI16 acetylation is upstream to STING activation. Control tubulin protein was acetylated in uninfected and infected cells, and an equal amount of IFI16 was pulled down in all groups (Fig 7E, fourth panel). IRF-3 was phosphorylated post-KSHV infection which was hampered in STING knockdown cells; however, total IRF-3 was detected in equal amounts in all the groups and results with tubulin showed equal loading (Fig 7E, last three panels). An increased level of IFN-β was observed in the supernatants of HMVEC-d cells infected with KSHV which was significantly reduced by STING knockdown (Fig 7F).
These studies demonstrated that acetylation during KSHV infection induced IFI16 acetylation is required for its cytoplasmic interaction with STING, pIRF-3 induction, and IFN-β production, IFI16 acetylation is upstream to STING activation and STING does not play any role in IFI16 acetylation.
We and others have shown that HSV-1 infection of HFF cells also induced the IFN-β gene and secretion of IFN-β which was dependent upon IFI16 and IRF-3 [16,22]. We utilized C-646 to determine whether IFI16 acetylation has any role in IFN-β production during HSV-1 infection. C-646 did not show any cytotoxic effects on HFF cells nor did it affect the infectivity of HSV-1 (S9A and S9B Fig). At 30 min post HSV-1 infection, 20 and 16 pg/ml of IFN-β was detected in untreated and C-646 treated supernatants, respectively (Fig 8A). At 6 h p.i., 317±16.5 pg/ml of IFN-β was detected in untreated cells whereas significant (>67%; p<0.001) inhibition of IFN-β production was observed in the C-646 treated cells (107±19.4 pg/ml; Fig 8A).
When we examined the phosphorylation of IRF-3 by IFA, compared to the uninfected cells, at 6 h p.i., appreciable levels of phosphorylated IRF-3 were detected in the nucleus and in the cytoplasm (Fig 8B, third panel). In contrast, C-646 treatment prior to infection reduced these levels especially in the nucleus (Fig 8B, fourth panel, and 8C).
In PLA studies, similar to KSHV infected HMVEC-d cells, we observed the cytoplasmic redistribution of acetylated IFI16 in HSV-1 infected HFF cells (6 h p.i.) which was inhibited by C-646 (Fig 8D). IFI16, which was predominantly nuclear in the uninfected HFF cells, was detected in the cytoplasm of HSV-1 infected cells which was abrogated by C-646 (Fig 8E). The observed reduction in the total as well as acetylated IFI16 levels is probably due to the degradation of IFI16 by HSV-1 via its ICPO protein [14].
When the whole cell lysates in the presence or absence of C-646 were IP-ed with anti-acetylated lysine antibody and WB for IFI16 and IRF-3, acetylation of IFI16 was observed as early as 30 min p.i., which was abolished by C-646 (Fig 8F, top panel, lanes 1–6). Acetylation of IRF-3 was not observed (Fig 8F, second panel). In IP-reactions with anti-STING antibodies, increased levels of IFI16 and IRF-3 were detected at 30 min and 6 h p.i. which demonstrated that IFI16 interacts with STING and STING interacts with IRF-3 (Fig 8F, third and fourth panels). These interactions were abrogated by C-646 (Fig 8F, third and fourth panels, lanes 4–6). The level of pIRF-3 increased in untreated HSV-1 infected cells, whereas it was absent in C-646 treated infected cells (Fig 8F, sixth panel, lanes 1–6). As expected, IFI16 levels decreased at 6 h p.i., and in contrast, the IFI16 level was unchanged with C-646 which further suggested that acetylation might be facilitating the stability of IFI16.
Similar to KSHV infected cells, IFI16 acetylation was not affected by STING knockdown during HSV-1 infection (Fig 8G, lanes 1–6). Equal levels of IFI16 was pulled down in both Si-control and in STING knockdown HSV-1 infected cells, and IRF-3 was activated in Si-control HSV-1 infected cells and not in STING knockdown cells (Fig 8G, lanes 1–6). HSV-1 infection induced IFN-β production was hampered in STING knockdown cells (Fig 8H).
Together, these results demonstrated that as in KSHV infected cells, IFI16 acetylation and its translocation to the cytoplasm in HSV-1 infected cells is also critical for its interaction with STING in the cytoplasm, subsequent STING interaction with IRF-3, phosphorylation of IRF-3, and nuclear translocation of pIRF-3 leading into IFN-β production.
Since recognition of KSHV, HSV-1 and EBV genome by IFI16 in the nucleus of infected cells leads to inflammasome activation [11,13,14], we determined whether acetylation of IFI16 is required for its ability to sense the viral genome. Cells were infected with BrdU-KSHV for 6 h in the presence or absence of C-646, IFA was performed for BrdU followed by PLA using anti-IFI16 mouse and rabbit antibodies (Fig 9A). IFI16 was mostly nuclear in the uninfected cells. At 6 h p.i., we observed the appreciable colocalization of IFI16 with KSHV genome in the nucleus of both untreated as well as C-646 treated HMVEC-d cells (Fig 9A, enlarged panels). As before, we observed IFI16 redistribution in the cytoplasm which was absent in C-646 treated cells (Fig 9A). Increased associations of acetylated IFI16 with BrdU-KSHV were observed in untreated cells (Fig 9B, enlarged panels, and white arrows) which were completely abrogated by C-646 (Fig 9B, lower enlarged panel, and 9D). The IFI16-KSHV genome colocalization spots in untreated and C-646 treated cells were similar and the difference was not statistically significant (Fig 9C). Interestingly, the levels of acetylated IFI16 molecules associated with BrdU-KSHV were about 50% less than that of the total IFI16 associated with viral genome (Fig 9C and 9D).
To confirm the direct association of IFI16 with KSHV genome, we performed PLA and chromatin immunoprecipitation (ChIP) assays. To detect the direct binding of IFI16 with KSHV genome, we infected HMVEC-d cells with KSHV with BrdU labeled genome and performed the PLA using anti-BrdU and anti-IFI16 antibodies as this will give signal only when KSHV genome and IFI16 interact and are at close proximity (<40 nm). In the PLA reactions, we observed that the number of IFI16-KSHV genome colocalization spots were similar in both the untreated as well as C-646 treated KSHV infected cells (Fig 9E, white arrows, and 9F). These results further corroborated the Fig 9A results and demonstrated that IFI16 acetylation does not play any role in viral genome recognition. Similar results were also observed in HFF cells infected with BrdU genome labeled HSV-1 (S9C and S9D Fig).
We carried out the ChIP assay of KSHV infected BCBL-1 cells with and without C-646 treatment by pulling down the DNA associated with IFI16 and performed qPCR using primers for two different locations of KSHV and with a control GAPDH primer (Fig 9G). We did not observe any significant changes in the binding of IFI16 with KSHV genome by C-646 treatment (Fig 9G).
These results suggested that a) IFI16 directly associates with KSHV and HSV-1 genomes, b) the acetylation of IFI16 is not required for genome recognition, c) IFI16 acetylation occurs as a dynamic post-genome recognition event, and d) post-acetylation, IFI16 probably moves away from the genome for the formation of its complexes and eventually leading to its cytoplasmic translocation.
IFI16, a member of the ALR family, has emerged as a critical sensor against both nuclear and cytoplasmic DNA with pivotal roles in inflammasome activation and IFN production [11, 21]. However, how the inflammasome formed as a consequence of recognition of herpesviral genomes in the nucleus by IFI16, followed by cytoplasmic accumulation of the IFI16-ASC complex, and how HSV-1 and KSHV genome recognition in the nucleus via IFI16 lead to STING-IRF-3 activation in the cytoplasm and subsequent IFN-β production were not known. Our comprehensive studies for the first time demonstrate that acetylation of IFI16 after recognizing the viral genome occurs as a dynamic post-genome recognition event that is common to the IFI16-mediated innate responses of inflammasome induction and IFN-β production during herpesvirus infections.
Several molecular mechanistic steps of nuclear innate sensing by IFI16 are revealed here (Fig 10). The first step is the recognition of nuclear foreign herpes viral genomes by IFI16 which is independent of acetylation and IFI16 interaction with ASC or STING. This is followed by IFI16’s association with p300 which mediates the acetylation of IFI16. This is a key molecular step common to both of the IFI16 mediated innate responses of inflammasome induction and IFN-β production as IFI16’s acetylation is essential for its interaction with ASC leading into procaspase-1 interaction and activation in the nucleus, interaction with RanGTPase, cytoplasmic translocation and IL-1β induction during KSHV, EBV and HSV-1 infection. Cytoplasmic translocation of acetylated IFI16 is also critical for the activation of STING resulting in the phosphorylation of IRF-3 and IFN-β production (Fig 10).
Crystal structures of overexpressed IFI16 proteins suggest that IFI16 binds to the sugar-phosphate backbone of dsDNA in a non-sequence specific manner with more affinity to superhelix and cruciform DNA [23, 24]. Herpesviral genomes enter the nucleus as a linear, naked dsDNA with nicks and breaks and undergo rapid circularization and chromatinization [25]. Our studies demonstrating that IFI16 recognized the KSHV genome soon after its entry into the nucleus coupled with the fact that this occurs in the absence of acetylation suggests that IFI16 has evolved for rapid recognition of incoming foreign DNA (Fig 10).
Studies with overexpressed proteins suggest that DNA sensing induces filamentous clusters of IFI16 due to homotypic PYD-PYD interactions and cooperative DNA binding that might amplify signals, stabilize IFI16-dsDNA complexes and could act as danger signal [26]. Our studies show such DNA recognition by IFI16 initiates its acetylation process which is essential for the innate immune functions of both inflammasome and interferon responses executed in the cytoplasm and nucleus. Colocalization of reduced levels of acetylated IFI16 with viral genomes (Fig 9) compared to the non-acetylated IFI16 levels suggest that acetylation probably changes the affinity and structure of IFI16 resulting in a dynamic post-genome recognition event of IFI16’s disassociation from the DNA to facilitate its interaction with other proteins and transport into the cytoplasm in a continuous fashion with genomes always occupied with another IFI16 molecule as shown in the KSHV and EBV latently infected cells. This scenario is also supported by observations such as the acetylation of histone prompts its structural extension and charge neutralization resulting in the weakening of DNA-histone interaction [27], and acetylation of KSHV LANA-1 resulting in its dissociation from KSHV genome [28].
Our studies show that acetylation is also critical for IFI16’s transport and interaction with STING, and subsequent IFN-β production in both HSV-1 and KSHV infected cells. Together with our earlier studies demonstrating that ASC is not required for IFN-β production [21, 22] and the absence of IFII6-ASC-procaspase- inflammasome formation and the translocation of acetylated IFI16 in ASC knockdown cells shown here suggested that acetylated IFI16, either alone or in combination with other yet to be identified protein(s), is also relocalized during herpesviral infection resulting in the interaction with STING. Further studies determining whether IFI16 interacts with STING alone or in association with another protein(s) are in progress.
KSHV infection of a PMA stimulated human monocytic THP-1 cell line has been shown to result in IL-1β and IFN-β production by a pathway that is independent of IFI16 [29]. This discrepancy may be due to the fact that KSHV may be undergoing abortive infection in the PMA stimulated cells as has been shown for HSV-1 in these cells [30] and DNA released from the lysosomes is probably recognized by AIM2, c-GAS, and others to stimulate the IL-1β and interferon responses. In contrast, during in vitro infection of permissive cells, viral DNA from the capsid enters the IFI16 rich nucleus resulting in the consequences presented by our studies.
Besides its role in inflammasome and interferon induction, IFI16 is also shown to be a transcriptional modulator of normal cells and the mechanisms are poorly defined [19]. Detection of a basal level of IFI16-p300 interaction and acetylated IFI16 in uninfected cells suggest that they may have roles in other cellular functions such as cell cycle regulation and transcription modulation. Increased IFI16-p300 interaction in infected cells suggests that a dynamic process is initiated; however, why the IFI16-p300 interaction increases in the presence of herpesviral DNA and whether IFI16 recruits p300 directly or via its interaction with other proteins needs to be evaluated further. The p300 HAT assay and HDAC assay performed with nuclear and cytoplasmic fractions of KSHV infected HMVEC-d cells and treatment with their respective specific inhibitors revealed the increased p300 activity in the nucleus and not in the cytoplasm, and thus supporting our conclusion that p300 acetylates the IFI16 in the nucleus after viral genome entry into the nucleus (Fig 3B). Simultaneously, the increased activity of HDAC, further demonstrates that the increase in acetylation was not due to decreased activity of HDACs but due to p300 (Fig 3C).
Our recent studies and others suggested that IFI16 promoted the addition of repressive heterochromatin markers and reduced the active euchromatin markers on HSV-1 gene promoters resulting in the reduced binding of transcription factors and RNA pol II [22]. Whether the regulatory functions of these genes are independent or dependent on the acetylation of IFI16 needs to be determined which is beyond the scope of the present studies.
Increased IFI16 oligomerization in KSHV infected cells due to acetylation suggests that acetylation mediated structural changes in IFI16 probably favors its increased binding with multiple ASC molecules leading into inflammasome assembly. Similarly, ligand mediated NLRC4 phosphorylation has been shown to be crucial for inflammasome activation [31]. In addition, ASC phosphorylation at the CARD domain has been shown to be critical for speck like aggregation and for NLRP3 and AIM2 mediated inflammasomes activation [32]. Though IFI16 has also been shown to be phosphorylated by pUL97 of HCMV that relocalizes IFI16 to the cytoplasm [33], its role in the context of innate immunity has not been evaluated. The Li et al., [15] studies with total cell extracts from human CEM-T lymphoblast-like cells identified six phosphorylation and nine acetylation sites on endogenous IFI16. Which of these sites undergo modifications during the recognition of nuclear viral genomes needs to be examined further and is beyond the scope of the present study. In addition, using uninfected U2OS transfected with DNA, Li et al., [15] demonstrated that acetylation at the NLS motifs of IFI16 results in the cytoplasmic retention of newly synthesized IFI16 by inhibiting nuclear import, and p300 regulated the cytoplasmic IFI16 acetylation during transfection of DNA. As the NLS motif is essential for IFI16 to enter the nucleus, studies with NLS mutants were not possible in our experimental approaches since these mutants IFI16 will stay in the cytoplasm and will not detect the herpes virus genome. Further understanding of IFI16’s modifications will shed additional light on its role in host innate responses as well as its cell cycle and transcription regulatory functions.
In summary, our studies demonstrate that the post-genome recognition event of IFI16’s acetylation by histone acetyltransferase p300 is required for the IFI16-mediated innate immune responses of inflammasome induction, IL-1β and interferon-β production during herpesvirus infections.
HMVEC-d and HFF cells (Clonetics, Walkersville, MD), TIVE, TIVE-LTC, BCBL-1, BJAB-KSHV, Raji, LCL, BJAB, and Ramos cells were grown as described before (11, 12,13, 14). Cells were routinely tested for mycoplasma and only mycoplasma free cells were used for experiments.
Protein A-Sepharose and Protein G-Sepharose CL-4B Fast Flow beads were from GE Healthcare Bio-Sciences Corp., Piscataway, NJ. Cyto Nuclear extract kit was from Active Motif, Carlsbad, CA. Trichostatin A (TSA), Leptomycin B (LPT), and nicotinamide were from Sigma-Aldrich. The CytoTox 96 non-radioactive cytotoxicity kit was from Promega, Madison, WI. SlowFade Gold Antifade reagent with DAPI was from Life Technologies. Verikine human IFN-β ELISA kit was from PBL Assay Science, Piscataway Township, NJ. IL-1β ELISA kit was from RayBiotech, Inc. Norcross, GA. The FLICA 660 Caspase-1 Assay kit was from Immunochemistry Technologies, Bloomington, MN. P300 and HDAC activity assay kits were from BioVision Inc., Milpitas, CA.
Mouse monoclonal anti-IFI16, rabbit polyclonal anti-p300, mouse monoclonal anti-IL-33 and rat polyclonal anti-BrdU antibodies were from Santa Cruz Biotechnology Inc., Santa Cruz, CA. Rabbit anti-BrdU antibody was from Rockland Inc., Gilbertsville, PA. Mouse monoclonal anti-β-actin and tubulin antibodies plus rabbit anti-human IFI16 antibodies were from Sigma-Aldrich. Mouse monoclonal anti-human IL-1β and caspase-1 antibodies were from R&D Systems, Minneapolis, MN, and Invitrogen, Carlsbad, CA, respectively. Goat polyclonal antibody against human ASC was from RayBiotech. Mouse monoclonal antibody against ASC was from MBL International, Woburn, MA. Mouse anti-human TATA binding protein (TBP), rabbit anti-human Ran and mouse anti-IRF-3 antibodies were from Abcam Inc., Cambridge, MA. Rabbit anti-cyclin B1, rabbit monoclonal anti-STING,-p-IRF-3,-histone H2B and H3 antibodies were from CST, Danvers, MA. Anti-rabbit, goat and mouse antibodies linked to horseradish peroxidase, Alexa Fluor-488, -594 and -647 were from KPL Inc., Gaithersburg, MD, or Molecular Probes, Eugene, OR. Anti-Mouse IgG (heavy-chain spcificity)-HRP conjugate goat antibody was from Alpha Diagnostics Intl. Inc. San Antonio, TX. Anti-mouse tagged with IR Dye 680RD secondary antibodies were from LI COR Biotechnology, Lincoln, NE.
Induction of the lytic cycle in BCBL-1 cells by phorbol ester, supernatant collection, and virus purification were described previously [11, 18]. For generating 5-bromo-2-deoxyuridine (BrdU) incorporated KSHV genome, BrdU labeling reagent (Life Technologies) was added to the culture medium in a 1:100 (v:v) ratio (from the supplied stock) [34]. KSHV DNA was extracted, copy numbers quantitated by real-time DNA-PCR, and infection was done with 30 genome copies/cell [11].
HMVEC-d or HFF cells pre-starved for 2 h in the presence or absence of inhibitors such as C-646, leptomycin or cycloheximide, washed, left uninfected or infected with 30 DNA copies/cell in serum free medium for different time points, washed and incubated in complete medium in the presence or absence of inhibitors for different time periods. KSHV positive BCBL-1, BJAB-KSHV and TIVE-LTC cells and uninfected BJAB and TIVE cells were incubated with inhibitors for different time points.
The KOS strain of HSV-1 was produced and titer determined by plaque assay on Vero cells as described [14]. To generate BrdU labelled genome HSV-1, we added BrdU labeling reagent (Life Technologies) to the culture medium at 8 h, 24 h and 48 h post infection in a 1:100 (v:v) ratio (from the supplied stock). HFF cells were starved for 2 h in the presence or absence of C-646, washed and infected with HSV-1 at a multiplicity of infection (MOI) of 1 PFU/cell (~25 genome copies/cell) in serum-free DMEM with or without C-646 for different times, washed with PBS, and incubated in DMEM supplemented with 2% FBS for different time points.
A non-radioactive cytotoxicity assay was performed according to the manufacturer’s protocol (Promega) to evaluate the various inhibitors used in this study. Briefly, HMVEC-d, BCBL-1, TIVE, TIVE-LTC and HFF cells cultured in 12 well plates were incubated with DMSO or varying concentrations of C646 in DMSO for 4 and 24 h in their respective complete media. 100 μl of culture medium of each group was taken carefully and treated with 10 μl of lysis solution (Promega) and incubated for 45–60 min in a 37°C incubator with 5% CO2. Plates were then centrifuged at 1,000 RPM for 3 min and 50 ul samples were transferred carefully into separate 96 well plates with 3 positive control wells of LDH (supplied in the kit). 50 μl of substrate was added to each well of the plate, incubated at RT for 15–30 min, and read at 490 nm using an ELISA reader. The positive control was considered as 100% cytotoxic.
HMVEC-d were starved for 2 h with or without C-646 and either left uninfected or infected with KSHV (30 DNA copies/cell) at 37°C for 2 h. These cells were washed, treated with trypsin-EDTA to remove non-internalized virus, and incubated for varying times of infection [18]. Nuclei were isolated using a Nuclei EZ Prep isolation kit (Sigma) according to the manufacturer’s instructions. Briefly, cells were lysed on ice for 5 min with a mild lysis buffer (Sigma), and nuclei were concentrated by centrifugation at 500xg for 5 min. Cytoskeletal components loosely bound to the nuclei were removed from the nuclear pellet by a repeat of the lysis and centrifugation procedures as described previously [11]. DNA was extracted from isolated nuclei using a DNeasy kit (Qiagen, Germantown, MD). Internalized nuclear KSHV DNA was quantitated by amplification of the ORF73 gene by real-time DNA PCR [2]. The KSHV ORF73 gene cloned in the pGEM-T vector (Promega) was used for the external standard. The CT values were used to generate the standard curve and to calculate the relative copy numbers of viral DNA in the samples.
Total RNAs from KSHV infected or uninfected HMVEC-d cells in the presence or absence of inhibitors were prepared using an RNeasy kit (Qiagen). To quantitate viral gene expression, total RNA was subjected to real-time RT-PCR using ORF73 gene-specific primers and TaqMan probes. A standard curve using the CT values of different dilutions of in vitro-transcribed transcripts was used to calculate relative copy numbers of the transcripts. These values were normalized to those for GAPDH (glyceraldehyde- 3-phosphate dehydrogenase) control reactions. To obtain p values between DMSO, C-646, Leptomycin-B and cycloheximide treated cells, an unpaired Student’s t test was used.
For de novo infection, peripheral blood mononuclear cells (PBMCs) were obtained from the University of Pennsylvania CFAR Immunology Core, and 1X107 PBMCs were either left uninfected or infected by KSHV or EBV as previously described [13]. Briefly, PBMCs were infected with KSHV or EBV in 1 ml of RPMI 1640 medium supplemented with 10% FBS and 5 ng/ml of polybrene (Sigma-Aldrich), incubated for 4 h at 37°C (time point 0), and infected and uninfected cells were centrifuged at 1,200g for 5 min. Cells were washed twice with RPMI medium, resuspended and cultured in six-well plates at 37°C in fresh RPMI medium with 10% FBS. At 24 h p.i., the cells were washed twice with 1X PBS and spotted on slides.
HMVEC-d cells pre-starved for 2 h in the presence or absence of C-646 were washed, uninfected or KSHV infected for 2 h, washed, and incubated with complete medium with or without C-646 for 24 h. BJAB and BCBL-1 cells were treated with C-646 for 24 h. These cells were lysed in HEPES-lysis buffer (100 mM NaCl, 40 mM HEPES [pH 7.5], 05% (v/v) glycerol, 0.1% (v/v) Nonidet P-40 [NP-40] supplemented with PIC). The lysates were cross-linked with 5 mM glutaraldehyde for 10 min, reaction terminated with the addition of 10 μl of 1M Tris-HCL (pH 8.0), samples boiled in SDS buffer, and analyzed by western blot to detect IFI16 oligomerization.
All Si-RNA oligonucleotides for ASC and p300 were from Santa Cruz Biotechnology, Inc. STING Si-RNA (smart pool: Si genome TMEM173, cat. No. M-024333-00-0010) were from GE-Dharmacon (Fisher-Scientific, Pittsburgh, PA) Primary HMVEC-d cells were transfected with Si-RNA using a Neon transfection system (Invitrogen) according to the manufacturer’s instructions. Briefly, subconfluent cells were detached from the culture flasks, washed once with PBS and resuspended in buffer R (Invitrogen) at a density of 1X107 cells/ml. 10 μl of the cell suspension was gently mixed with control Si-RNA or 100 pmol of target specific Si-RNA and then microporated at room temperature using a single pulse of 1,350 V for 30 ms. After microporation, cells were distributed into pre-warmed complete medium and placed at 37°C in a humidified 5% CO2 atmosphere. At 48 h post-transfection, cells were infected with KSHV as described earlier and incubated for 24 h, whole cell lysates using NETN or cytoplasmic/nuclear extract buffers were isolated, knockdown efficiency evaluated by western blotting, and then subjected to co-IP and western blotting.
KSHV infection-induced protein acetylation and other protein-protein interactions were evaluated by co-IP experiments using equal amounts of WCL as well as cytoplasmic and nuclear lysates. The lysates were first incubated for 2 h with 15 μl of Protein A/G sepharose beads and then the pre-cleared lysates incubated for 2 h with immunoprecipitating antibody (anti-acetylated lysine,-IFI16,-ASC,-caspase-1,-p300 antibodies) at 4°C. The immune complexes were captured using 15 μl of Protein A/G-Sepharose beads, washed 4 times with lysis buffer, 3 times in PBS, boiled with SDS-PAGE sample buffer, resolved by 10% SDS-PAGE, and subjected to western blotting.
HMVEC-d cells grown on fibronectin-coated 8 well chamber glass slides for 48 h were serum-starved in the presence or absence of inhibitors for 2 h, washed and then either left uninfected or infected with KSHV (30 DNA copies/cell) for 2 h. Cells were washed with PBS, incubated in complete medium for various time points, washed, fixed in 4% paraformaldehyde for 10 min and permeabilized with 0.2% Triton X-100 for 5 min. BJAB, BCBL-1 and BJAB-KSHV suspension cells treated or untreated with C-646 for 24 h were fixed and permeabilized with pre-chilled acetone. Cells were washed and blocked with Image-iT FX signal enhancer (Invitrogen) for 20 min at RT, and incubated with specific antibodies diluted in 2% BSA for 2 h at 37°C. After washing, cells were incubated with Alexa-Fluor conjugated appropriate secondary antibodies for 1 h at 37°C, washed, mounted in DAPI, imaged with Nikon Eclipse 80i fluorescence microscope and analyzed by Nikon Elements software.
A DuoLink PLA kit from Sigma-Aldrich was used to detect protein–protein interactions as per manufacturer’s protocol. Cells were infected with KSHV (30 DNA copies/cell) or HSV-1 (1 PFU/cell; ~25 genome copies/cell), fixed and permeabilized as described in the IFA section and blocked with DuoLink blocking buffer for 30 min at 37°C. These cells were incubated with target specific primary antibodies diluted in DuoLink dilution buffer. After washing, the cells were incubated for another 1 h at 37°C with species specific PLA probes (PLUS and MINUS) under hybridization conditions and in the presence of 2 additional oligonucleotides to facilitate hybridization of PLA probes if they were in close proximity (<40 nm). A ligation mixture and ligase were then added to join the two hybridized oligonucleotides to form a closed circle. Amplification solution was added to generate a concatemeric product extending from the oligonucleotide arm of the PLA probe. Finally, a detection solution consisting of fluorescently labeled oligonucleotides was added, and the labeled oligonucleotides were hybridized to the concatemeric products. The signal was detected as a distinct fluorescent dot in the Texas red or FITC green channel and analyzed by fluorescence microscopy. Negative controls consisted of samples treated as described but with only secondary antibodies. In some experiments, BrdU staining was performed before PLA to detect viral genome in the infected cells.
BJAB and BCBL-1 cells were subjected to the FLICA 660-YVAD-FMK Caspase-1 assay to detect the active caspase-1 in C-646 treated or untreated cells. The BCBL-1 cells were incubated with C-646 (p300 inhibitor) overnight, washed and FLICA 660-YVAD-FMK caspase-1 detection reagent was applied for 1 h to stain the cells with active caspase-1 in untreated or C-646 treated cells. The cell permeable FLICA 660-YVAD-FMK caspase-1 detection reagent efficiently diffuses into cells and irreversibly binds to activated caspase-1 enzymes. Cells without active caspase-1 have a non-fluorescent status after the wash step. The cells were fixed into the fixation media provided by the manufacturer. The cells were washed to remove the unbound FLICA 660 and subjected to flow cytometry (LSRII, BD Biosciences) at the Flow Cytometry Facility at Rosalind Franklin University of Medicine and Science.
The whole cell protein lysates (WCL) from uninfected and KSHV infected cells were prepared using NETN lysis buffer (100 mM NaCl, 20 mM Tris-HCl [pH 8.0], 0.5 mM EDTA, 0.5% (v/v) Nonidet P-40 [NP-40]) supplemented with 10 μM TSA, 5 mM nicotinamide and protease inhibitor cocktail. Cells were incubated on a rocker at 4°C for 15 min and sonicated at 40 amplitude three times with pulses of 15 seconds on and 10 seconds off. Lysates were clarified by centrifugation for 15 min at 4°C at 15000 x g. The nuclear and cytoplasmic extracts were prepared following the manufacturer’s procedure (Active Motif, Carlsbad, CA). Equal amounts of samples were resolved by 10–20% SDS-PAGE, subjected to western blot, immunoreactive bands developed by enhanced chemiluminescence reaction (NEN Life Sciences Products, Boston, MA), and the bands scanned and quantitated using an AlphaImager (Alpha Innotech Corporation, San Leonardo, CA). The bands were scanned and quantitated using FluorChemFC2 software and an AlphaImager system (Alpha Innotech Corporation, San Leonardo, CA). To detect the tubulin in IP samples, secondary anti-mouse (IRdye 680 labelled) antibody was used and immunoblots were visualized by using the LI-COR Odyssey system
The HFF or HMVEC-d cells in 6 well plates were starved and infected with HSV-1 or KSHV, respectively, for 30 min or 2 h with or without C-646 and incubated for 6 h. Culture supernatants were centrifuged and subjected to ELISA for detection of IFN-β. HMVEC-d cells were starved for 2 h and infected with KSHV-1 with or without C-646 for 2 h, washed and incubated for 24 h and culture supernatant was used to detect IL-1β by ELISA performed as per manufacturer’s instructions. Briefly, the culture supernatants and standards were incubated in the pre-coated wells for 1 h, washed with the washing buffer provided in the kit and probed with IFN-β antibody for 1 h. These wells were washed, incubated with HRP tagged antibodies for 1 h, washed, incubated with substrate (TMB) for 15 min and the reaction was terminated with a stop solution. After 5 min, readings were taken at 450nm and calculations done using a standard curve.
The p300 histone acetyltransferase activity was measured using the p300 HAT fluorometric assay kit from Biovision (Mountain View, CA) as per the manufacturer’s instructions with slight modifications. Briefly, HMVEC-d cells were either left uninfected or infected with KSHV for 24 h in the presence or absence of C-646, and cytoplasmic and nuclear fractions were isolated. From each group 5 μg protein was incubated with the p300 substrate (H3 peptide and acetyl CoA) at 30°C for 30 min in a 96 well plate. The reaction was stopped by adding pre-chilled isopropyl alcohol followed by addition of thiol detecting probe and incubated at room temperature for 15 min. The plate was read for fluorescence at Ex/Em = 392/482 nm in a plate reader (BioTek, Winooski, VT).
The HDAC activity was measured using a fluorometric assay kit from BioVision (Mountain View, CA) as per the manufacturer’s instructions. Protein samples were prepared as in the p300 activity experiment of HMVEC-d cells infected with KSHV for 24 h in the presence or absence of Tricostatin-A (TSA). From each group 10 μg of nuclear and cytoplasmic proteins were mixed in HDAC assay buffer, the fluorometric substrate was added to each well of the 96 well plate and incubated at 37°C for 30 min. The reaction was stopped by adding Lysine developer and mixed well followed by incubation at 37°C for 30 minutes. The samples were analyzed in a fluorescence plate reader (Ex/Em = 350–380/440–460 nm).
To detect the direct association of IFI16 with viral DNA, Chromatin Immunoprecipitation (ChIP) assay was performed as per manufacturer’s instructions. Briefly, untreated BCBL-1 cells (3 x 107) or cells treated with 1μM C-646 for 24 h were fixed with 1% (v/v) methanol-free formaldehyde in fixing buffer for 5 min, crosslinking was quenched using quenching buffer for 5 min at RT, cells were washed twice with cold PBS then incubated in lysis buffer to break the cell membrane. Intact nuclei were collected by centrifugation at 1,700 x g for 5 min at 4°C, resuspended in shearing buffer containing protease inhibitors and chromatin shearing performed on an AFA (Adaptive Focused Acoustics) ultrasonicator (Covaris M220).
Following chromatin shearing, ChIP was performed as described previously (22). Briefly, cellular debris was cleared from the sheared chromatin by centrifugation and the supernatant was incubated overnight at 4°C with 1.5 μg of IFI16 antibody. Samples were incubated in ChIP grade Protein G Magnetic Beads for 2 h at 4°C to collect immune complexes and then washed successively with low salt wash buffer (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris, [pH 8.1]; 150 mM NaCl), then high salt wash buffer (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris, [pH 8.1]; 500 mM NaCl), and then LiCl wash buffer (0.25 M LiCl; 1% NP-40; 1% deoxycholate; 1 mM EDTA; 10 mM Tris, [pH 8.1]). DNA-protein complexes were eluted in 1% SDS prepared in 0.1 M NaHCO3. Crosslinking was reversed by adding 1 μL RNase A and NaCl (0.3 M) and incubating at 65°C for 5 h. Protein was removed by incubating lysate with proteinase K at 55°C for 1 h. Subsequently, DNA was purified using the Wizard SV Genomic DNA Purification System (Promega) and resuspended in nuclease-free water. Real-time PCR was performed with the following KSHV genome (NCBI Reference NC_009333.1) specific primers, Primer Set 1: CAAGGTTAAAGTGGGTTTGCTG, GGTTATTGGCCGTTTCTGTTTC and Primer Set 2: GCGTAATTACTTCCGAGACTGA, TTAACTCCACTTTGCA CCAAAC. As a cellular control, human positive control primer set GAPDH-2 from Active Motif was used. ChIP results are represented as fold enrichment over IgG control.
Results are expressed as means ± SD of at least three independent experiments (n≥3). The p value was calculated using a Student’s T test. In all tests, p<0.05 was considered statistically significant.
|
10.1371/journal.pgen.1004905 | An AP Endonuclease Functions in Active DNA Demethylation and Gene Imprinting in Arabidopsis | Active DNA demethylation in plants occurs through base excision repair, beginning with removal of methylated cytosine by the ROS1/DME subfamily of 5-methylcytosine DNA glycosylases. Active DNA demethylation in animals requires the DNA glycosylase TDG or MBD4, which functions after oxidation or deamination of 5-methylcytosine, respectively. However, little is known about the steps following DNA glycosylase action in the active DNA demethylation pathways in plants and animals. We show here that the Arabidopsis APE1L protein has apurinic/apyrimidinic endonuclease activities and functions downstream of ROS1 and DME. APE1L and ROS1 interact in vitro and co-localize in vivo. Whole genome bisulfite sequencing of ape1l mutant plants revealed widespread alterations in DNA methylation. We show that the ape1l/zdp double mutant displays embryonic lethality. Notably, the ape1l+/−zdp−/− mutant shows a maternal-effect lethality phenotype. APE1L and the DNA phosphatase ZDP are required for FWA and MEA gene imprinting in the endosperm and are important for seed development. Thus, APE1L is a new component of the active DNA demethylation pathway and, together with ZDP, regulates gene imprinting in Arabidopsis.
| DNA cytosine methylation (5-methylcytosine, 5-meC) is an important epigenetic mark, and methylation patterns are coordinately controlled by methylation and demethylation reactions during development and reproduction. In plants, REPRESSOR OF SILENCING (ROS1) is one of the well characterized 5-meC DNA glycosylases that initiate active DNA demethylation by 5-meC excision. Our previous work showed that a 3′-DNA phosphatase, ZDP, functions downstream of ROS1 during active DNA demethylation in Arabidopsis. Here we found that the apurinic/apyrimidinic endonuclease APE1L functions downstream of ROS1 in a ZDP-independent branch of the active DNA demethylation pathway in Arabidopsis. In plants, gene imprinting requires the 5-meC DNA glycosylase Demeter (DME) that has been proposed to initiate a base excision repair pathway for active DNA demethylation in the central cell in female gametophyte. However, besides DME, no other base excision repair enzymes have been found to be important for gene imprinting. Our results show that APE1L and ZDP act jointly downstream of DME to regulate gene imprinting in plants, and suggest that DME-initiated active DNA demethylation in the central cell and endosperm uses both APE- and ZDP-dependent mechanisms.
| DNA methylation is a stable epigenetic mark that regulates numerous aspects of the genome, including transposon silencing and gene expression [1]–[7]. In plants, DNA methylation can occur within CG, CHG, and CHH motifs (H represents A, T, or C). Genome-wide mapping of DNA methylation in Arabidopsis has revealed that methylation in gene bodies is predominantly at CG context whereas methylation in transposon- and other repeat-enriched heterochromatin regions can be within all three motifs [8]. Although the function of abundant CG methylation within genic regions remains unclear, DNA methylation generally correlates with histone modifications that repress transcription activities [1], [9], [10]. DNA methylation patterns are coordinately controlled by methylation and demethylation reactions. In Arabidopsis, symmetric CG and CHG methylation can be maintained by DNA METHYLTRANSFERASE 1 (MET1) and CHROMOMETHYLASE 3 (CMT3), respectively, during DNA replication. In contrast, asymmetric CHH methylation cannot be maintained and is established de novo by DOMAINS REARRANGED METHYLASE 2 (DRM2), which can be targeted to specific sequences by the RNA-directed DNA methylation (RdDM) pathway [1], [10], [11]. DNA methylation is antagonized by an active DNA demethylation pathway that includes the DNA glycosylases REPRESSOR OF SILENCING1 (ROS1), DEMETER (DME), DEMETER-LIKE2 (DML2) and DEMETER-LIKE3 (DML3) [12]–[14].
ROS1, DME, DML2 and DML3 are all bifunctional DNA glycosylases that initiate active DNA demethylation by removing the 5-methylcytosine (5-meC) base and subsequently cleaving the phosphodiester backbone by either β- or β, δ-elimination [12], [14]–[16]. When β, δ-elimination occurs, a gap with a 3′-phosphate group is generated. Our previous work demonstrated that the 3′ DNA phosphatase ZDP catalyzes the conversion of 3′-phosphate group to a 3′-hydroxyl (3′-OH), enabling DNA polymerase and ligase activities to fill in the gap [17]. The β-elimination product, a gap with a blocking 3′-phosphor-α, β-unsaturated aldehyde (3′-PUA), also must be converted to a 3′-OH to allow completion of the demethylation process through single-nucleotide insertion or long patch DNA synthesis by DNA polymerase and ligase [18]. However, the enzymes that may function downstream of ROS1 and DME in the β-elimination pathway have not been identified.
The mutation of ROS1 leads to hypermethylation and transcriptional silencing of a luciferase reporter gene driven by the RD29A promoter, as well as of the endogenous RD29A gene [13]. ROS1 dysfunction also causes DNA hypermethylation in thousands of endogenous genomic regions [19]. zdp mutants also show hypermethylation in the RD29A promoter and many endogenous loci. However, the hypermethylation in the RD29A promoter caused by zdp mutations is not as high as that caused by ros1 mutations, and there are many ROS1 targets that are not hypermethylated in zdp mutants [17]. These observations suggest that there may be an alternative, ZDP-independent branch of the DNA demethylation pathway downstream of ROS1 and other DNA glycosylases/lyases.
Although ROS1 functions in almost all plant tissues [13], DME is preferentially expressed in the central cell of the female gametophyte and is important for the regulation of gene imprinting in the endosperm [20]–[22]. In Arabidopsis, the imprinted protein-coding genes include FWA (Flowering Wageningen), MEA (MEDEA) and FIS2 (Fertilization-Independent Seed 2) and the list is expanding [21]–[25]. The loss-of-function mutation of DME results in aberrant endosperm and embryo development because of DNA hypermethylation and down-regulation of the maternal alleles of imprinted genes [26]. DME is also necessary for DNA demethylation in the companion cells in the male gametophyte [27]–[29]. SSRP1, a chromatin remodeling protein, was identified as another factor required for gene imprinting and the mutation of SSRP1 gives rise to a maternal lethality phenotype similar to that caused by DME mutations [30]. Therefore, it is possible that ZDP and other protein(s) acting downstream of the 5-meC DNA glycosylases/lyases may also affect gene imprinting in Arabidopsis. Intriguingly though, neither ZDP mutants nor mutants in other DNA repair enzymes that may be downstream of DNA glycosylases/lyases show developmental phenotypes associated with defective gene imprinting.
In this study, we characterized the functions of Arabidopsis APE-like proteins in the processing of 3′-blocking ends generated by ROS1 and examined methylome changes induced by ape mutations. We found that purified APE1L can process 3′-PUA termini to generate 3′-OH ends. APE1L also displays a weak activity in converting 3′-phosphate termini to 3′-OH ends. ape1l-1 mutants show altered methylation patterns in thousands of genomic regions. Interestingly, we found that the ape1l+/−zdp−/− mutant is maternally lethal, giving rise to a seed abortion phenotype resembling that of dme mutants. The maternal alleles of the imprinted genes FWA and MEA are hypermethylated, and their expression levels are reduced in the endosperm of such abnormal seeds of the double mutant. Thus, APE1L functions downstream of the ROS1/DME subfamily of DNA glycosylases/lyases in active DNA demethylation and genomic imprinting in Arabidopsis.
The Arabidopsis genome encodes three AP endonuclease-like proteins: APE1L, APE2 and ARP [31]. We purified recombinant full-length APE1L, APE2 and ARP proteins, and found that all three enzymes exhibit AP endonuclease activity in vitro. APE1L, but not APE2 or ARP, also displayed a 3′-phosphatase activity (Fig. 1A). We next wanted to determine if these proteins can process the 3′-PUA termini generated by ROS1 after the β-elimination reaction. We first incubated ROS1 with a 51-mer duplex DNA substrate containing a 5-meC residue at position 29 in the 5′-end labeled strand (Fig. 1B). As expected, the DNA glycosylase/lyase activity of ROS1 generated a mixture of β- and β, δ -elimination products, with either 3′-PUA or 3′-phosphate ends, respectively (Fig. 1B, lane 1). These products were then purified and combined with either APE1L, APE2 or ARP proteins. We found that APE1L efficiently processed the 3′-PUA to generate a 3′-OH terminus. In comparison, 10-fold higher amounts of APE2 or ARP proteins displayed either weak [32] or undetectable (APE2) activity against 3′-PUA ends (Fig. 1B).
To confirm that APE1L is responsible for the detected enzymatic activity we generated an APE1L mutant, N224D. Residue N224 corresponds to N212 of human APE1, which is essential for the enzymatic activity of the mammalian protein [33]. Substitution of N224 by aspartic acid almost completely abolished the activity of APE1L on the 3′-PUA termini (Fig. 1C). The mutation also greatly reduced the AP endonuclease activity on a synthetic AP site and the 3′ phosphatase activity on 3′-phosphate ends. Altogether these results indicate that, in addition to its AP endonuclease activity, APE1L possesses a potent 3′-phosphodiesterase activity that can efficiently process the 3′-PUA blocking ends generated by ROS1.
ROS1 remains bound to its reaction products, which contributes at least partially to the highly distributive behavior of the enzyme in vitro [34]. To determine whether APE1L is able to process 3′-PUA and/or 3′-phosphate termini in the presence of ROS1, we incubated ROS1 and a duplex DNA substrate containing a single 5-meC residue, with WT or N224D APE1L (Fig. 2A). We found that a 3′-OH terminus is efficiently generated in the presence of WT but not mutant APE1L (Fig. 2A). The emergence of the 3′-OH terminus is concomitant with the loss of both 3′-PUA and 3′-phosphate ends, suggesting that the 3′-OH terminus is produced by the 3′-phosphodiesterase activity of APE1L on the 3′-blocking ends generated by ROS1. Quantification of the reaction products revealed that the total amount of strand incision is not increased in the presence of APE1L (Fig. 2B). To assess whether APE1L modulates the DNA glycosylase/lyase activity of ROS1, we performed the reaction in the absence of Mg2+, which is required for APE1L but not ROS1 activity. We found that the enzymatic activity of ROS1 is not increased in the presence of APE1L (S1 Fig.). Thus, APE1L is able to access the 3′-blocked termini generated by ROS1 but does not increase the turnover of this DNA glycosylase. These results suggest that APE1L does not displace ROS1 from DNA.
We next used in vitro pull-down assays to test whether ROS1 and APE1L can physically interact (Fig. 3A). His-tagged ROS1 (His-ROS1) was incubated with either Maltose Binding Protein (MBP) or MBP-APE1L bound to an amylose column. We found that MBP-APE1L, but not MBP, associates with His-ROS1, suggesting that APE1L and ROS1 directly interact in vitro.
To gain insights into the transfer of DNA demethylation intermediates between ROS1 and APE1L, we performed electrophoretic mobility shift assays with a gapped DNA substrate (Fig. 3B–C). MBP-APE1L alone is not able to form a stable complex with the substrate, judging by the smeared band next to the position of the free probe (Fig. 3B, lanes 2 and 1). A mobility shift was observed when the DNA substrate was incubated with His-ROS1, consistent with complex formation (Fig. 3B, lane 3). As we have previously reported [17], part of the labeled probe remained trapped in the wells, hinting at the formation of insoluble His-ROS1-DNA complexes. Next, we incubated the gapped DNA substrate and His-ROS1 with increasing concentrations of MBP-APE1L to assess complex formation. With increasing MBP-APE1L, the band corresponding to the ROS1-DNA complex and the labeled material in the well gradually disappeared, concomitant with the appearance of a discrete, new band (Fig. 3B, lanes 4–7). Importantly, this band was only detected when both ROS1 and APE1L were present in the binding reaction. These results suggest that ROS1, APE1L, and gapped DNA form a ternary complex and that ROS1 is required for APE1L to stably associate with the DNA substrate.
To further examine complex formation we performed supershift experiments using antibodies against MBP-APE1L and His-ROS1 (Fig. 3C). We found that adding anti-MBP to a binding reaction containing MBP-APE1L, His-ROS1 and DNA generated an additional shift, thus confirming the presence of APE1L in the complex. However, a supershift was not observed in the presence of the anti-His antibody. We reasoned that access to the His epitope on His-ROS1 might be restricted in the complex. Therefore, as an alternative approach, we compared the mobility shifts generated from binding reactions containing the DNA gapped substrate, MBP-APE1L and either His-ROS1 or MPB-ROS1 (Fig. 3D). We found that MBP-ROS1, which has a higher molecular weight than His-ROS1, gave rise to a higher molecular weight gel shift, thus confirming that ROS1 is also present in the complex. The most likely interpretation for these results is that ROS1, APE1L, and the gapped DNA substrate form a ternary complex. To further confirm the interaction between APE1L and ROS1, we performed a firefly luciferase complementation imaging assay [35] in tobacco leaves. We found that APE1L can interact with ROS1 in the tobacco leaves (Fig. 3E).
Our previous data show that ZDP, a component of the active DNA demethylation pathway, co-localizes with ROS1 in subnuclear foci [17]. To determine the subnuclear localization of APE1L protein, we generated antibodies specific to APE1L and used them for immunolocalization of APE1L in Arabidopsis leaf nuclei. As shown in Fig. 4A, APE1L is broadly distributed throughout the nucleus. In 62% of the cells examined, APE1L is enriched in the nucleolus whereas in 38% of the cells, APE1L localizes to small nucleoplasmic foci. Only very weak signals were observed when the antibodies were applied to nuclei preparations of ape1l-1 mutant plants, indicating that the staining patterns in wild type plants reflect APE1L localization rather than non-specific binding of the antibody (Fig. 4A).
To test whether APE1L co-localizes with ROS1 or ZDP, we performed co-immunofluorescence. In our experiments, FLAG-tagged ROS1 was expressed from its native promoter in ros1-1 mutants and visualized with anti-FLAG antibodies. We observed APE1L co-localization with ROS1 within both nucleoplasmic foci and also the nucleolus in about 10% of cells, as shown by the strong yellow signals (Fig. 4B). In 54% of the cells, APE1L co-localizes with ROS1 in the nucleolus but not in nucleoplasmic foci, whereas in 36% of the cells, APE1L and ROS1 do not substantially co-localize (Fig. 4B). APE1L and ZDP also co-localize in nucleoplasmic foci in approximately 28% of cells (Fig. 4C). Thus, APE1L co-localizes with components of the DNA demethylase machinery in distinct subnuclear structures in a subset of cells.
To evaluate possible roles of APE1L in active DNA demethylation initiated by the ROS1 subfamily of DNA glycosylases/lyases, two T-DNA insertion lines were isolated for APE1L (S2A Fig.). RT-PCR analysis with APE1L-specific primers corresponding to the full-length open reading frame of the gene detected the expected product in wild-type plants in both the Ws and Col backgrounds, but not in ape1l-1, which is in Ws. In contrast, ape1l-2 shows almost the same expression level as wild-type plants (S2B Fig.). Since the endonuclease ARP shows weak activity against the 3′-PUA blocking ends generated by ROS1 in vitro, we also isolated two T-DNA insertion lines for ARP (S2A Fig.) and confirmed by RT-PCR that they have a complete loss of mRNA expression (S2B Fig.). One of the mutants, arp-1, was used for further experiments.
To examine the general DNA methylation status in the ape1l-1 and arp-1 mutants, we compared the susceptibility of 5S rDNA and 180-bp centromeric repeat regions to the restriction enzymes HpaII and MspI. These enzymes recognize the same site (CCGG), but HpaII cleavage is methylation-inhibited whereas methylation does not affect cleavage by MspI. DNA cleavage was assessed by Southern analysis. Similar to the zdp-1 and ros1-4 mutations, the ape1l-1 or arp-1 mutation does not affect the DNA methylation levels at the 5S rDNA or 180-bp centromeric repeats (S3 Fig.), suggesting that the ape1l-1 and arp-1 mutants do not have changes in their overall DNA methylation patterns.
We performed whole genome bisulfite sequencing using DNA from 14-day-old ape1l-1, arp-1, zdp-1 and their corresponding wild-type control plants. The CG methylation levels in wild type (Col-0) and zdp-1 mutant are similar, but the CHG and CHH levels are mildly elevated in zdp-1 (S4A Fig.). For ape1l-1, its overall genome methylation level in CG, CHG and CHH contexts is slightly higher than that in Ws (S4B Fig.). In total, we identified 6389 DMRs (differentially methylated regions) in ape1l-1 mutant plants, including 3497 hyper-DMRs that have a significant increase in methylation and 2892 hypo-DMRs that have a significant reduction in methylation (S4C Fig.; S3 Table). In contrast, arp-1 only affects methylation levels at 403 genomic regions, including 162 hyper-DMRs and 241 hypo-DMRs (S4C Fig.). 1559 hyper-DMRs and 612 hypo-DMRs were identified from zdp-1 (S4C Fig.; S4 Table). The hyper-DMRs and hypo-DMRs identified in ape1l-1, zdp-1 and ros1-4 are evenly distributed along the five chromosomes (S4D Fig.). To determine whether APE1L and ZDP mutations affected DNA demethylation in specific genomic regions, we analyzed intergenic regions, transposable elements (TEs) outside of genes, TEs overlapping with genes and genic regions. Unlike zdp-1, ros-1 mutants and ros1-3;dml2-1;dml3-1 (rdd) triple mutants, which have less than 43% of hypermethylated (hyper-) DMRs distributed in gene regions, in ape1l-1 and arp-1 more than 60% of the hyper-DMRs are distributed in gene regions (S4E Fig.). In contrast, the percentages of hyper-DMRs distributed in TEs in ape1l-1 and arp-1 are lower than those in zdp-1, ros-1 and rdd mutants (S4E Fig.). These data indicate that the APE1L and ARP mutations preferentially impact DNA demethylation of gene regions while the ZDP and ROS1 mutations have a greater impact on TE regions. The distribution patterns of classified hypo-DMRs are different from those of hyper-DMRs. The percentages of hypo-DMRs in gene regions are higher than 70% in rdd, zdp-1 and ape1l-1. The arp-1 has a low percentage of hypo-DMRs in gene regions but high percentage of hypo-DMRs in intergenic regions (S4E Fig.). The ape1l-1 mutation affects CHG and CHH demethylation more profoundly than CG demethylation, both in gene regions or in TEs (S5A Fig.). We also examined the effect of APE1L mutation on TEs of different lengths and found that the ape1l-1 mutation has a bigger impact on shorter genes but longer TEs (S5B and S5C Figs.). Unlike ape1l-1, zdp-1 shows almost the same DNA methylation pattern for both gene regions and TEs (S5 Fig.).
Compared to the high level of overlap (70.9%) between zdp-1 and rdd hyper-DMRs, less than 50% of the hyper-DMRs in ape1l overlap with those in rdd (S4C and S5D–S5E Figs.). One reason for this relatively low level of overlap may be the difference in genetic backgrounds; the ape1l-1 mutant is in the Ws background whereas the other mutants are in Col. When the hyper-DMRs in ros1-1 (C24 background) and ros1-4 (Col-0 background) were compared, the overlap was also quite low (52%). For the hyper-DMRs, the level of overlap between ape1l-1 and zdp-1 is also very low (14%) (S5F Fig.) even though some loci do show hypermethylation in ape1l-1 as well as zdp-1 (S5G–S5J Fig.). These results are consistent with the notion that APE1L and ZDP largely represent two different mechanisms (AP endonuclease vs 3′-phosphatase) downstream of the DNA glycosylases/lyases, despite their redundant functions (as 3′-phosphatases).
ROS1 and ZDP mRNA levels are decreased in RdDM pathway mutants [17]. We examined the expression of APE1L and ARP in the RdDM mutants nrpd1-3 and nrpe1-11, and found no substantial decreases in the mRNA levels in the mutants compared to Col (S6A Fig.). We also measured the expression levels of ROS1 and ZDP in ape1l-1 and arp-1 mutants, and found that the expression levels are similar in the mutants compared to those in the Col or Ws wild type control plants (S6B Fig.). Also, unlike the zdp-1 mutant, which is hypersensitive to MMS induced DNA damage, the ape1l-1 and arp-1 mutants show a sensitivity level similar to that of wild-type plants (S7A Fig.).
To study the potential genetic interactions between APE1L and ZDP, we crossed ape1l-1 and zdp-1 mutant plants. Interestingly, we found that ape1l+/−zdp−/− and ape1l−/−zdp+/− plants produce many aborted seeds, suggesting that the double mutations of APE1L and ZDP are lethal (Fig. 5A). We grew the viable seeds, genotyped the seedlings, and found no ape1l−/−zdp−/− plants (Table 1). The ratio of aborted seeds is 48.7% in self-pollinated ape1l+/−zdp−/− plants and 26.5% in self-pollinated ape1l−/−zdp+/− plants (S5 Table). Approximately seven days after pollination, the seeds fated to abortion show white color and plump phenotypes (S8A Fig.). The endosperm in those seeds fails to undergo cellularization and the growth of their embryos is arrested (S8B Fig.). Later, those seeds accumulate brown pigments and collapse.
The 48.7% seed abortion ratio of self-pollinated ape1l+/−zdp−/− plants suggests that the lethality of this mutant may be maternally regulated. Also, because APE1L and ZDP may act downstream of ROS1 and DME and some of the characteristics of seed abortion in ape1l+/−zdp−/− and ape1l−/−zdp+/− mutants resemble those in the dme+/− mutant, we examined whether double mutations of APE1L and ZDP, like dme mutation, are also maternally lethal. If so, all seeds derived from a female gametophyte with APE1L and ZDP double mutations will abort irrespective of the paternal allele. We crossed ape1l+/−zdp−/− (♀) with ape1l+/+zdp−/− (♂) and the cross resulted in about 50% aborted seeds and 50% viable seeds. When we crossed them in a reverse direction, we observed 100% viable seeds (Fig. 5B and S5 Table). Furthermore, when we crossed ape1l+/−zdp−/− (♀) plants with wild type plants, approximately 50% of the seeds aborted. These data indicate that ape1+/−zdp−/− mutant is indeed maternally lethal (Fig. 5B and S5 Table). However, the ape1l−/−zdp+/− mutant is not maternally lethal, based on the fact that few seeds aborted when we crossed ape1l−/−zdp+/− to Col or ape1l−/−zdp+/+ in either directions (Fig. 5B and S5 Table). This is consistent with its seed abortion ratio (26.5%) (S5 Table) and segregation ratio (ape1l−/−zdp+/+∶ape1l−/−zdp+/−∶ape1l−/−zdp−/− = 0.98∶2∶0) (Table 1) when self pollinated.
We examined the morphology of aborting seeds from ape1l+/−zdp−/− and ape1l−/−zdp+/− mutants using differential interference contrast microscopy. The major defects of aborting seeds are arrested embryo growth at the heart stage or earlier (S8B Fig.) and abnormal sizes of endosperm nuclei (S8C Fig.). In some aborting seeds, the embryos are invisible, indicating that the embryos are arrested very early in development. The aborting seeds of ape1l+/−zdp−/− and dme+/− both display arrested embryo growth. Unlike ape1l+/−zdp−/− mutant seeds, dme+/− mutant seeds display clumps of unknown structures but there were no aberrant endosperm nuclei (S8B–C Fig.).
We noticed that the ape1l+/−zdp−/− mutant has abnormal segregation ratio (4.07∶1∶0), which does not fit the expected segregation ratio of maternally lethal plant (1∶1∶0). Alexander staining and in vitro germination assay were carried out to examine the pollen development in different mutants. The ape1l+/−zdp−/− mutant showed defects in pollen development and germination (S9 Fig.), suggesting that the ape1l+/−zdp−/− mutation not only leads to maternal lethality but also gives rise to paternal defects.
Maternal lethality phenotypes can be caused by aberrant expression of maternally imprinted genes and defects in the central cell or the endosperm [12], [26], [30]. FWA and FIS2 are two well-studied maternally imprinted genes, and their maternal expression in the endosperm relies on active DNA demethylation initiated by DME [12], [23]. We investigated whether the methylation of the FWA and FIS2 promoters in endosperm tissues is affected by APE1L and ZDP double mutations (Fig. 6A). The ape1l+/−zdp−/− plants were backcrossed to zdp−/− plants three times to minimize the Ws background. To examine the methylation levels of DME target genes in our mutants, we employed the method of Buzas et al. [36] where the DNA methylation specific restriction enzyme McrBC is used to digest DNA before doing q-PCR in seeds at 3 days post manual pollination. We found that after digestion with McrBC, the amount of DNA recovered from FWA and FIS2 promoter regions (where is methylated in wild type leaf) was reduced in both dme and ape1l−/−zdp−/− mutants compared with wild type, but there was no difference in the unmethylated FWA gene body region (Fig. 6A). These results indicate that the ape1l−/−zdp−/− endosperm has hypermethylation in FWA and FIS2 promoter regions.
In order to measure the mRNA levels of FWA and MEA in Col and ape1l−/−zdp−/− endosperms, we carried out real-time PCR and found that the expression levels of FWA and MEA but not the DME and FIE mRNAs are down-regulated in the ape1l−/−zdp−/− mutant endosperm (Fig. 6B). To confirm and further analyze the FWA expression change, we introduced a pFWA::ΔFWA-GFP reporter into the ape1l+/−zdp−/− and ape1l−/−zdp+/− mutants by crossing the mutants with a transgenic line expressing the reporter [23]. Both ape1l+/−zdp−/− and ape1l−/−zdp+/− plants produce about 50% seeds defective in pFWA::ΔFWA-GFP expression (Fig. 6C–6D and S6 Table). To our surprise, ape1l−/−zdp+/−mutant also produced 50% GFP-off seeds even though it is not maternally lethal and it produces about 75% viable seeds (Table 1). It turns out that hypermethylation of pFWA::ΔFWA-GFP promoter and silencing of FWA-GFP can occur in mutants which do not show maternal lethality. In addition, it seems that GFP-off seeds can be viable, so 75% viable seeds may be comprised of 50% GFP-on seeds and 25% GFP-off seeds. Taken together, our data suggest that DNA hypermethylation and down-regulation of imprinted genes occur and may be the cause of defects in the ape1l−/−zdp−/− endosperm.
Active DNA demethylation in plants is initiated by the ROS1 subfamily of 5-meC DNA glycosylases/lyases and presumably completed through a base excision repair pathway [2], [37]. Previous work has reported that the 3′-phosphatase ZDP and the scaffold DNA repair protein XRCC1 also function in active DNA demethylation in Arabidopsis [17], [38]. AP endonucleases are known to catalyze post-excision events during base excision repair. Our study here demonstrates that APE1L, one of the Arabidopsis AP endonucleases, functions in active DNA demethylation by processing β-elimination products of the bifunctional 5-meC DNA glycosylases/lyases and generating a 3′-OH group. APE1L-mediated reaction comprises a new branch of the DNA demethylation pathway downstream of ROS1, DME, DML2 and DML3 (Fig. 7). Our biochemical data show that APE1L has an additional, weak 3′-phosphatase activity, and thus may also function in the other branch, perhaps redundantly with ZDP, to process β, δ-elimination products. Interestingly, it has been recently reported that the wheat homolog of APE1L also possesses 3′-phosphatase and 3′-phosphodiesterase activities [39]. Our results suggest that APE1L not only functions downstream of ROS1, DML2 and DML3 in vegetative tissues to prevent DNA hypermethylation but also functions together with ZDP downstream of DME to control DNA demethylation and gene imprinting in the central cell and endosperm and is thus important for seed development.
Active DNA demethylation in mammals can be initiated through the deamination of 5meC by AID to generate thymine, or the oxidation of 5meC to generate 5-hydroxymethylcytosine (5hmC), and further to 5-formylcytosine (5fC) and 5-carboxycytosine (5caC) by the TET family of DNA dioxygenases [2], [40]–[42]. 5fC and 5caC can be excised by the monofunctional DNA glycosylase TDG, whereas thymine can be removed by the monofunctional DNA glycosylase MBD4. Thus, a base excision repair pathway is required for completing the DNA cleavage and cytosine insertion steps during active DNA demethylation in mammals. Little is known about the DNA repair factors involved in active DNA demethylation in mammals, but it is likely that mammalian APE functions in active DNA demethylation downstream of the DNA glycosylases.
The ape1l-1 mutation leads to DNA hypermethylation in thousands of genomic regions, indicating that APE1L is required for DNA demethylation in these regions in Arabidopsis. Like mutations in 5-methylcytosine DNA glycosylases/lyases such as ROS1, mutations in DNA repair enzymes downstream of these enzymes are expected to preclude active DNA demethylation and cause hypermethylation. Coordinating the DNA glycosylase/lyase and repair activity would be predicted to prevent an otherwise fatal accumulation of strand breaks throughout the genome [17]. APE1L and ROS1 physically interact in vitro and co-localize in vivo, strongly suggesting that these proteins form a complex which coordinates their activities. One may ask why the DNA demethylation pathway includes both lyase activity of ROS1 and AP endonuclease activity of APE1L. In a recent study, it was reported that Wheat APE1L has weak endonuclease activity but robust 3′-repair phosphodiesterase and 3′-phosphatase activities [43]. Even though we detected the endonuclease activity of Arabidopsis APE1L in vitro, it is possible that, like Wheat APE1L, Arabidopsis APE1L is weak in cleaving DNA backbone at AP sites when involved in DNA demethylation. In this case, the lysase activity of ROS1 is required for generating the DNA gap.
The Arabidopsis ARP endonuclease also processes the 3′-PUA generated by ROS1 in vitro, although this activity is much weaker than APE1L. Our whole genome bisulfite sequencing data identified only a small number of DMRs in the arp-1 mutant. Therefore, ARP is unlikely to play a major role in DNA demethylation, at least under normal growth conditions.
Interestingly, we detected many genomic regions that are hypomethylated in the ape1l-1 mutant. APE1L is a multifunctional enzyme; its APE and 3′-phosphatase activities may contribute to other DNA repair pathways in addition to active DNA demethylation, Thus, APE1L dysfunction may affect many DNA-related processes that directly or indirectly cause DNA hypomethylation. Compared to the ros1-4 and rdd mutations, ape1l and arp mutations induce higher percentages of hypermethylation in genic regions, whereas the zdp mutation induces a higher percentage of hypermethylation in TEs. The mechanisms underlying this genomic specificity are unclear, but it is possible that APE1L and ARP function redundantly in the demethylation of TEs, such that mutating either one individually does not cause hypermethylation.
Unlike ZDP, which processes 3′-phosphate blocking ends and promotes the release of ROS1 from its products, APE1L converts both 3′-phosphate and 3′-PUA to 3′-OH, but does not increase the turnover of ROS1. Although both ZDP and APE1L interact with ROS1 in vitro and co-localize with ROS1 in vivo, ZDP and APE1L do not show extensive co-localization. It is possible that ZDP and APE1L exist mostly in two different protein complexes (Fig. 7). ZDP dysfunction caused DNA hypermethylation and transcriptional silencing of a luciferase reporter driven by the RD29A promoter, although the mutant phenotype was less severe than ros1 mutants [17]. We hypothesized that at some DNA demethylation target regions, such as the RD29A promoter, the DNA glycosylases/lyases may use both β- and β, δ-elimination activities and thus require both APE and ZDP to process the intermediates and prevent transcriptional silencing. However, we found that the ape1l-1 and arp mutations did not affect expression of the reporter gene (S7B Fig.). It is possible that APE1L may function redundantly with ARP and/or ZDP in demethylation of the RD29A promoter. zdp mutant showed sensitivity to MMS but ape1l and arp mutants are not sensitive to MMS probably because they carry out different reactions. In addition, APE1L, APE2 and ARP may play redundant roles in repairing MMS-induced DNA damage, such that the single mutation or double mutations are not sufficient to induce sensitivity to MMS.
The choice between the APE branch and the ZDP branch of the active DNA demethylation pathway depends on the elimination mechanism used by the DNA glycosylases/lyase enzymes. It is unclear when and where a DNA glycosylases/lyase employs β-elimination, β, δ-elimination, or both. Knowing which genomic regions depend on APE1L and which depend on ZDP for demethylation would be helpful. However, because the zdp-1 and ape1l-1 mutants are in different ecotypes, it is not ideal to compare the genomic regions targeted by the two different branches of the demethylation pathway.
The double mutations of APE1L and APE2 are embryonic lethal, but not paternally or maternally lethal based on our results and the segregation ratio of selfed ape1l+/−ape2−/− reported previously [31]. It is possible that the lethal phenotype caused by APE1L and APE2 double mutations reflect deficiencies of DNA repair. Interestingly, we found that the ape1l+/−zdp−/− mutant shows a maternal lethality phenotype, which has been shown to occur in other mutants that are defective in DNA demethylation, such as the dme and ssrp1 mutants [26], [30]. Unexpectedly, only the ape1l+/−zdp−/− mutant shows maternal lethality but the ape1l−/−zdp+/− mutant is not maternally lethal. As a result of maternal lethality, about 50% of seeds abort in dme+/− and ape1l+/−zdp−/− mutants. In contrast, about 25% seeds abort in the ape1l−/−zdp+/− mutant. All of the aborting seeds display embryos arrested at early growth stages presumably because an abnormal endosperm cannot support normal growth of the embryo. The morphology of aborted seeds in the ape1l+/−zdp−/− and ape1l−/−zdp+/− mutants is almost the same as that in the ape1l+/−ape2−/− mutant, which is not maternally lethal and gives about 25% aborted seeds [31]. It is likely that the ape1l+/−ape2−/− mutant is also defective in DNA demethylation. Alternatively, this type of morphology (arrested embryo and aberrant endosperm) may reflect deficiencies of base excision repair.
It is likely that APE1L and ZDP function downstream of DME in the active DNA demethylation pathway that controls seed development. However, the aborting seeds in ape1l+/−zdp−/− mutants have varied sizes of endosperm nuclei but the aborting seeds in dme+/− mutants have endosperm nuclei of uniform sizes. This phenotypic difference may arise because APE1L and ZDP have multiple functions in DNA demethylation and repair, whereas DME only participates in DNA demethylation. As in dme+/− mutants, the seed abortion phenotype in ape1l+/−zdp−/− mutants is associated with the hypermethylation of the FWA promoter and the MEA ISR, and reduced FWA and MEA expression. Similar to the ape1l+/−zdp−/− mutant, the ape1l−/−zdp+/− mutant also produces about 50% GFP-off seeds, suggesting that these two types of mutants are similarly defective in DNA demethylation of imprinted genes. The phenotype of the ape1l−/−zdp+/− mutant in pFWA-GFP silencing (50% GFP-off) and seeds viability (25% aborted and 75% viable) resembles that of the recently discovered atdre2 mutant [44]. Some other factors beyond DNA demethylation or some dosage effects must be differentially involved in different types of mutants, leading to maternally lethality in some mutants but not in others, even though they are all defective in the expression of imprinted genes. In summary, our results show that APE1L and ZDP are important regulators of gene imprinting in plants, and suggest that DME-initiated active DNA demethylation in the central cell and endosperm employs both APE- and ZDP-dependent mechanisms.
Full-length APE1L and APE2 cDNAs were subloned into pMAL-c2X (New England Biolabs) to generate MBP-APE1L and MBP-APE2 fusion proteins. The full-length ARP cDNA was subcloned into pET28a (Novagen) to generate a His-ARP fusion protein. Expression was induced in Escherichia coli BL21 (DE3) dcm− Codon Plus cells (Stratagene). MBP-APE1L and MBP-APE2 were purified by amylose affinity chromatography (New England Biolabs) and His-ARP was purified by affinity chromatography on a Ni2+-nitrilotriacetic acid column (Amersham Biosciences). His-ROS1 and MBP-ROS1 were expressed and purified as previously described [15], [34].
Site directed mutagenesis of APE1L was performed using the Quick-Change II XL kit (Stratagene) according to the manufacturer's instructions. The N224D mutation was introduced into pMal-APE1L by using the oligonucleotides APE1LN212D_F4 and APE1LN212D_R4 (see S1 Table). The mutant sequence was confirmed by DNA sequencing, and the construct was used to transform E. coli strain BL21 (DE3) dcm− Codon Plus cells (Stratagene). Mutant protein was expressed and purified as described above for APE1L.
Oligonucleotides used to prepare DNA substrates (see S2 Table) were synthesized by Integrated DNA Technologies [45] and purified by PAGE before use. Double-stranded DNA substrates were prepared by mixing a 5 µM solution of a 5′-fluorescein-labeled or 5′-Alexa Fluor-labeled oligonucleotide (upper strand) with a 10 µM solution of an unlabeled oligomer (lower strand). For preparation of 1-nt gapped DNA, a 5 µM solution of the corresponding 5′-labelled oligonucleotide was mixed with 10 µM solutions of unlabelled 5′-phosphorylated oligonucleotides P30_51 and CGR. Annealing reactions were performed at 95°C for 5 min, followed by slow cooling to room temperature.
To detect 5-meC DNA glycosylase/lyase activity, purified His-ROS1 (35 nM) was incubated at 30°C for 4 h with a Alexa Fluor-labeled DNA duplex (20 nM), containing a single 5-meC, in a reaction mixture containing 50 mM Tris–HCl pH 8.0, 1 mM DTT, 0.1 mg/ml BSA. In reactions containing APE1L, the mixture also included 200 mM NaCl and 1 mM MgCl2. Reactions were stopped by adding 20 mM EDTA, 0.6% sodium dodecyl sulfate, and 0.5 mg/ml proteinase K, and the mixtures were incubated at 37°C for 30 min. DNA was extracted with phenol/chloroform/isoamyl alcohol (25∶24∶1) and ethanol precipitated at −20°C in the presence of 0.3 mM NaCl and 16 µg/ml glycogen. When the ROS1 reaction products were used as purified substrates for AP endonucleases (see below), samples were resuspended in 5 µl of distilled water. Otherwise, they were resuspended in 10 µl of 90% formamide, heated at 95°C for 5 min, and separated in a 12% denaturing polyacrylamide gel containing 7 M urea. Alexa Fluor-labeled DNA was visualized using the blue fluorescence mode of the FLA-5100 imager and analyzed using Multigauge software (Fujifilm).
The AP endonuclease activity was detected using a DNA substrate containing a synthetic AP site (tetrahydrofuran, THF) opposite G. The 3′-phosphatase activity was assayed on a 1-nt gapped substrate containing 3′-phosphate and 5′-phosphate ends. The 3′-phosphodiesterase activity was tested on purified ROS1 products, which contain a mixture of fragments with 3′-PUA and 3′-phosphate termini. In all assays, purified AP endonucleases were incubated with DNA substrates (20 or 40 nM) at 30°C for the indicated times in a reaction mixture containing 50 mM Tris–HCl pH 8.0, 200 mM NaCl, 1 mM DTT, 0.1 mg/ml BSA and 1 mM MgCl2. Reactions were stopped and products analyzed as indicated above.
Purified MBP alone or MBP-APE1L (200 pmol) in 100 µl of Column Buffer (20 mM Tris, pH 7.4, 1 mM EDTA, 1 mM DTT, 0.5% Triton X-100) was added to 100 µl of amylose resin (New England Biolabs) and incubated for 1 h at 4°C. The resin was washed twice with 600 µl of Binding Buffer (10 mM Tris, pH 8.0, 1 mM DTT, 0.01 mg/ml BSA). Purified His-ROS1 (15 pmol) was incubated at 25°C for 1 h with either MBP or MBP-APE1L bound to resin. The resin was washed twice with Binding Buffer. Bound proteins were analyzed by Western blot using antibodies against His6 tag (Novagen).
EMSAs were performed using an Alexa Fluor-labeled duplex containing a gap flanked by 3′-phosphate and 5′-phosphate termini prepared as described above. The labeled duplex substrate (10 nM) was incubated with MBP-APE1L and/or His-ROS1 at the indicated concentrations in DNA-binding reaction mixtures (10 µl) containing 10 mM Tris HCl, pH 8.0, 1 mM DTT, 10 µg/ml BSA. After 15 min incubation at 25°C, reactions were immediately loaded onto 0.2% agarose gels in 1× Tris acetate/EDTA. Electrophoresis was carried out in 1× Tris acetate/EDTA for 40 min at 80 V at room temperature. Alexa Fluor-labeled DNA was visualized in a FLA-5100 imager and analyzed using MultiGauge software (Fujifilm).
To investigate the interaction between APE1L and ROS1, two constructs was generated: APE1L-Cluc and ROS1-Nluc. The BamHI and SalI sites were used for cloning APE1L genomic DNA into pCAMBIA1300-CLUC vector. ROS1 was introduced to NLUC by In-Fusion HD Cloning Kit (Clontech). For protein interaction analysis, two combinatory constructs were transformed simultaneously into Nicotiana benthamiana leaves. To prevent the silencing of those genes, a virus p19 protein gene containing construct was transformed at the same time. After 3 d, 1 mM luciferin was sprayed onto the lower epidermis and kept in the dark for 5 min, then a CCD camera (1300B; Roper) was used to capture the fluorescence signal at 21°C.
Two T-DNA insertion mutants of the APE1L gene (At3g48425), INRA Flag240B06 and Salk_024194C, were used and they were referred to as ape1l-1 and ape1l-2 respectively. T-DNA insertions are present in the fifth exon and fourth intron of ARP in arp-1 (SALK_021478) and arp-2 (SAIL_866_H10) respectively. For all plants, seeds were sown on 1/2 MS plates containing 2% sucrose and 0.7% agar, stratified for 48 hours at 4°C and grown under long day conditions at 22°C. They were collected at 14 days or transplanted to soil.
Immunofluorescence was performed in 2- to 3-week-old leaves as described by Pontes et al., [46]. Nuclei preparations were incubated overnight at room temperature with rabbit anti-APE1L (anti-APE1L antibodies were generated by injecting rabbits with a recombinant full length APE1L protein that was purified by affinity chromatography), anti-ZDP [17] and mouse anti-Flag (F3165, Sigma). Primary antibodies were visualized using mouse Alexa 488-conjugated and rabbit Alexa-594 secondary antibody at 1∶200 dilution (Molecular Probes) for 2 h at 37°C. DNA was counterstained using DAPI in Prolong Gold (Invitrogen). Nuclei were examined with a Nikon Eclipse E800i epifluorescence microscope equipped with a Photometrics Coolsnap ES.
Total RNA was extracted from 2-week-old seedlings using the RNeasy Plant Mini Kit (QIAGEN). 2-µg RNA was used for the first-strand cDNA synthesis with the Super script III First-Strand Synthesis System (Invitrogen) for RT-PCR following the manufacturer's instructions. The cDNA synthesis reaction was then diluted five times, and 1 µl was used as template in a 20-µl PCR reaction with iQ SYBR Green Supermix (Bio-Rad). All reactions were carried out on the iQ5 Multicolour Real-Time PCR Detection System (Bio-Rad). The comparative threshold cycle (Ct) method was used for determining relative transcript levels (Bulletin 5279, Real-Time PCR Applications Guide, Bio-Rad), with TUB8 as an internal control.
DNA was extracted from 2 g of 12-day-old seedlings grown in a growth chamber and sent to BGI (Shenzhen, China) for bisulfite treatment, library preparation, and sequencing.
Images of seed phenotypes were captured using an Olympus SZX7 microscope equipped with a Canon Powershot A640 camera. For cleared whole-mount observation, immature seeds, that are 8 days after pollination, were cleared using chloral hydrate, glycerol, and water (8 g: 1 ml: 2 ml) and photographed using a Leica DM6000 B differential interference contrast microscope equipped with a Leica DFC 425 camera. Fluorescence was detected with an Olympus BX53 fluorescence microscope equipped with an Olympus DP80 digital camera.
The McrBC assay was performed according to Buzas et al [36]. Briefly, wild type and the apel1+/−zdp−/− mutant were pollinated with Ler pollen. 3 days after pollination, pools of GFP-on and GFP-off seeds were selected under a dissecting fluorescence microscope and more than 300 seeds were used for DNA extraction. Genomic DNA concentration was measured by Nanodrop. Approximately 1 µg of DNA was digested with 1 µL of McrBC overnight at 37°C. After digestion, DNA methylation levels at the specific loci were determined by real-time PCR using absolute quantification against a 1∶1 mixture of genomic DNA extracted from Col-0 and Ler leaves. Primers are listed in S1 Table.
Female ape1l+/−zdp−/− plants (Col-0) were crossed with male wild type plants. The endosperm plus seed coat fraction was collected for RNA purification using the Trizol method. DNAase treatment and LiCl precipitation were applied to remove DNA and polysaccharide contaminations, respectively. RNA was reverse transcribed into cDNA by the SuperScript III First-Strand Synthesis System (Invitrogen) with an oligo dT primer. Real-time PCR analysis was performed using SYBR Premix Ex Taq (TaKaRa) and CFX96 real-time system (Bio-Rad). ACT11 was used as the internal control.
We used whole-genome bisulfite sequencing to analyze the methylomes of Ws, ape1l-1, arp-1 and zdp-1 mutant plants. The data set was deposited at NCBI (GSE52983).
|
10.1371/journal.pgen.1006597 | Bovine and murine models highlight novel roles for SLC25A46 in mitochondrial dynamics and metabolism, with implications for human and animal health | Neuropathies are neurodegenerative diseases affecting humans and other mammals. Many genetic causes have been identified so far, including mutations of genes encoding proteins involved in mitochondrial dynamics. Recently, the “Turning calves syndrome”, a novel sensorimotor polyneuropathy was described in the French Rouge-des-Prés cattle breed. In the present study, we determined that this hereditary disease resulted from a single nucleotide substitution in SLC25A46, a gene encoding a protein of the mitochondrial carrier family. This mutation caused an apparent damaging amino-acid substitution. To better understand the function of this protein, we knocked out the Slc25a46 gene in a mouse model. This alteration affected not only the nervous system but also altered general metabolism, resulting in premature mortality. Based on optic microscopy examination, electron microscopy and on biochemical, metabolic and proteomic analyses, we showed that the Slc25a46 disruption caused a fusion/fission imbalance and an abnormal mitochondrial architecture that disturbed mitochondrial metabolism. These data extended the range of phenotypes associated with Slc25a46 dysfunction. Moreover, this Slc25a46 knock-out mouse model should be useful to further elucidate the role of SLC25A46 in mitochondrial dynamics.
| Mitochondria are essential organelles, the site of numerous biochemical reactions, with a critical role in delivering energy to cells, particularly in the nervous system. Consequently, disrupted mitochondrial function often results in neurodegenerative diseases, in humans and in other mammals. Herein, we determined that the “Turning calves syndrome”, a new hereditary sensorimotor polyneuropathy in the French Rouge-des-Prés cattle breed was due to a single substitution in SLC25A46, a gene encoding a protein of the mitochondrial carrier family. We created a mouse knock-out model and determined that disruption of this gene dramatically disturbed mitochondrial dynamics in various organs that resulted in altered metabolism and early death, indirectly confirming the gene identification in cattle. Moreover, our novel findings extended the range of phenotypes associated with polymorphisms of this gene and help to elucidate the role of SLC25A46 in mitochondrial function.
| Mitochondria are eukaryotic organelles with a wide range of functions. In addition to delivery of energy to cells via oxidative phosphorylation (OXPHOS), they are involved in various other bioenergetic reactions, including Krebs cycle, β-oxidation of fatty acids and heme biosynthesis. Furthermore, they have roles in calcium signaling, stress response and apoptosis [1–3]. Consequently, they are a vital organelle. Not surprisingly, mitochondrial dysfunction is shown to be responsible for an increasing number of diseases, inherited or not [2,4].
To enable a variety of cells to respond to variable physiological conditions, particularly to adapt to varying energy demands, mitochondrial morphology is highly dynamic, with three main mechanisms: fusion, fission and cristae remodeling [5–7]. The balance between fusion and fission is particularly critical to regulate mitochondrial shape, size and number. In mammals, mitochondrial morphology is regulated by the following GTPase proteins: DRP1 (Dynamin related protein 1) for fission, mitofusin MFN1 and MFN2, and OPA1 (Optic atrophy 1) for fusion. All these proteins are essential for development [8–10] and despite ubiquitous expression, their mutations primarily cause neurological diseases, as is common for proteins involved in mitochondrial dynamics [11–13], probably due to neurons being energy-intensive cells [14].
Fusion proteins, for example, are involved in diverse syndromes. Dominant mutations of OPA1 cause Autosomal Dominant Optic Atrophy (ADOA), affecting mitochondrial morphology (aggregated and fragmented) and content (reduced content of mitochondrial DNA (mtDNA) and reduced ATP production) [15–18]. Mutations in MFN2 cause Charcot-Marie-Tooth type 2A (CMT2A) disease in humans, a sensorimotor axonopathy with aggregated swollen mitochondria and altered structural integrity of inner and outer mitochondrial membranes [19,20]. Mutations of orthologous genes cause neurodegenerative diseases in other mammals, with for example, different mutations of MFN2 causing respectively an early axonopathy in Tyrolean Grey breed [21] and fetal-onset neuroaxonal dystrophy in dog [22].
Recently, human patients with combined ADOA and CMT2 phenotypes were identified as having recessive mutations in SLC25A46 [23]. This gene encodes a protein belonging to the mitochondrial carrier transporter family [24], anchored on the outer mitochondrial membrane [23]. 53 proteins belong to this family. Most of them are responsible for the transport of a quantity of diverse metabolites across the inner mitochondrial membrane, which are necessary for all the metabolic pathways taking place in mitochondria [25–27] However, the observed phenotypes linked to SLC25A46 dysfunction suggested that SLC25A46 is rather involved in mitochondrial dynamics, and particularly may act as a pro-fission factor [23].
In cattle, due to massive inbreeding and bottlenecks effects in each selected breed, recessive mutations are likely to be transmitted to a large proportion of the population, leading to emergences of hereditary diseases [28].
In the late 2000’s, such an outbreak was described in the French Rouge-des-Prés breed with a new sensorimotor polyneuropathy named “Syndrome des veaux tourneurs” (“Turning calves syndrome”) because of a propensity of the affected calves to turn around themselves before falling down [29]. This neurodegenerative disease is characterized by an early onset of ataxia, especially of hindlimbs, and paraparesia affecting young calves (2–6 weeks old). Despite symptomatic care, nervous symptoms progress over the next months, leading to repetitive falls and ultimately resulting in permanent recumbency and inevitably euthanasia. Degenerative lesions involve both the general proprioceptive sensory and upper motor neuron motor systems [29].
The number of cases in this breed has rapidly increased in a few years (based on statistics from the French National Observatory for Bovine Genetic Diseases), prompting a genetic study to identify the causal mutation. We identified herein by homozygosity mapping the 3Mb haplotype associated to this disease on bovine chromosome 7.
Further examination of this genetic interval allowed us to determine that this disease resulted from a single nucleotide polymorphism in the coding region of the SLC25A46 gene, leading to an apparently damaging amino acid substitution. The eradication of the “Turning calves syndrome” was undertaken, through the selection of non-carrier males so the number of reported affected calves rapidly dropped to zero.
Therefore, a novel mouse knockout model of Slc25a46 was produced to elucidate the function of the encoded protein. The resulting phenotype described below included nervous symptoms but had more widespread effects, including alterations in mitochondrial dynamics and metabolism that caused premature death, thus extending the range of phenotypes associated with polymorphisms of this gene.
Calves from the Rouge-des-Prés breed presenting an ataxic gait and paraparesis of hindlimbs as described in [29] were examined by a veterinarian, and diagnosis was confirmed by histopathology. Pedigree analysis of 11 of them confirmed the autosomal recessive determinism of the “Turning calves syndrome” and the involvement of a predominant founder ancestor (S1 Fig). This bull, born in 1973, was a historical sire of the Rouge-des-Prés breed (contributing 6% of the breed). Genotyping of 12 affected calves followed by homozygosity mapping identified a single 3.1 Mb homozygous interval at the telomeric end of bovine chromosome 7 (S1 Table). This information was used to design a genetic indirect test, based on the haplotype associated to the disease, allowing to begin the selection against the “Turning calves syndrome” of the Rouge-des-Prés breed.
To identify the causative mutation, whole-genome sequencing was performed on two affected cattle, one heterozygous carrier and one wild-type (WT). The detected polymorphisms (SNP and small indels) were filtered in several steps. First, the genotype/phenotype correlation had to be perfect, i.e. affected cattle had to be homozygous for the polymorphism, and the WT and carrier cattle had to be homozygous or heterozygous, respectively, for the WT allele. Second, since this mutation is supposedly specific to the Rouge-des-Prés breed, with relatively recent emergence, polymorphisms were discarded if they were already present in the dbSNP database and/or in the Illumina SNP chip. Finally, polymorphisms were filtered according to their predicted effects on transcript and/or protein, based on the hypothesis that this mutation is very deleterious (Table 1).
The two remaining putative causal SNPs were a single substitution in exon 15 of MAN2A1 gene and a single substitution in exon 4 of SLC25A46 gene (Table 2).
These two polymorphisms were further tested (Sanger sequencing and Taqman assay) on an extended DNA multibreed panel, including 93 Rouge-des-Prés cattle, and 321 samples from 12 French cattle breeds. The MAN2A1 polymorphism was discarded because the genotype/phenotype association was not always found, and because it was present in other breeds. The SLC25A46 polymorphism had a perfect genotype/phenotype association. All the affected animals were homozygous for this mutation.
All the proteins from the mitochondrial carrier family share a common structure with three tandemly repeated homologous domains about 100 amino acids long. Each domain contains two transmembrane alpha-helices forming a funnel-shaped cavity allowing the binding and the transport of the substrate from the intermembrane space to the matrix by a conformational transition [26,27,30].
The C/T SLC25A46 substitution leads to replacement of an arginine by a cysteine, in the first transmembrane helix of the protein (Fig 1A–1C). This amino acid is highly conserved throughout evolution in SLC25A46 proteins (Fig 1D). When compared to the other mitochondrial carriers, as described in [30], it appears that this amino-acid is not conserved across this family. The most frequent amino-acid at this position is a threonine, and is shared by only 19 of the 53 mitochondrial carriers. This suggests a role more related to the specific biological function(s) of SLC25A46. Based on SIFT software [31], this substitution was expected to affect the function of SLC25A46.
The mutated protein was expressed normally in brain and liver of affected animals and was present in mitochondrial-enriched protein extracts, consistent with a typical mitochondrial localization (Fig 1E).
Affected calves have characteristic degenerative microscopic lesions in the central nervous system (CNS) and peripheral nervous system (PNS), both in grey matter (brain stem lateral vestibular nuclei and spinal cord thoracic nuclei) and white matter (dorsolateral and ventromedial funiculi of the spinal cord), in addition to demyelination in certain peripheral nerves [29]. Electron microscopy confirmed this neuropathy phenotype, with discrete lesions of demyelination and a few enlarged nodes of Ranvier (Fig 1F).
As mentioned above, selection against this disease in the affected breed was undertaken for obvious economic reasons as soon as the genetic test was commercially available. Thus, affected animals were rapidly unavailable, limiting the range of phenotypic investigations that could be performed to analysis of previously collected tissue samples. To better characterize the function of SLC25A46, construction of mouse models was initiated.
SLC25A46 mouse models were constructed, using TALEN (Translation Activator-Like Effector Nuclease) technology, by targeting mouse exon 3, the exon homologous to the one mutated in the bovine gene. Following microinjection of the TALEN mRNA and screening of the resulting mice, two transgenic lines were established in a pure FVB/N genetic background: 1) Tg26 line with a 75 bp DNA deletion inducing exon 3 aberrant splicing and resulting in a truncated protein of 159 amino acids; and 2) Tg18 line with a 15 bp insertion / 3 bp deletion, causing replacement of 2 amino acids from the first transmembrane domain by six modified amino-acids (Fig 2A, S2A Fig). Heterozygous mice were viable and appeared as fit as their WT counterparts (they were monitored for at least 12 months). Transmission of the mutated allele followed Mendelian inheritance.
In both lines, Slc25a46 was undetectable by western blot analysis, on both total protein and on protein extracts enriched for mitochondrial proteins (Fig 2B and 2C). However, in Tg18 line, Slc25a46 mRNA levels were unchanged in homozygous mutant animals, except in peripheral nerves (S2B Fig). It suggests that a repression of the translation of the Slc25a46 mRNA occurred in Tg18 mice and/or more likely that the mutated protein was not properly associated with the mitochondrial membrane and consequently was rapidly degraded. A degradation mechanism must also occur in Tg26 mice for the putatively translated truncated protein, in addition to a noticeable reduction in amount of Slc25a46 mRNA, probably due to mRNA decay (S2C Fig). Thus, homozygous mutants from both lines were regarded as functional knock-outs and will now be referred to as Tg-/- mice.
At birth, Tg-/- pups from the two lines were indistinguishable from each other and from their WT and heterozygous littermates, despite reported expression of the Slc25a46 gene early during mouse embryogenesis in various EST databases. However, their growth was reduced compared to the WT pups from the end of the 1st week of life, and from the 2nd week, they stopped gaining weight (Fig 3A and 3B). The observed reduced growth rate started despite a normal feeding behavior during the first weeks, while the pups were still nursed by their mother, as evidenced by the presence of milk in their stomach (Fig 3C) and by a normal behavior in the cage (i.e. all the pups were regularly seen under their mother, and none of them was left alone in the cage). Yet, at 3 weeks of age, intestinal tracts of Tg-/-mice were less filled than their WT counterparts, with reduced feces (Fig 3D), consistent with their cachectic state. Furthermore, there were intestinal hemorrhages in the oldest Tg-/- animals (Fig 3E).
Intestinal length and diameter were smaller in Tg-/- than in WT mice. However, histological staining did not reveal any obvious change in the intestine from the Tg-/- mice which could explain their reduced growth (S3A Fig).
An ataxic gait was apparent from the 2nd week of life, especially on the hind limbs (Fig 3F and 3G). Tg-/- mice walked on the tip of their toe, instead of putting the whole foot sole on the soil (Fig 3F and 3G). This was evocative of a proprioception defect, as it is described in the “Turning calves syndrome” [29]. However, it did not evolve to permanent recumbency, perhaps due to the short lifespan of the Tg-/- mice (see below). Moderate hyperreflexia was also evidenced on hindlimbs when pinching the mice’s toes. An epileptic-like phenotype was also noticed from the 2nd week of life (S1 Movie).
All Tg-/- mice died between the 3rd and 4th weeks of life, either spontaneously in the cage, or by euthanasia for evident ethic reasons (Fig 3H).
Post-mortem examination revealed that several other organs were affected in Tg-/-mice. Thymus and spleen were significantly smaller relatively to the body mass (Fig 3I and 3J). This was expected as they are described as metabolic state sensors, with rapid atrophy in case of malnutrition [32]. Liver was also significantly smaller (Fig 3I and 3J), and biochemical blood analyses showing increased biliary acids, bilirubin and cholesterol in Tg-/- mice were indicative of a cholestasis, and consistent with a stress of the liver (Table 3). Liver histology was nonetheless almost normal (S3B Fig). Muscle damage was also suspected, based on a general decrease of muscle mass combined with increased creatine kinase and aspartate amino transferase (Table 3), but muscle histopathology was also unchanged in Tg-/- mice (S3C Fig).
Biochemical analysis revealed a highly-disturbed metabolism in Tg-/- mice, confirming the general alteration of their state (Table 3). Severe hypoglycemia was noted, which may be linked to the observed growth defect. Low plasma iron concentrations combined with high ferritin were indicative of defective iron metabolism and/or storage.
Therefore, the phenotype of the Tg-/- mice was distinctly different from that of the above-mentioned bovine sensorimotor polyneuropathy, presenting a wider range of symptoms.
Since Tg-/- mice displayed symptoms evocative of proprioception and motor involvement, investigations were then undertaken on the nervous system of Tg-/- mice. However, no major defect of the CNS was detected in Tg-/- mice (based on HES and Luxol blue staining), with only minimal lesions consisting of rare vacuolated neurons in the lateral vestibular nuclei (Fig 4A). Peripheral nerves lacked visible degenerative lesions, although the presence of macrophages containing lipid debris suggested a possible degenerative process (Fig 4B). Axon diameters and myelin sheath thickness was comparable in both genotypes (Fig 4C). Thus, even if it was not possible to exclude a peripheral neuropathy in Tg-/- mice, the fast evolution of the disease up to death may limit it to an early very mild form.
Furthermore, the study of the optic nerve could not highlight any difference between WT and Tg-/- pups (Fig 4D), nor degenerative lesions in the Tg-/- axons, an observation recalling the lack of reported vision defect in the “Turning calves” [29] but contrasting with consistency of this phenotype in recently reported human cases [23,33,34].
However, axons from CNS and PNS had abnormal round, small and aggregated mitochondria as evidenced in myelinated and non-myelinated fibers (Fig 5A and 5B), indicating a fusion/fission imbalance, an observation also noticed in tissues from affected “Turning calves” (Fig 5C). Moreover, most mitochondria in Tg-/- mice had abnormalities of their internal architecture, namely abnormal membranes and cristae. These abnormal mitochondria were also detected in the enteric nervous system of Tg-/- mice, in intestinal Auerbach plexus cells (Fig 5D).
In accordance with ubiquitous expression of Slc25a46 in mice (S2B and S2C Fig), there were abnormal mitochondria in other organs from Tg-/- mice, indicating a generalized mitochondrial defect. Skeletal muscles also had numerous aggregated mitochondria, although their morphology generally remained normal (Fig 5E), as well as muscular layers in the intestinal tract. Hepatocytes had numerous and smaller mitochondria with vesicular cristae, rarely attached to the inner mitochondrial membrane (Fig 5F–5H).
Since mitochondrial internal architecture and morphology were altered in several tissues, we searched for effects on mitochondrial metabolism. Regarding activity of each respiratory chain complex in mice, there were significant decreases for complexes I, III, and IV in brain and muscle from Tg-/- mice (Fig 6A). However, there was an opposite trend in liver from Tg-/- mice, with a significant increase for complexes III and IV activities. This particular response could be explained by the specific effect of physiological stresses on the mitochondrial metabolism in various tissues and specifically the liver [35,36].
Krebs cycle enzymes, localized in the mitochondrial matrix, generally had no change in activities, except aconitase which was increased in muscles from Tg-/- mice, and fumarase which was increased in liver (Fig 6B).
Proper fusion/fission equilibrium is necessary to maintain a homogeneous and healthy population of mitochondria [37]. For example, several missense mutations of MFN2 causing autosomal dominant optic atrophy ‘plus’ phenotype induce a respiratory chain defect and mtDNA deletions and eventually mtDNA depletion in muscle cells [38,39]. Moreover, loss of mtDNA is also found in Ugo1p depleted cells, Ugo1p being SLC25A46’s homolog in yeast [40]. However, there was no significant mtDNA depletion or large deletion in liver, muscle or brain from homozygous mutant mice (Fig 6C).
Comparative MS-MS analysis was conducted in brain protein extracts after enrichment of mitochondrial proteins, in order to detect changes in protein expression induced by disruption of Slc25a46 in mouse. Amongst the detected proteins, only 26 were significantly up- or downregulated (Table 4). Interestingly, five downregulated proteins belonged to the Hsp70 (70 kDa Heat-shock protein) family (Grp78, Hs71l, Hs71b, Hsp7c, Hs74), as well as three others for which the p-value nearly reached the significance threshold (Hsp72, Grp75, Hs90b). Such observations might suggest a role, direct or indirect, of Slc25a46 in the mitochondrial-Endoplasmic Reticulum (ER) contact sites (see below). Alterations were also noted in mitochondrial membrane proteins associated with glucose transport (Hk1, Hk1-sb), and fatty acid metabolism (Gpdm, Echa). Notably, hemoglobin subunits were significantly upregulated (Hba, Hbb1, Hbb2). This upregulation may be linked to an iron dysregulation, as evidenced by biochemical analyses on Tg-/- mice.
While this manuscript was first submitted, a paper was published, with evidence of interaction between SLC25A46 and fusion proteins MFN2 and OPA1, and MIC60 and MIC19 proteins belonging to the MICOS complex [34]. Notably MS-MS results did not show any significant change of expression for these proteins, their level was then monitored by western blot on brain extracts (S4A Fig). Tg-/- mice did not display significant expression level for these proteins. Thus the knock-out of Slc25a46 in mouse does not lead to a reduction of Mic60 and a potential disruption of the MICOS complex, contrary to the fibroblasts treated with siRNA, as described in [34]. Moreover, it is not compensated by a change in the expression of fusion proteins Opa1 and Mfn2. Expression of OPA1, MFN2 and MIC60 was also monitored by Western blot on bovine brain and liver protein extracts, but we could not infer a significant change in the expression of these proteins, especially because the number of biological samples was very low (S4B Fig).
In the present study, we provided reliable evidence that the “Turning calves syndrome”, a recessive sensorimotor polyneuropathy reported in the French Rouge-des-Prés breed in the late 2000’s, was caused by a point mutation in SLC25A46 gene. The single amino acid substitution did not affect protein expression nor its proper location within the mitochondria (based on western blot). However, it affected a highly conserved amino acid, in the first transmembrane helix of the protein. Amongst the mitochondrial carrier proteins, 14 are known to be associated to rare metabolic diseases [27,41]. Mutations are mostly located in functional domains of the proteins, including the substrate binding sites and the matrix and cytosolic gates (which respectively open/close the carrier to the mitochondrial matrix and towards the cytosol) [41] Interestingly, even the mutated amino-acid is not conserved amongst the mitochondrial carrier family, it is located just in the matrix gate area [30], which is known to be critical for the conformational change.
Electron microscopy confirmed axonal lesions in affected cattle and identified abnormal round and aggregated mitochondria in axons. This phenotype is reminiscent of mutations in mitochondrial fusion proteins such as MFN2 in CTM2A disease [20], consistent with an fusion/fission imbalance. However, SLC25A46 function in fusion or fission remains elusive. Ugo1p, which is SLC25A46 homolog in yeast, plays a crucial role in fusion, in close interaction with Fzo1p and Mgm1p (MFN1/2 and OPA1 homologs, respectively) [40,42,43]. Ugo1p mutants had fragmented mitochondria, and a loss of mtDNA [40]. In humans, a pro-fission role of SLC25A46 was proposed, due to an increase of mitochondrial branching in fibroblasts derived from a patient carrying a homozygous missense mutation in the carrier domain of the protein [23]. In contrast, there was another report of a SLC25A46 mutation in a splice site, leading to a truncated transcript and perhaps to a knock-out [33]. In this case, the mitochondrial network was fragmented, suggesting a fusion role for SLC25A46.
To better understand SLC25A46 function, and because the selection against “Turning calves syndrome” made new biological material collection difficult in cattle, mouse knock-out models were constructed. In Tg-/- mice, nervous degenerative phenotypes (ataxic gait and epilepsy) were apparent from the 2nd week of life. Furthermore, Tg-/- mice also had pronounced weight loss and metabolic defects leading to premature death around weaning. Although histopathology did not account for this drastic phenotype, electron microscopy implicated involvement of mitochondria in several tissues. There was a fusion/fission imbalance (similar to cattle), with numerous round aggregated mitochondria in the central and peripheral nervous systems, including the enteric nervous system. Abnormal mitochondria were present in Auerbach plexus cells. This may have contributed to dysmotility of the intestinal tract, and the subsequent observed weight loss, at least partially, as often described for multi-systemic mitochondrial diseases such as Mitochondrial Neurogastrointestinal Encephalopathy Syndrome. [44] These small and numerous mitochondria were also detected in muscle and liver. Clearly, effects of disruption of SLC25A46 were not restricted to the nervous system, consistent with ubiquitous expression of the gene.
Mitochondria regulate their shape in accordance with the metabolic state of the cell. In case of starvation, mitochondrial length is increased, by phosphorylation of the fission protein Drp1, leading to decreased fission [45] and/or by oligomerization of fusion protein OPA1 [46]. In mitochondrial fusion-incompetent cells, mitochondria cannot fuse and are degraded, leading to cell death. Tg-/- mice which experience weight loss, are in a metabolic state mimicking starvation. Consequently, the absence of elongated mitochondria suggests an impaired fusion in these animals.
In addition to the fusion/fission imbalance, mitochondria from Tg-/- mice had disturbed internal architecture, with distorted and vesicular-like cristae, and cristae less frequently attached to the membrane. Cristae morphology is maintained and regulated mainly by OPA1 and by the MICOS complex [1,47]. This complex is composed of six subunits in yeast, with all of them inserted in the inner mitochondrial membrane [48]. Mutations in genes encoding these subunits result in an altered internal architecture, i.e. loss of cristae junctions, and cristae organized as membrane stacks [48–50]. MIC60, also known as Mitofilin, is one of the key players of the MICOS complex. In yeast, the MIC60 homolog Fcj1p interacts with SLC25A46 homolog Ugo1p, forming close contact sites between outer and inner mitochondrial membranes [51]. The interaction between MIC60 and SLC25A46 has been recently documented in human [23,34]. In the report from Janer et al., the absence of SLC25A46 resulted in a marked decrease in the steady-state level of MIC60 in studied human fibroblasts [34]. Based on abnormalities of mitochondrial architecture detected in Tg-/- mice, we inferred that Slc25a46 (potentially in association with Mic60), may contribute to establishment of a proper contact between outer and inner mitochondrial membranes in mammals. However, Mic60 was only marginally downregulated in Tg-/- mice (MS/MS analysis), with the p-value nearly reaching the threshold of significance, and this downregulation could not be observed by Western blot analysis.
Furthermore, MS/MS analysis did not highlight any downregulation of fusion factors interacting with Slc25a46, such as OPA1 and MFN2, nor did specific analysis of these proteins by Western blotting. These differences suggest either cell-type (fibroblast vs brain cells) and/or species’ specificities.
Since cristae contain OXPHOS subunits (i.e. respiratory complexes I to V), disorganization of cristae often decreases activity of OXPHOS subunits [49,52,53] and disturbs assembly of respiratory supercomplexes, with profound reduction in respiration efficiency [54]. Mitochondrial metabolism is indeed affected in Tg-/- mice, with a marked decrease in complexes I, III, IV activities in brain and muscle, and an increase in complexes III and IV activities in liver. This discrepancy is not unlikely, given the specificity of each tissue and each cell type in the response to physiological stresses [35,36] or to mutations [55,56].
Our proteomic analysis highlighted a potential interaction between Slc25a46 and Hsp70 proteins; eight of the latter were down-regulated in Tg-/- mice. These chaperone proteins, participate in the protein folding [57–59] and in the protein import across the outer mitochondrial membrane [60,61] in close interaction with Tom70, which is also significantly downregulated. Thus, in Tg-/- mice, importation of proteins may be downregulated, either by a direct interaction between Slc25a46 and the import machinery (interactions between Slc25a46 and Hsp90, Grp75 and Grp78 were recently evidenced by immunoprecipitation [23]), or by a general alteration of the outer mitochondrial membrane structure. Interestingly, Grp78 also known as BiP, which is one of the most significantly downregulated protein in the Tg-/- mice, is considered as a major regulator of the ER, due to its multiple roles in the ER function [62], and is shown to act at the ER-mitochondria interface under stress conditions [63,64]. The recent observation that SLC25A46 interacts with the Endoplasmic Reticulum Membrane Complex (EMC) and may participate to the regulation of the phospholipid flux between ER and mitochondria appears to support the pivotal role of SLC25A46 between ER and mitochondria [34].
Alternatively, since all these Hsp70 proteins function under the dependence of ATP, the affected mitochondrial metabolism may be insufficient to provide enough ATP, which could downregulate Hsp70 protein expression.
Collectively, there was good evidence for a pivotal function of SLC25A46 between the outer and inner mitochondrial membranes. Disruption of Slc25a46 in mouse not only affected the subtle equilibrium between fusion and fission, but also disturbed the internal architecture and the link with a pool of Hsp70 chaperone proteins and potentially the mitochondrial-ER trafficking. Consequently, mitochondrial and general metabolisms were severely impacted, leading to premature death. This model seemed similar to an affected infant that died seven days after birth [33].
In contrast, in cattle affected by the “Turning calves syndrome”, the mutated protein was still present and we inferred that it retained a portion of its activity, as in humans carrying homozygous missense mutations. According to the localization of the mutations (in cytosolic, transmembrane or inter mitochondrial membrane domains), interactions with various proteins could be disturbed, affecting only a part of SLC25A46’s functions.
Furthermore, alteration of SLC25A46’s functions might also result in species-specific phenotype, as all human cases reported so far suffer from optic atrophy, which is not observed in the bovine [29] and mouse models reported here (Table 5). However, it should be mentioned that FVB/N mice carry two mutations which result in severe vision impairment: a mutation in the tyrosinase gene (TyrC) causing an albino phenotype and the retinal degeneration mutation (Pde6brd/rd11) [65]. Consequently, FVB/N mice suffer from early onset retinal degeneration and blindness around weaning [66] which might interfere with the observation of the phenotype.
Overall, our data in both models provided a basis for the wide range of human phenotypes described for SLC25A46 mutations. Furthermore, there was clear evidence that SLC25A46 should be added to the list of candidate genes causing premature neonatal death, with a potential link between early deaths and SLC25A46 mutations that result in the absence or in a drastic reduction of the amount of the protein (see Table 5). Finally, we produced the first Slc25a46 knock-out mouse model, which should be useful to further elucidate the function of SLC25A46 in mitochondrial dynamics.
All procedures involving animals conformed to the Guide for the Care and Use of Laboratory Animals (NIH Publication No.85-23, revised 1996). All efforts were made to minimize suffering.
Blood samples were collected from cattle by veterinarians or by trained and licensed technicians during routine blood sampling for paternity testing, genomic selection or annual prophylaxis. Affected calves were euthanized for ethical reasons, due to the absence of effective treatment. All samples and data were obtained with permission of breeders or breed organizations.
For mice, protocols were approved by the Animal Experimentation Ethics Committee and the French Ministry of Research (APAFIS#1227–2015100516164803 v3), and the Haut Conseil des Biotechnologies (HCB n°6461).
Twelve affected calves were examined clinically and confirmed to have the disease [29]. Blood samples were collected from these calves and their parents, and DNA was extracted with a Genisol Maxi-Prep kit. Blood samples were also collected from control animals known to be unaffected based on our genotypes database. In total, 123 unaffected adult cattle, all of the Rouge-des-Prés breed, were selected (including eight bulls used for artificial insemination and 115 cows from the La Greleraie INRA experimental facility). All of these cattle were genotyped by Labogena with the Bovine SNP50 Beadchip V1 (Illumina). Mapping was carried out by homozygosity mapping with in-house HOMAP software, as described [28].
Whole genome sequencing was performed at the Get-PlaGe platform (http://genomique.genotoul.fr/) on a HiSeq 2000 Illumina sequencer producing 100-bp long, paired end reads. Four animals had their entire genomic DNA sequence determined (two affected, one carrier and one healthy). Reads were quality checked and mapped on the UMD3.1 reference genome using BWA aln software (version 0.5.9-r16). Alignments were filtered with a minimum MAPQ value of 30. Reads that mapped to multiple localizations were removed. The target region was selected on each produced.bam file using Samtools (Version 0.1.18). Local indel realignment and base quality recalibration were applied using GATK toolkit. The SNPs were predicted with samtools mpileup and bcftools, and annotated with Ensembl Variant Effect Predictor tool and SNPs were filtered according to the animals’ phenotype-genotype correlation.
A total of 93 living Rouge-des-Prés cattle were tested for SLC25A46 and MAN2A1 polymorphisms, including 27 with clinical symptoms of distal axonopathy (with or without subsequent histopathological confirmation). In addition, 31 more historical Rouge-des-Prés animals were also tested, as well as 321 other cattle from 12 French breeds. For all of these, DNA was extracted from blood samples (Genisol Maxi-Prep kit or QIAsymphony DNA Kit (Qiagen)).
Sanger sequencing was performed using standard methods on the two potential polymorphisms identified after whole-genome sequencing, in SLC25A46 and MAN2A1 genes. Primers (S2 Table) were designed using Primer3. The PCR products were amplified using 200 ng of DNA, with standard GoTaq PCR reagents (Promega), on a Master Thermal Cycler (Eppendorf).
The SNP of SCL25A46 gene was genotyped using PCR-LAR (Ligation Assay Reaction) by Labogena. A pair of PCR primers (Turn_F and Turn_R) flanking the mutation was designed with Primer3.vo4 software, based on the genomic sequence of the bovine gene SCL25A46, according to the UMD3.1 assembly (S3 Table).
The PCR amplification was performed in a final volume of 10 μl using a Qiagen Multiplex PCR Kit, 10−50 ng of template DNA and 2.0 pmol of each primer. Reactions were run for 30 cycles in an MJ thermal cycler (Model PTC-200). The PCR amplification included an initial activation step at 95°C for 15 min, denaturation at 94°C for 30 s, primer annealing at 60°C for 90 s, extension at 72°C for 1 min, and final extension at 60°C for 30 min.
The following tagged probes were designed for the ligation assay, Turn_LAR-M ending with the mutated nucleotide and Turn_LAR-S ending with the non-mutated nucleotide, and Turn_2p, a double-phosphorylated primer (S3 Table).
The PCR product (10 μl) was used for allele discrimination using the Ligation Assay Reaction. The reaction contained 2 pmol of each probe, 1.5 U of Taq DNA Ligase and reaction buffer (New ENGLAND BioLabs). Reactions were run in an MJ thermal cycler (Model PTC-200). The ligation reaction included an initial activation step at 95°C for 2 min and the following thermocycling profile was repeated 35 times: denaturation at 94°C for 30 s and probe annealing at 60°C for 3 min. Finally, the reacting solution was held at 99°C for 10 min to deactivate Taq DNA Ligase.
Following the ligation reaction, an Applied Biosystems 3730xl DNA analyser with GeneMapper Analysis software (Applied Biosystems) was used to analyze fluorescently tagged fragments.
Exon3 of mouse Slc25a46 gene was targeted at the site of the original mutation associated with phenotype (chromosome:GRCm38:18:31604753:31606764:1 (reverse complement)). The target sequence was chosen with the ZiFiT Targeter program (http://zifit.partners.org). A potential TALEN target sequence identified by the program was selected empirically with a preference for an 18-16-18 combination (16 bases for the spacer). The chosen sequences were TGTGCTGGCCCATCCTTG for the left TALEN and CAGTGTCAGGTAAATATA for the right. No homology with the targeted sequence was identified (Blast NCBI) at any other location in the genome that could represent a potential off-target site The TALEN kit used for TALE assembly was a gift from the Keith Joung laboratory (Addgene kit # 1000000017).
The TALEN were constructed according the REAL (Restriction Enzyme And Ligation) assembly method, as described [67].The left TALEN was constructed by assembling units of the kit in the following order: 9, 15, 19, 22, 30, 14, 19, 22, 27, 12, 16, 25, 27, 12, 20, and 25 (by groups of four units). The entire insert was subcloned into the final JDS74 plasmid opened at the bsmb1 sites. Similarly, the right TALEN was constructed by assembling of units of the kit in the following order: 6, 15, 16, 25, 30, 15, 16, 22, 27, 15, 19, 21, 27, 11, 17, and 25. The entire insert was then subcloned into the final JDS74 plasmid. All inserts of final plasmids were entirely sequenced with primers 2978 (TTGAGGCGCTGCTGACTG) and 2980 (TTAATTCAATATATTCATGAGGCAC). To prepare RNA from each plasmid for microinjection, 5 μg of TALEN plasmid was linearized with 20 U of Age1 enzyme (New England Biolabs,) for 8 h at 37°C in 100 μl. The linearized fragment was purified by migration on an agarose gel and a Qiagen Gel extraction column kit (Qiagen). Messenger RNA was produced on 1 μg of purified linearized plasmid with the ARCA T7 capRNA pol kit (Cellscript, TEBUbio France) and polyadenylated with the polyA polymerase tailing kit (Epicentre) according to the manufacturer's instructions. Messenger RNA was purified with a Qiagen RNEasy minikit (Qiagen France), re-suspended in distilled water, and RNA concentration was estimated with a nanodrop photometer (Thermoscientific). The concentrated RNA was then diluted (100 ng/μl) in injection buffer (Millipore, France) and stored at -80°C until used.
The day of injection, 5 ng/μl of each TALEN RNA were mixed (final concentration, 10 ng/μl) in injection buffer and microinjected into pronuclei of murine FVB/N embryos, which were transferred into pseudo-pregnant mice. Resulting offspring were genotyped by DNA analysis of tail biopsies. Transgenic mice were crossed with FVB/N mice to derive F1 offspring that were used to produce the mentioned transgenic FVB/N Tg18 and Tg26 lines.
Transgenic mice were genotyped using a couple of primers (S2 Table) surrounding the TALEN restriction site. Since the AT content of the amplified fragment was high (71%), the PCR used KAPA2G Robust Taq (Kapa Biosystems), with KAPA2G buffer A and KAPA Enhancer, in accordance with manufacturer’s recommendations. The PCR products were amplified on a Master Thermal Cycler (Eppendorf) including an initial activation step at 95°C for 3 min, 40 cycles of denaturation at 95°C for 15 s, primer annealing at 60°C for 15 s, extension at 72°C for 15 s and final extension at 72°C for 3 min, followed by electrophoresis on a 3% agarose gel.
In accordance with their general condition, between 2 and 4 weeks of age, WT and Tg-/- mice were euthanized by cervical dislocation or anesthetic overdose (ketamine-xylazine) according to protocols approved by the Animal Experimentation Ethics Committee. Blood was collected (heparinized tubes) from the posterior vena cava [68], centrifuged (10 min at 3500 rpm and 4°C), and plasma was removed and frozen (-20°C) pending analyses. Various tissues were dissected, and either frozen at -80°C for RNA/protein extraction, or rinsed in PBS and fixed overnight in 4% PFA/PBS, then processed in a TP1020-1-1 tissue processor (Leica) and embedded in paraffin (EG 1160 Embedding Center; Leica).
Affected calves were euthanized for ethical reasons by intravenous administration of euthanasia solution (T-61, embutramide 200 mg/mL, mebezonium iodure 50 mg/mL, tetracaine chlorhydrate 5 mg/mL, 1 dose of 0.1 mL/kg, Intervet, Angers, France). Tissue were dissected and frozen at -80°C for subsequent protein extraction.
Plasma biochemical analyses were done at the Institut Clinique de la Souris (ICS, Strasbourg), and at the Laboratoire de Biochimie (Hôpital Bicêtre, Paris). Following parameters were measured: glucose, sodium, potassium, chloride, calcium, phosphorus, magnesium, urea, iron, ferritin, total proteins, albumin, total bilirubin, bile acids, total cholesterol, triglycerides, creatinine, beta hydroxybutyrate, lactate; as well as creatine kinase, aspartate aminotransferase and alanine aminotransferase.
Fixed and embedded tissues from WT and Tg-/- (n≥3) were sectioned (5 μm) and stained (Hemalun-Eosin-Saffron) in a VV24/4 VARISTAIN automatic slide stainer (Thermo Electron) according to a standard protocol. After staining, slides were scanned on a 3DHISTECH scanner (Sysmex).
Total RNA was extracted from various tissues using Qiazol reagent (Invitrogen) and the RNeasy mini kit (Qiagen) with DNase I treatment. Then, 500 ng RNAs were reverse-transcribed using a RT-vilo kit (Invitrogen), according to the manufacturer’s instructions.
To sequence transcripts, cDNA were amplified by PCR with PCR primers for Slc25a46 and Rpl13 genes (S4 Table). Following PCR reactions, products were electrophoresed on a 2% agarose gel and amplified cDNA fragments of Slc25a46 were sequenced.
Frozen tissues were ground with an Ultra-Turrax either in Mitochondria Isolation kit for Tissue (ThermoFisher) for mitochondria-enriched protein extracts, or in 10 mM Tris solution (pH 7.2, 2 mM MgCl2, 0.5% NP40, 1 mM DTT, with Halt Protease Inhibitor Cocktail EDTA free; Roche Diagnostics) for total protein extracts. Protein contents were assayed with the 2-D Quant kit (GE Healthcare) and 100 μg of proteins were separated on home cast 12.5% SDS-PAGE at 80 V for 20 min and 150 V for 50 min, with 10 μl of MW Marker (Nippon Genetics) added to the gel before electrophoresis. After separation, proteins were transferred on nitrocellulose membrane with a Trans-blot Turbo apparatus (BioRad) for 7 min at 2.5 A and 25 V. The membrane was rinsed twice, 5 min each, in MilliQ water and proteins were stained in 5X Rouge Ponceau solution for 5 min. Proteins were destained in TBS solution for 20 min and the membrane was blocked with 5% dry milk TBS-0.1% Tween20 solution for 1 h at room temperature. The membrane was incubated with polyclonal anti-SLC25A46 Novus antibody at a 1/1000 ratio in TBS-0.1% Tween20 solution for 1 h. The membrane was washed for 5 min and then twice for 15 min in fresh TBS-0.1% Tween20 solution. Thereafter, the membrane was incubated with goat anti-rabbit horseradish peroxydase-conjugated antibody (Santa Cruz) diluted at 1/10 000 in TBS-0.1% Tween20 for 1 h, and then washed as previously described. Bands were visualized by enhanced chemiluminescence (ECL Prime, GE Healthcare) and detected on ChemiDoc Touch (BioRad) in automatic mode.
The membrane was washed 5 min in TBS solution, 2 x 10 min in TBS-0.1% Tween20 and then was incubated with MTCO2 polyclonal antibody (ProteinTech) against cytochrome c oxidase II protein (COX2) at a 1/1000 ratio in TBS-0.1% Tween20 solution for 1 h, and then processed as described above.
Additional western blots were performed with rabbit polyclonal anti-Mfn2 (Santa Cruz) antibody at a 1/200 ratio, mouse monoclonal anti-MIC60 (Abcam) antibody at a 1/500 ratio and anti-OPA1 (Santa Cruz) antibody at a 1/200 ratio. For incubation of mouse samples with mouse antibodies, a preliminary saturation of mouse IgG was performed with AffiniPure Fab Fragment Donkey anti-mouse IgG (Jackson ImmunoResearch) at a 30 μg/mL ratio in TBS/Tween20 0.1% solution. Successive steps were as described above with secondary goat respective anti-rabbit or anti-mouse horseradish peroxydase-conjugated antibody (Santa Cruz) diluted at 1/10 000 in TBS-0.1% Tween20 for 1 h.
Mitochondria-enriched protein extracts were prepared as described in the above paragraph. Each lane of gel was cut into 20 gel pieces and analyzed separately, except the last two which were paired. After protein in-gel reduction (10 mM dithiothreitol), alkylation (50 mM iodoacetamide) and trypsin digestion (overnight incubation at 37°C with 100 ng of trypsin in 25 mM ammonium bicarbonate), the resulting peptides were extracted with 40% ACN/0.1% TFA (v/v). Tryptic peptides were dried and re-suspended with 40 μL of 2% ACN/0.08% TFA (v/v).
Peptide analysis was performed on a nanoLC system (Ultimate 3000, Thermo Scientific) coupled to an LTQ-Orbitrap Discovery mass spectrometer (Thermo Scientific) with a nanoelectrospray interface. The sample was loaded into a trap column (PepMap100, C18, 300 μm i.d. × 5 mm, 5 μm, Thermo Scientific) at a flow rate of 20 μl.min−1 for 3 min with 2% ACN/0.08% TFA (v/v). Peptides were then separated on a reverse phase nanocolumn (PepMap100, C18, 75 μm i.d. × 150 mm, 2 μm, Thermo Scientific) with a two-step gradient from 1 to 25% B for 42 min and from 25 to 35% B for 5 min at 300 nl.min−1 at 40°C (buffer A: 2% ACN/0.1% FA (v/v), buffer B: 80% ACN/0,1% FA (v/v)). Ionization was performed in positive mode (1.4 kV ionization potential) with a liquid junction and a sillica tip emitter (10 μm id, NewObjective). Peptide ions were analyzed using Xcalibur 2.0.7 with a data-dependent method including two steps: (i) full MS scan (m/z 300–1,400) and (ii) MS/MS (normalized collision energy fixed to 35%, dynamic exclusion time set to 45 s). The MS and MS/MS raw data were submitted for protein identification and quantification by spectral counting using the X!TandemPipeline 3.3.4 (version 2015.06.03, http://pappso.inra.fr/bioinfo/xtandempipeline/) with X! Tandem search engine (version Piledriver, 2015.04.01, http://www.thegpm.org/TANDEM) and Uniprot SwissProt mus musculus database (version 2015.10.14; 25248 entries). Search criteria used were trypsin digestion, carbamidomethyl (C) set as fixed modification and oxidation (M) set as variable modification, one missed cleavage allowed, mass accuracy of 10 ppm on the parent ion and 0.5 Da on the fragment ion. The final search results was filtered using multiple threshold filter: -4.0 protein log (E-value) identified with at least two different peptides with E-value < 0.01.
Differential analyses were performed using the Bioconductor Limma R package, with a voom transformation [69,70] and a Benjamini-Hochberg correction for multiple testing. Proteins with the lowest counts were removed using a threshold of 10 spectra for the sum over all replicates.
Samples of central and peripheral nervous systems were fixed for 3 h in 2.5% glutaraldehyde in Sœrensen buffer and osmificated for 1 h in 1% OsO 4, as described [71]. Afterwards, they were rinsed in Sœrensen buffer, dehydrated in graded acetone, and embedded in Epon. Semi-thin sections (1 μm) were stained with toluidine blue. Ultra-thin sections were stained with uranyl acetate and lead citrate and viewed using a JEOL electron microscope.
Liver was fixed with 2% glutaraldehyde in 0.1 M Na cacodylate buffer (pH 7.2), for 4 h at room temperature. Samples were then contrasted with 0.5% Oolong Tea Extract (OTE) in cacodylate buffer, postfixed with 1% osmium tetroxide containing 1.5% potassium cyanoferrate, gradually dehydrated in ethanol (30 to 100%) and substituted gradually with a mixture of propylene oxyde-epon and embedded in Epon (Delta Microscopie–Labège France). Thin sections (70 nm) were collected onto 200 mesh cooper grids, and counterstained with lead citrate. Grids were examined with an Hitachi HT7700 electron microscope operated at 80kV (Elexience–France), and images were acquired with a charge-coupled device camera (AMT).
Samples of brain, liver and muscle were transferred to ice-cold isolation buffer (20 mM Tris, 0.25 M sucrose, 40 mM KCl, 2 mM EGTA, and 1 mg/mL BSA; pH 7.2) and homogenized on ice by five strokes with a glass-Teflon potter homogenizer. Citrate synthase (CS), complex I (NADH-ubiquinone oxidoreductase) (CI), complex II (succinate-ubiquinone oxidoreductase) (CII), complex III (ubiquinone-cytochrome c oxidoreductase) (CIII), complex IV (cytochrome c oxidase) (CIV), aconitase (ACO), isocitrate dehydrogenase (IDH), α-ketoglutarate dehydrogenase (AKGDH) and fumarate hydratase (FH) activities were spectrophotometrically measured at 37°C as described [72,73]. All enzymatic activities were expressed as nanomoles per minute and per milligram of protein.
Total DNA was extracted from brain, liver and muscle homogenates using a standard procedure [74]. The mtDNA copy number per cell was measured by quantitative PCR based on the ratio of mtDNA (MTCO2 gene) to nDNA (PPIB gene), as described [74]. Long-range PCR was performed to detect large mtDNA deletions with primers F1 ACGGGACTCAGCAGTGATAAAT;R1 GCTCCTTCTTCTTGATGTCTT (expected size, 15144 bp).
|
10.1371/journal.ppat.1000754 | TgMORN1 Is a Key Organizer for the Basal Complex of Toxoplasma gondii | Toxoplasma gondii is a leading cause of congenital birth defects, as well as a cause for ocular and neurological diseases in humans. Its cytoskeleton is essential for parasite replication and invasion and contains many unique structures that are potential drug targets. Therefore, the biogenesis of the cytoskeletal structure of T. gondii is not only important for its pathogenesis, but also of interest to cell biology in general. Previously, we and others identified a new T. gondii cytoskeletal protein, TgMORN1, which is recruited to the basal complex at the very beginning of daughter formation. However, its function remained largely unknown. In this study, we generated a knock-out mutant of TgMORN1 (ΔTgMORN1) using a Cre-LoxP based approach. We found that the structure of the basal complex was grossly affected in ΔTgMORN1 parasites, which also displayed defects in cytokinesis. Moreover, ΔTgMORN1 parasites showed significant growth impairment in vitro, and this translated into greatly attenuated virulence in mice. Therefore, our results demonstrate that TgMORN1 is required for maintaining the structural integrity of the parasite posterior end, and provide direct evidence that cytoskeleton integrity is essential for parasite virulence and pathogenesis.
| The disease toxoplasmosis is the result of uncontrolled growth and proliferation of the intracellular parasite Toxoplasma gondii, which is pathogenic for most warm-blooded animals. If growth of the parasite is blocked, then it does not cause disease, even though it may persist in the host as a chronic infection. Proper assembly of the cytoskeleton of T. gondii is known to be essential for its growth, and consequently required for virulence. In this study, we investigated the function of a novel cytoskeletal protein, TgMORN1, in T. gondii. TgMORN1 is a major component of the basal complex, a novel cytoskeletal assembly located at the posterior end of the parasite. We found that TgMORN1 is required for maintaining the structural integrity of the parasite posterior end and is important for ensuring successful separation of daughters at late stage of parasite replication. In addition, infection with parasites deficient in TgMORN1 not only failed to kill mice but also provided protective immunity against a lethal challenge infection, indicating the importance of TgMORN1 in T. gondii growth both in vitro and in vivo.
| Toxoplasma gondii is one of the most successful human parasites, infecting ∼30% of the total world population. It is the most common cause of congenital neurological defects in humans, and an agent for devastating opportunistic infections in immunocompromised patients. T. gondii is also a member of the phylum Apicomplexa, which contains thousands of species of obligate intracellular parasites [1]. Like T. gondii, many of these parasites (such as Plasmodium spps, the causative agents for malaria) pose serious health threats to human beings. The damage caused by these parasites absolutely depends on their ability to replicate. For instance, T. gondii causes severe lytic cerebral and ocular lesions when the immune system fails to control its proliferation. Massive proliferation of Plasmodium parasites often results in hemolytic anemia, parasite-mediated destruction of red blood cell; and cerebral malaria, caused by parasite-engorged erythrocytes clogging blood vessels in the brain [2]–[4]. An understanding of the growth and division of these parasites is therefore crucial for developing effective therapies. The T. gondii cytoskeleton provides the framework for organellar partitioning, maintains cell shape and drives invasion, thus is essential for parasite survival and proliferation. Furthermore, it is rich in structural features that are unique to the parasites, thus highly attractive potential drug targets for designing parasite specific drugs.
The cytoskeleton of T. gondii is complicated but highly ordered. Each parasite contains one cytoskeletal apical complex (made of 3 ring structures and 14 filaments of a novel tubulin polymer), and 22 cortical microtubules [5],[6]. Overlying the microtubules is a regular two-dimensional (2D) meshwork formed by intermediate-filament like proteins, subtended beneath a tri-layer membrane containing a highly ordered 2D array of intra-membranous particles [7]–[12]. Its actin cytoskeleton is extremely dynamic. Most of its actin are kept in the monomeric form, and only undergo very transient polymerization in extracellular parasites for driving parasite motility and invasion into the host cell [13]–[15]. The entire daughter cytoskeleton is assembled afresh in a reproducible temporal sequence within the mother during each round of parasite replication [12],[16],[17].
Previously, we located a number of new cytoskeletal proteins, including TgMORN1 (Membrane Occupation and Recognition Nexus 1), TgCentrin2, and TgDLC- a member of the dynein light chain family, to a novel cytoskeletal structure at the extreme basal end of the parasite [18]–[20]. Due to its unique location and molecular composition, we named this structure “the basal complex” [18],[19]. We also discovered that although it will eventually become the extreme basal end of the parasite, the basal complex is initially constructed at the very beginning of daughter formation (Figure 1A), and in juxtaposition to the future apical end [18],[19]. It first appears as a TgMORN1-containing ring that caps the growing ends of the daughter cortical cytoskeletons. The cap is maintained throughout daughter development (Figure 1B), and eventually constricts into a cone at the basal end of the parasite when the daughter parasite becomes mature. It was a great surprise to find that the basal complex is formed so early, and equally surprising that it is initiated at the same site where the very first elements of the future apical complex are laid down. This highly suggestive combination of unexpected timing and unexpected location prompted the hypothesis that components of the basal complex might play some unknown critical role in the assembly of the daughter cortical cytoskeleton.
To address the role of the basal complex in T. gondii physiology, we decided to dissect the function of TgMORN1, as it is a major basal complex component and is also the earliest basal complex component identified so far. We found that TgMORN1 formed rings and fibers when ectopically expressed in bacteria, supporting the notion that it might function as a structural protein. We also made a knock-out mutant of TgMORN1 (ΔTgMORN1) and discovered that the structure of the parasite posterior end was grossly altered upon the loss of TgMORN1. Interestingly, ΔTgMORN1 parasites displayed cytokinesis defects, apicoplast segregation defects and growth defects in vitro. In mice, these parasites not only were avirulent but also provided protective immunity against a lethal challenge infection.
The MORN-domain is a structural module conserved from bacteria to human. MORN-domain containing proteins have been found in large protein complexes and are thought to mediate protein-protein or protein-lipid interactions [21]–[25]. TgMORN1 is mainly formed of 14 MORN repeats [18]–[20]. When ectopically expressed in E. coli, 6XHIS-mCherryFP-TgMORN1 was assembled into rings and fibers in the absence of other T. gondii proteins (Figure 1C), suggesting that the tendency of TgMORN1 to self-assemble might be involved in forming the basal ring structure during daughter construction (Figure 1A). 6XHIS-TgMORN1 and 6XHIS-eGFP-TgMORN1 formed similar structures when expressed in E. coli, with higher tendency to form fibers (data not shown).
To further understand its function, we decided to generate a knockout mutant of TgMORN1. Several attempts to eliminate TgMORN1 expression by one-step homologous recombination in RHΔHXGPRT (RHΔHX) parasites failed (unpublished results). These failures were not due to low homologous recombination frequency, as a parasite line in which endogenous TgMORN1 gene was replaced by homologous recombination with “LoxP-TgMORN1-HXGPRT-LoxP” was obtained fairly easily in the same set of experiments (one out of nine clones screened was positive) (Figure 2A), suggesting that the loss of TgMORN1 confers serious growth disadvantage to the parasite. In the “LoxP-TgMORN1-HXGPRT-LoxP” parasite line, the TgMORN1 coding sequence was flanked by two LoxP sites, allowing for the excision of the [TgMORN1-(HXGPRT expression cassette)] fragment after the transient transfection of a plasmid expressing Cre-eGFP into the parasite (Figure 2A). 6-thioxanthine (6-TX) selection was then applied to select for parasites that had lost HXGPRT activity, and five TgMORN1 negative clones (out of total 41 clones screened by immunofluorescence using a rat anti-TgMORN1 antibody) were obtained. Genomic PCR and western blot analysis confirmed the disruption of the TgMORN1 locus and the complete loss of TgMORN1 protein expression in the TgMORN1 knockout parasite (ΔTgMORN1) (Figure 2B&C).
It was immediately noticeable that the posterior ends of the ΔTgMORN1 parasites were highly irregular and heterogeneous and that the distribution of the width of IMC1 basal gap among these parasites was much more wide-spread than those of the parental strain (LoxP-TgMORN1-HXGPRT-LoxP) and the complement (ΔTgMORN1/eGFP-TgMORN1) (Figure 3A). The morphologies of ΔTgMORN1 parasites posterior ends can be clustered into three major groups. About 37% of the parasites had a basal IMC1 gap that was more than 1 µm (Figure 3B; white arrowheads), much wider than the average of the basal IMC1 gap of the parental strain, which was ∼0.65 µm. In the second group (∼35%), the basal IMC1 gap was either small or nonexistent, but the posterior ends of the parasites still appeared to be significantly wider than that of the parental strain, giving the parasite a “triangular” shape (Figure 3B; white arrow). In the third group (∼28%), the width of the parasite posterior end looked normal, however, irregular basal IMC structure was often present (Figure 3B; purple arrow). Likely a direct result of distorted parasite shape, ΔTgMORN1 parasites never formed a “rosette”- a common organization of wild-type parasites in vacuoles containing more than 8 parasites (Figure 3C). These defects were fully corrected in the ΔTgMORN1/eGFP-TgMORN1 line (Figure 3A–C). The arrangement of cortical microtubules around ΔTgMORN1 parasite cortex appeared to be normal (Figure S1).
As TgMORN1 is the earliest component recruited to the basal complex identified so far, we examined how the loss of TgMORN1 affected the localization of several other basal complex components. In ΔTgMORN1 parasites, eGFP-TgCentrin2 basal complex localization was undetectable, although the localization of eGFP-TgCentrin2 to the apical complex, the centrioles and the peripheral annuli was not affected (Figure 4A). Similarly, when eGFP-TgDLC was expressed in ΔTgMORN1 parasites, it was incorporated into the apical complex and spindle pole/centriole assembly as in the parental strain, but an eGFP-TgDLC concentration could not be detected at the parasite posterior end (Figure 4B). There could be two plausible explanations for the undetectability of an eGFP-TgCentrin2 or TgDLC basal concentration in ΔTgMORN1 parasites. It is possible that these two proteins failed to be recruited to the parasite basal complex because of the loss of TgMORN1. Alternatively, these proteins could spread to larger area in the disorganized parasite posterior end, which would dilute the signal, rendering it undetectable.
To assess the defects in parasite invasion in ΔTgMORN1 parasites, we performed invasion and gliding motility assays. We did not detect a significant difference in invasion among the parental, ΔTgMORN1 and the complemented parasites (Figure 5). P values for the comparison between ΔTgMORN1 and parental parasites; and ΔTgMORN1 and complemented parasites were 0.27 and 0.35 respectively.). We also did not observe qualitative differences among the trails deposited by the parental, ΔTgMORN1 and the complemented parasites in gliding motility assays (data not shown).
To assess the defects in parasite replication in ΔTgMORN1 parasites, we examined their intracellular growth (Figure 6). We found that a significant percentage of vacuoles contained parasites displaying cytokinesis defects, where daughter parasites failed to separate after budding (c.f. Figure 4B). This percentage increased from 21% (12 hours post infection) to 38% (24 hours post infection) as the total number of cell division events increased over time (Figure 6A, top). For the parental strain or ΔTgMORN1/eGFP-TgMORN1 parasites, less than 0.5% of the total vacuoles contained parasites displaying similar defects. The construction of new daughter parasites was observed in parasites where cytokinesis had failed (Figure 6A, bottom), suggesting that the completion of cytokinesis and the initiation of the next round of daughter formation are not tightly coupled, confirming previous observations [12],[26],[27].
To examine if ΔTgMORN1 parasites are defective in organelle biogenesis, we examined the segregation of the apicoplast, the mitochondrion as well as the de novo formation of the secretory organelles: dense granules, micronemes, and rhoptries. While we did not detect any significant defects in the segregation or synthesis of other organelles in these parasites, we found that parasites in ∼24% of the vacuoles contained no apicoplast (revealed by an antibody recognizing the Acyl Carrier Protein (ACP), an apicoplast protein [28]), indicating a modest apicoplast segregation defect (n = 250, Figure 6B). No apicoplast negative parasites were observed in either the parental or complemented parasites (n = 50).
When numbers of parasites per vacuoles were counted for 18, 24, 30 and 40 hours post-infection, we found a trend showing a decrease in replication rate upon the loss of TgMORN1 (Figure 6C). This assay also showed that asynchronous replication within the parasitophorous vacuole occurred much more frequently in ΔTgMORN1 parasites, where ∼15% of vacuoles at 24–40 hours post-infection, contained “odd” number (≠ 2n and <16) of parasites, comparing with less than 0.4% for the parental strain and ΔTgMORN1/eGFP-TgMORN1 parasites (Figure 6C).
To evaluate how the infectivity of the parasite was affected by TgMORN1 deficiency, we performed plaque assays and found that ΔTgMORN1 parasites formed plaques significantly smaller than those formed by the parental and ΔTgMORN1/eGFP-TgMORN1 parasites (Figure 7A). Because of this growth defect observed in vitro, the effect of TgMORN1 deletion on parasite virulence in mice was investigated. CD1 outbred mice were infected intraperitoneally with either 103 or 104 parental; 103, 104, 2×104 or 105 ΔTgMORN1; or 103 or 104 ΔTgMORN1/EGFP-TgMORN1 tachyzoites (Figure 7B). Mice that were infected with parental parasites started to show signs of disease (i.e. ascites due to tachyzoites in the peritoneum, ruffled fur) at day 5 post-infection (pi), and died between day 7 and day 9 pi, as expected. In contrast, mice that were challenged with 103 or 104 ΔTgMORN1 parasites showed no signs of disease and remained alive. Mice that received 2×104 or 105 ΔTgMORN1 tachyzoites showed a slightly swollen abdomen as sole sign of disease and remained alive. EGFP-TgMORN1 complementation restored parasite virulence, as mice that were infected with complemented parasites followed the same pattern as mice that were infected with the parental strain (Figure 7B). At day 21 pi, surviving mice “immunized” with 103 or 104 ΔTgMORN1 parasites were challenged with 10,000 wild type RH tachyzoites (LD100 = 1). Compared to naïve mice, which died between day 7 and day 10 pi, mice that were pre-infected with ΔTgMORN1 parasites were protected from lethal challenge, where all mice immunized with 104 ΔTgMORN1 parasites and 75% of mice immunized with 103 ΔTgMORN1 parasites remained alive and healthy more than 60 days after the challenge (Figure 7C).
Several knock-out strategies have been developed to study protein function in T. gondii. One-step homologous replacement [29] has been used to study the function of many non-essential genes. For single-copy essential genes, however, no viable knock-out mutant strain can be obtained using this method, because T. gondii grown in lab culture is asexual and contains a haploid genome. In fact, for any gene the loss of which results in ∼20% reduction in growth rate per cell cycle, it is practically impossible to acquire a knockout mutant using this method, as after ∼21 cell cycles (i.e. ∼7 days, the usual timeframe for drug selection before cloning), the percentage of the knockout parasite in the population will drop to less than ∼0.03% (0.821×0.03; 0.03 is the ratio of homologous vs nonhomologous events estimated in [29]). The failure to generate TgMORN1 knock-out mutants using one-step homologous replacement led us to implement the Cre-LoxP recombination technique [30],[31] (Personal Communications, Drs Gusti Zeiner, Michael Reese and John Broothroyd at Stanford University). There are two potential advantages of this strategy compared with the previously developed methods [29],[32]. First of all, it can be used for generating knock-out for genes whose function is sensitive to its expression level, because in the “LoxP-GENEX-HXGPRT-LoxP” intermediate, the expression of the target gene is driven by its endogenous promoter. Secondly, this strategy makes it easier to obtain a clonal population of knock-out mutants that have severe growth defects, because the loss of the target gene via recombination by Cre recombinase simultaneously results in the loss of HXGPRT, which allows for the enrichment of the knockout mutants through a subsequent 6-TX selection that highly inhibits the growth of parasites where recombination has not occurred.
The initiation, construction and the maturation of different parts of T. gondii cytoskeleton occur in a highly reproducible sequence. We have previously demonstrated that the basal complex was assembled at the beginning of daughter construction, prompting our hypothesis that the basal complex might play a guiding role in the initiation of the parasite cortical cytoskeleton [18],[19]. We were surprised to see, however, that although TgMORN1 is a prominent component of the basal complex, ΔTgMORN1 parasites managed to construct a cortical cytoskeleton functional enough to support parasite survival in vitro. This indicates that if the basal complex dictates the initiation of the daughter cortical cytoskeleton, it involves basal complex components other than TgMORN1. The prediction is that if such protein exists, its recruitment to the basal complex should be early and independent of TgMORN1.
Once the daughters are constructed, they need to be properly segregated to become a functional entity. We have previously proposed that the basal complex might play a role in this maturation step based on the correlation between the timing of the basal complex constriction and that of cytokinesis initiation [19]. The phenotypes of ΔTgMORN1 parasites support this hypothesis. The architecture of the posterior end in these parasites was clearly perturbed, which correlated with a significant increase in defective cytokinesis. This might be a direct result of losing TgMORN1, a major structural protein in the basal complex. Alternatively, the indirect effect on the localization of other basal complex components, such as TgCentrin2, might also play a role in producing this structural defect, as we found previously that TgCentrin2 containing basal structure underwent contraction when the intracellular calcium concentration was elevated [19]. We do not know at this time if TgMORN1 itself plays an active role in recruiting these basal complex components or if the basal structure formed by TgMORN1 provides a necessary platform for protein association. These defects in the biogenesis of the basal complex are likely the direct cause for the inefficient daughter parasite separation in ΔTgMORN1 parasites. Moreover, TgMORN1 most likely works with other proteins in the basal complex to drive cytokinesis, as the “penetrance” of the cytokinesis defect was not complete in the TgMORN1 deficient parasites. Besides cytokinesis defects, ΔTgMORN1 parasites displayed a modest apicoplast segregation defect, which also likely contributed to the growth defects of these parasites, as “apicoplast-less” parasites were shown to have a “delayed-death” phenotype [33]–[35]. It was proposed that the apicoplast is anchored to the centrosome during apicoplast division based on the close proximity between these two organelles [36],[37]. It will be interesting to elucidate in the future whether a direct connection between these two organelles exists and if TgMORN1 plays a role in the formation/maintenance of such a connection, or the apicoplast segregation defect in ΔTgMORN1 parasites is an indirect result of other defects in these parasites.
In sum, we have implemented a new knock-out strategy that is generally applicable to studying the functions of genes that are important for parasite growth. In addition, ΔTgMORN1 mutant will provide a convenient background to generate multi-gene knockouts for systematic dissection of the function of the basal complex and the interaction among its proteins. It will also facilitate mutagenesis analysis to understand the structural role of the MORN domain in general. Finally, our study provides direct evidence that cytoskeleton integrity is essential for parasite virulence and pathogenesis, as TgMORN1 deficiency has a profound effect on parasite virulence in vivo.
All mice were maintained under specific-pathogen-free conditions in accordance with institutional guidelines of Institute of Parasitology, McGill University, Ste-Anne-de-Bellevue, H9X3V9, QC, Canada. The animal protocol was approved by the McGill University Macdonald campus Facility Animal Care Committee.
The constructs for generating the parental strain: “LoxP-TgMORN1-HXGPRT-LoxP” parasite were produced using PTKO2_II as the backbone, and PTKO2_II was constructed based on PTKO (a kind gift from Drs Gusti Zeiner, Michael Reese and John Broothroyd at Stanford University). PTKO contains total two LoxP sites and two multiple cloning sites (MCS), with one MCS (MCS1) placed at the 5′ end of the first LoxP site and the second MCS (MCS2) placed at the 3′ end of the second LoxP site. In PTKO2_II, an additional MCS (MCS3) (AGATCTGTTTAAACGCGATCGCGGTCCGAGGCCT) was added to the 3′ end of the first LoxP site in PTKO, and 3′ portion of MCS1 in PTKO was modified to (GGTACCCTCGAGGATATCTACGAATTC). To construct PTKO2_II, a 521bp DNA fragment (Table S1) containing these changes was synthesized and cloned into PUC-57 (Genscript, Inc, Piscataway, NJ), digested with KpnI and SpeI, and ligated into PTKO to replace the corresponding KpnI-SpeI fragment on PTKO. To construct the “LoxP-TgMORN1-HXGPRT-LoxP” plasmid, first, a 2.1kb fragment located to the 3′ end of TgMORN1 coding sequence in the genome was amplified using HK180 and HK181 (Table S1), digested with NheI and ApaI and ligated into NheI and ApaI sites of PTKO2_II (c.f. Figure 2A), which resulted in the plasmid PTKO2_II_3′UTR. Secondly, a ∼2.3kb fragment located to the 5′ end of TgMORN1 coding sequence in the genome was amplified using primers HK182 and HK183 (Table S1), digested with EcoRI and KpnI, and ligated into EcoRI and KpnI sites of PTKO2_II_3′UTR (c.f. Figure 2A), which resulted in the plasmid PTKO2_II_TgMORN1_5_3′UTR. Lastly, TgMORN1 coding sequence was amplified from pmin-eGFP-TgMORN1 [18] using primers HK191 and HK193 (15 bp of TgMORN1 Kozak sequence was included in the primer HK191) (Table S1), digested with BglII and StuI and ligated into BglII and StuI sites (c.f. Figure 2A) of PTKO2_II_TgMORN1_5_3′UTR to give the “LoxP-TgMORN1-HXGPRT-LoxP” plasmid. Genomic DNA for amplifying the 5′ and 3′ UTR of TgMORN1 was harvested from RH parasites using Qiagen DNeasy Blood & Tissue kit (Cat# 69504, Qiagen).
ptub-Cre-GFP was generated by amplifying Cre-GFP from pCAG-Cre:GFP (Plasmid 13776, Addgene) using primers HK223 and HK224 (Table S1). The PCR product was then digested with NheI and AflII and ligated immediately downstream of the ptub promoter in place of mCherryFP-eGFP/ NheI-AflII in ptub-mCherryFP-eGFP (a kind gift from Dr. John Murray, University of Pennsylvania). pmin-Cre-GFP was constructed by replacing eGFP-TgDLC/NheI-AflII in pmin-eGFP-TgDLC [18] with Cre-GFP/NheI-AflII from ptub-Cre-GFP.
To construct pQE30-6xHIS-TgMORN1, TgMORN1/BglII-AflII from pmin-eGFP-TgMORN1 was ligated into pQE30-DIP13_AflII [38] to replace DIP13/BamHI-AflII. To construct pQE30-6xHIS-mCherryFP-TgMORN1, mCherryFP-TgMORN1/NheI-AflII from pmin-mCherryFP-TgMORN1 [19] was ligated into PQE30-DIP13_AflII_NheI [38] to replace DIP13/NheI_AflII. The construction of pQE30-6xHIS-TgMORN1 and pQE30-6xHIS-mCherryFP-TgMORN1 was carried out by Biomeans Inc. (Sugar Land, TX).
Frozen stocks of BL21(DE3)pLysS bacteria transformed with pQE30-6XHIS-mCherryFP-TgMORN1 were made from cultures of single colonies. The frozen stock were then streaked on LB agar plates containing 100µg/ml of ampicillin, 50µg/ml chloramphenicol and grown at 37°C for 24 hours, then at 4°C for ∼24–48 hours. Alternatively, liquid cultures were grown from frozen stocks at 37°C in 100 ml LB containing 100µg/ml of ampicillin, 50µg/ml chloramphenicol (LB-amp-cap) for ∼24 hours. For unknown reasons, 6XHIS-mCherryFP-TgMORN1 expression in E. coli varied among cultures when grown in suspension and in a given experiment usually about 1 in 3 cultures expressed 6XHIS-mCherryFP-TgMORN1 well, forming rings and fibers in ∼100% of the bacteria. The expression of 6XHIS-mCherryFP-TgMORN1 in bacteria grown on LB agar plate was more consistent. Samples were processed as described in [39] before imaging.
T. gondii tachyzoites were used in all experiments, and the maintenance of parasites by continuous passage in human foreskin fibroblasts (HFFs) and parasite transfections were performed as previously described [40].
33 µg LoxP-TgMORN1-HXGPRT-LoxP plasmid was linearized with ApaI (c.f. Figure 2A) and transfected into ∼1×107 RHΔHXGPRT (RHΔHX) parasites [41], and selected by MPA (25µg/ml) and xanthine (50µg/ml). GFP negative parasites were collected by flow cytometry and 3 parasites were sorted into each well of a 96 well plates containing HFF. Single clones were then amplified, and screened by PCR for parasites where the endogenous TgMORN1 locus has been replaced by LoxP-TgMORN1-(HXGPRT expression cassette)-LoxP. This parasite line, named LoxP-TgMORN1-HXGPRT-LoxP /parental strain was then transfected with 25 µg pmin-Cre-GFP plasmid to excise the fragment between LoxP, and then placed under 80µg/ml 6-thioxanthine (6-TX) selection. Single clones were chosen after the second and third passage of the 6-TX resistant population, and TgMORN1 knockout (ΔTgMORN1) parasites were first selected by immunofluorescence using a rat anti-TgMORN1 antibody and subsequently confirmed by genomic PCR and western blotting.
To generate ΔTgMORN1/eGFP-TgMORN1 parasites, 7×106 ΔTgMORN1 parasites were transfected with 25µg pmin-eGFP-TgMORN1 plasmid [18] and grown without applying any drug selection. After ∼16 days, 100% of the parasites expressed eGFP-TgMORN1, which were then used for assessing the effectiveness of the complementation. pmin-eGFP-TgMORN1 plasmid contains no T. gondii selectable marker. Therefore like ΔTgMORN1 parasites, ΔTgMORN1/eGFP-TgMORN1 parasites were also HXGPRT deficient.
Single BL21(DE3)pLysS bacterial colonies containing pQE30-6xHIS-TgMORN1 plasmid were grown overnight in LB-amp-cap at 37°C. Cultures were then diluted 1∶20 in 2 liter LB-amp-cap and grown till OD600 reached ∼0.6–0.8 before the addition of isopropyl β-D-1-thiogalactopyranoside (IPTG) to 1 mM. Cultures were then grown for additional 4 hours at 37°C. Cells were then pelleted at 6,000 rpm for 25 minutes, resuspended in 40 ml of cold lysis buffer (8 mM Tris-Ac pH 7.5, 3 mM Trisbase, 100 mM KAc, 1 mM MgAc) containing 9.6g of cell lytic express (∼8 vials, Cat# C1990, Sigma), 1 µM TAME (Cat# T4626, Sigma) and 1 µM PMSF (Cat# P7626, Sigma), and incubated at 4°C for ∼60 minutes. Cells were sonicated∼five times for 30 seconds each with 1 minute cooling between each cycle, then centrifuged at 15,000 rpm for 15 minutes at 4°C. 1.9 ml packed Talon resin (Cat# 635501, Clontech) equilibrated with lysis buffer was then added to the supernatant and gently mixed at 4°C for 1 hour. The resin was then washed with lysis buffer with 10mM imidazole 4 times and eluted with ∼1.5 ml 1XLDS sample buffer and reducing reagents buffer (Cat# NP0007, Invitrogen).
Eluted 6XHIS-TgMORN1 proteins (see above) were loaded on 4–12% Bis-Tris gel and gel slices containing 6XHIS-TgMORN1 were then used to inject rats for antibody production (Cocalico Biological, Inc). The affinity purification of TgMORN1 antibody was carried out as described in [42]. Briefly, PVDF membrane blot with immobilized recombinant 6XHIS-TgMORN1 was blocked in 1% “Blotto”(Cat# 1152709001, Roche) in TBS (20mM Tris base pH 7.4, 150mM NaCl) for 30min; diluted rat anti-TgMORN1 serum (1∶10, diluted in 0.5%“Blotto” in 1xTBS+ 0.1% (v/v) Tween-20 (TBS-T)+10mM Na N3) was then incubated with the blot for overnight at room temperature (∼18–20hours). The blot was then washed twice in 1xTBS-T for 15 minutes, and once in 1×PBS for 15 minutes at room temperature. Absorbed antibody was then eluted by incubating with 0.2M glycine pH 2.5 at room temperature for 3 minutes. The pH of the eluted antibody solution was then neutralized by adding 0.09V of 1M Trisbase (pH unadjusted). NaN3 was then added to the solution to 10mM. The purified antibody was stored at 4°C.
For each sample, 5×106 extracellular parasites were lysed by incubating in 1× SDS sample buffer (62.5 mM Tris pH 6.8, 2% (w/v) sodium deodeoyl sulfate, 10% (v/v) glycerol and ∼0.5 mg of bromophenol blue) containing 50 mM DTT at 100°C for 10 minutes. Western blot was performed as described in [38]. Affinity purified rat anti-TgMORN1 was diluted 1∶10 and mouse-anti-tubulin B-5-1-2 (Cat# T6074, Sigma, U.S.A) was diluted 1∶4,000 in TBS-T containing 0.5% (v/v) blocking buffer (Cat# 1152709001, Roche). Goat anti-rat IgG HRP (Cat# NA935V, GE Healthcare, United Kingdom) was diluted 1∶1000, and goat anti-mouse/rabbit HRP (Cat# 1152709001, Roche) were diluted 1∶20,000 in TBS-T containing 0.5% blocking buffer.
Intracellular parasites were fixed with 3.7% formaldehyde in 1×PBS for 15 minutes, permeabilized with 0.25 or 0.5% TX-100 in 1×PBS for 15 minutes, then blocked with 1 or 3% BSA in 1×PBS (blocking buffer) for 30 minutes at room temperature. The cells were then incubated in primary and subsequently secondary antibody solutions (diluted in blocking buffer) for 60 minutes each. Primary antibody dilutions were as follows: mouse anti-IMC1, 1∶1000 (A kind gift from Dr. Gary Ward, University of Vermont); rabbit anti-ACP (Kind gifts from Dr. Manami Nishi at McGill University and Dr. Dhanasekaran Shanmugam at University of Pennsylvania) 1∶500 or 1∶10; affinity purified rat anti-TgMORN1, 1∶10. Secondary antibody dilutions were as follows: goat anti-mouse Alexa 488(Cat# A11029, Molecular Probes-Invitrogen), 1∶2000; goat anti-mouse Alexa 568 (Cat# A11031, Molecular Probes-Invitrogen), 1∶2000; goat-anti-rat Cy3 (Cat#112-165-167, Jackson ImmunoResearch), 1∶500; goat-anti-rat Alexa 488 (Cat# A11006, Molecular Probes-Invitrogen), 1∶2000; goat anti-mouse Cy3 (Cat# 115-165-166, Jackson ImmunoResearch), 1∶500; goat anti-rabbit Alexa488 (Cat# A11034, Molecular Probes-Invitrogen), 1∶1000; goat anti-rabbit Cy3 (Cat# 111-165-144, Jackson ImmunoResearch), 1∶500; donkey anti-rabbit Alexa 488 (Cat# A21206, Molecular Probes-Invitrogen), 1∶1000; donkey anti-mouse Alexa 594 (Cat# A21203, Molecular Probes-Invitrogen), 1∶1000.
3D image stacks were collected at room temperature at z-increments of 0.3 µm on an Applied Precision Delta Vision imaging station constructed on an Olympus IX-70 inverted microscope base. A 100× oil immersion lens (NA = 1.4) and immersion oil at refractive index 1.518 were used for all the imaging. Deconvolved images were computed using the point-spread functions and software supplied by the manufacturer. All fluorescent images were maximum intensity projections of deconvolved 3D stacks unless otherwise stated. The brightness and contrast of images used in the final figures were optimized for color prints.
Measurement was performed on parasites in vacuoles containing one or two parasites. Specifically, parasites were labeled with IMC1 antibody as described above. 3-D stacks were acquired as described above, and images focused at the mid-section of the parasite were used for measuring the width of basal IMC1 gap in the parasites using Softworx (Applied Precisions, Inc.).
Invasion assays were performed as previously described [43] with the following modifications. 1×107 extracellular parental, ΔTgMORN1 or ΔTgMORN1/eGFP-TgMORN1 parasites were added to confluent HFF monolayers and incubated at 37°C for 1 hour. Cells were fixed in 3.1% formaldehyde and 0.06% glutaraldehyde (diluted in PBS) for 15 minutes at room temperature. Extracellular parasites were labeled with mouse anti-SAG1 antibody (Cat #11-132, Argene, North Massapequa, NY, diluted 1∶500) and visualized with goat anti-mouse Alexa 568 (1∶1,000). Cells were then permeabilized with 0.25% TX-100 diluted in PBS for 15 minutes at room temperature and intracellular and extracellular parasites were labeled with mouse anti-SAG1 and visualized with goat anti-mouse Alexa 488 (1∶1000). All antibody incubations were performed for 30 minutes. The number of invaded (green only) parasites was calculated by subtracting the number of extracelluar (dual-labeled) parasites from the total number of parasites on the coverslip. Images were taken from 6 randomly chosen fields at 10× magnification and counting was performed using MetaMorph® software. Results were from three independent experiments. Motility assays were performed as previously described [13].
Equal numbers of parental, ΔTgMORN1 or ΔTgMORN1/eGFP-TgMORN1 parasites were added to confluent HFF monolayers and grown for 12, 18, and 24 hours. Immunofluoresence assay with rat anti-TgDIP13 diluted 1∶400 (for visualizing the apical complex), mouse anti-IMC1 diluted 1∶500, and DAPI diluted to 1µg/ml was performed as described above. To assess the effect of TgMORN1 deficiency on cytokinesis, the number of vacuoles containing at least one parasite displaying cytokinesis defects was counted for a total of 200 vacuoles per time point in each of 3 independent experiments. The counting was restricted to the vacuoles with fewer than 16 parasites, because in larger parasitophorous vacuoles ΔTgMORN1 parasites were too disorganized for assessing the level of cytokinesis defect accurately.
Replication assays were performed as previously described [44]. The number of parasites/vacuole in 200 vacuoles per time point (18, 24, 30, and 40 hours) was counted in each of 3 independent experiments. Parasites that failed cytokinesis were counted as two. For the 40-hour time point, vacuoles containing one and two parasites were not included in the counting, as they were most likely secondary vacuoles formed by parasites that egressed from large primary vacuoles and reinvaded.
To analyze apicoplast segregation, parental, ΔTgMORN1 and ΔTgMORN1/eGFP-TgMORN1 parasites were grown for 20–24 hours and immunofluoresence was performed using an anti-ACP antibody as described above. For parental and ΔTgMORN1/eGFP-TgMORN1 parasites, total 50 vacuoles were counted. For ΔTgMORN1 parasites, total 250 vacuoles were counted (from 5 independent experiments; 50 vacuoles per experiment) and vacuoles were scored as apicoplast positive (i.e. all parasites in the vacuole contain apicoplast), apicoplast negative (i.e. none of the parasites in the vacuole contains apicoplast) or mixed (i.e. vacuoles contains both apicoplast positive and negative parasites).
Equal numbers of parental, ΔTgMORN1, and ΔTgMORN1/eGFP-TgMORN1 parasites were allowed to infect and grow in fully confluent HFF for 11 days. The cultures were then fixed and permeablized in cold methanol (−20°C) for 15 minutes and stained with Coomassie® Brilliant Blue G-250 dye (BioRad, Catalog#: 500-0006) at room temperature for 2–3 hours, then 4°C overnight before scanning.
Freshly lysed out tachyzoites were filtered (3 µm), spun down and parasite pellet was resuspended in PBS. Parental (103 or 104), ΔTgMORN1 (103, 104, 2×104 or 105) or ΔTgMORN1/eGFP-TgMORN1 (103 or 104) tachyzoites (in 0.1 ml of PBS) were injected intraperitoneally (IP) into 6–8 week old CD1 outbred female mice (total 8 groups at n = 4; Charles River, QC). After 21 days, surviving mice “immunized” with 103 or 104 ΔTgMORN1 parasites were challenged IP with 10,000 wild type RH tachyzoites.
TgMORN1 (583.m05359); TgCentrin2 (50.m03356); TgIMC1 (44.m00004); T. gondii Acyl Carrier Protein (TgACP; 55.m00019); T. gondii α1 –tubulin (TgTubA1; 583.m00022); T. gondii dynein light chain (TgDLC; 41.m01383).
|
10.1371/journal.pcbi.1003395 | Utilizing a Dynamical Description of IspH to Aid in the Development of Novel Antimicrobial Drugs | The nonmevalonate pathway is responsible for isoprenoid production in microbes, including H. pylori, M. tuberculosis and P. falciparum, but is nonexistent in humans, thus providing a desirable route for antibacterial and antimalarial drug discovery. We coordinate a structural study of IspH, a [4Fe-4S] protein responsible for converting HMBPP to IPP and DMAPP in the ultimate step in the nonmevalonate pathway. By performing accelerated molecular dynamics simulations on both substrate-free and HMBPP-bound [Fe4S4]2+ IspH, we elucidate how substrate binding alters the dynamics of the protein. Using principal component analysis, we note that while substrate-free IspH samples various open and closed conformations, the closed conformation observed experimentally for HMBPP-bound IspH is inaccessible in the absence of HMBPP. In contrast, simulations with HMBPP bound are restricted from accessing the open states sampled by the substrate-free simulations. Further investigation of the substrate-free simulations reveals large fluctuations in the HMBPP binding pocket, as well as allosteric pocket openings – both of which are achieved through the hinge motions of the individual domains in IspH. Coupling these findings with solvent mapping and various structural analyses reveals alternative druggable sites that may be exploited in future drug design efforts.
| Drug resistance has recently entered into media conversations through the lens of MRSA (methicillin-resistant Staphylococcus aureus) infections, but conventional therapies are also failing to address resistance in cases of malaria and other bacterial infections, such as tuberculosis. To address these problems, we must develop new antibacterial and antimalarial medications. Our research focuses on understanding the structure and dynamics of IspH, an enzyme whose function is necessary for the survival of most bacteria and malaria-causing protozoans. Using computer simulations, we track how the structure of IspH changes in the presence and absence of its natural substrate. By inspecting the pockets that form in the normal motions of IspH, we propose a couple new routes by which new molecules may be developed to disrupt the function of IspH. It is our hope that these structural studies may be precursors to the development of novel therapies that may add to our current arsenal against bacterial and malarial infections.
| In the past couple decades, antimicrobial drug resistance has risen dramatically and greatly hampered the efficacy of currently available therapies for bacterial and malarial infections [1]–[9]. Whereas (multiple-)drug-resistant bacterial infections are a ubiquitous problem, affecting both the Western world and developing nations, the burdens of malaria fall disproportionately on the poorest regions of the world, with over 219 millions cases and 666,000 deaths reported in 2010 [3]. Beyond the common problems associated with decreased lifetimes for drug efficacy due to rapid development of resistance [1], [2], [5], [6], [9], advances in the fight against bacterial and malarial infections have also been plagued by diminished attention from major pharmaceutical companies toward the development of new therapies and drugs [4], [5], [10]. Consequently, there is urgent need for the development of new drugs with novel modes of action, for administration either independently or in combination with established regimen, both to combat bacterial and malarial infections, as well as to address the propensity of each for rapidly developing drug resistance [1], [4], [6], [7], [9].
The nonmevalonate pathway for isoprenoid biosynthesis has recently been revealed as a novel target for both antibacterial and antimalarial drugs. Isoprenoids comprise essential metabolites derived from the 5-carbon biomolecules, isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP, Figure 1), examples of which include sterols that provide structural support to membranes, chlorophylls used in photosynthesis, and quinones that participate in electron transport chains [11]–[14]. In contrast, animals acquire IPP and DMAPP in a distinctive manner via a mevalonate-dependent pathway. Given this metabolic difference, the proteins involved in the nonmevalonate pathway provide novel targets for the development of antibacterial and antimalarial drugs that are both broadly specific to pathogenic species such as H. pylori, M. tuberculosis and P. falciparum and without known human analogs [15]–[18].
The ultimate step in the nonmevalonate pathway is the generation of IPP and DMAPP through a 2-electron reductive dehydroxylation of (E)-1-hydroxy-2-methyl-but-2-enyl pyrophosphate (HMBPP) by IspH, a [4Fe-4S] protein (Figure 1) [11], [19]–[21]. The catalytic mechanism of IspH has been a topic of great debate, largely due to uncertainties introduced by the iron-sulfur cluster [18], [22]. Initial structures of IspH from Aquifex aeolicus [23] and Escherichia coli [24] solved by X-ray crystallography resemble cloverleaves and comprise three sequentially different domains with pseudo-C3 symmetry, each tethered to a [Fe3S4]+ cluster via a conserved cysteine residue. The A. aeolicus [Fe3S4]+ IspH structure (PDB ID: 3DNF; henceforth referred to as [Fe3S4]+(open, substrate-free) IspH) assumes an open conformation, with a 10×20 Å cavity where the HMBPP molecule is expected to bind at the cluster [23]. In contrast to the A. aeolicus crystal structure, the [Fe3S4]+ E. coli counterpart is closed around an inorganic diphosphate molecule (PPi) that sits in the vicinity of the centrally located [Fe3S4]+ cluster. Various conserved polar and charged residues, including Glu-126, Thr-167, Asn-227, His-41, His-74, His-124, Ser-225, Ser-226 and Ser-269 (E. coli numbering scheme), coordinate the PPi molecule, likely via hydrogen bonding or salt bridge interactions [24]. The orientations of these conserved residues in the E. coli structure are distinct from their A. aeolicus counterparts due to a tilt of a single domain that enables co-localization of charged and polar residues around the PPi in the case of the former.
While results from electron paramagnetic resonance (EPR) spectroscopy have shown [Fe3S4]+ IspH to be catalytically active [25], reconstituted IspH displays EPR and Mossbauer signatures of a [Fe4S4]2+ cluster [26], [27]. Groll and co-workers provide further support for the catalytically relevant form of IspH containing a [Fe4S4]2+ cluster with their work in crystallizing IspH in the presence of its substrate, HMBPP. This HMBPP-bound crystal structure (PDB ID: 3KE8, henceforth referred to as [Fe4S4]2+(closed, HMBPP-bound) IspH) assumes a closed conformation having a domain tilt similar to that of the [Fe3S4]+ E. coli structure, with HMBPP bound via its terminal hydroxyl moiety to an unliganded iron of a [Fe4S4]2+ cluster (Figure 2) [28]. The coordination sphere of the HMBPP ligand is virtually identical to the inorganic diphosphate molecule, while its terminal hydroxyl moiety interacts with Glu-126, Thr-167 (E. coli numbering) and an ordered water molecule to make a hydrogen bond network that is proposed to facilitate proton transfer during catalysis [28]. While these structural data provide a good picture of the [Fe4S4]2+ IspH structure with HMBPP bound, the structure of the 4Fe-form in the absence of substrate, as well as a detailed understanding of how IspH changes conformation upon ligand binding, are not fully understood.
Drawing from insight gained from the aforementioned structural work, as well as various spectroscopic and mutational studies, multiple groups have contributed to drug discovery efforts on the IspH target [29]–[34]. To the best of our knowledge, IspH inhibitor development has fallen under two classes: (1) HMBPP analogues [29]–[31] and (2) pyridine or alkenyl/alkynyl diphosphates and bisphosphonates [32]–[34]. In the case of HMBPP analogues, inhibitor binding emulates the natural substrate, while leveraging improved interactions with the Fe-site (e.g. binding of a thiol instead of an alcohol) [30], [31]. Alternatively, Oldfield and co-workers have created novel inhibitors of IspH by utilizing olefinic and pyridine groups to form π/σ “metallacycle” complexes and η1-complexes, respectively, coupling these metal binding groups to phosphate skeletons that preserve the hydrogen bond and salt bridge interactions present in IspH-HMBPP complexation [32]–[34]. These initial drug discovery efforts may be enhanced, both in terms of finding new lead compounds and developing already discovered leads, by obtaining a better description of the IspH binding pocket and possible allosteric sites that may be targeted.
Given that there exists no high-resolution structural data for substrate-free, [Fe4S4]2+ IspH, this work employs accelerated molecular dynamics (aMD) simulations to describe the dominant conformations available to IspH having a fourth iron atom in the absence of HMBPP. Characterization of these dominant conformations reveals an expanded binding pocket and allosteric sites that may be targeted with future rational drug design efforts. Additional attention is directed toward understanding how IspH dynamics are altered upon ligand binding, allowing us to propose a mechanism for how IspH-HMBPP complexation is achieved.
Consistent with the nomenclature used by Gräwert, et al. [28], descriptions of IspH from this point forward will use the nomenclature D1, D2 and D3 to describe the domains containing residues 14–96; 97–193; and 194–281, 1–13, respectively (A. aeolicus numbering, Figure 2). We perform 3×100 ns aMD simulations of [Fe4S4]2+(open, substrate-free) IspH, starting from the A. aeolicus crystal structure with a fourth iron modeled into the cluster, as described in the Methods. All trajectories are aligned to the [Fe3S4]+(open, substrate-free) IspH crystal structure by the backbone atoms of all D1 residues, since these residues display significantly lower fluctuation throughout the simulation than those in D2 and D3 [23]. The root-mean-square deviation (RMSD) for the backbone atoms of all residues after alignment is given in Figure 3a. From this RMSD analysis, it is apparent that each independent trajectory samples conformational space differently. The large changes in RMSD correspond to opening and closing motions of the D2 and D3 domains, providing a more dynamic description of the [Fe4S4]2+(open, substrate-free) state than is acquired from a static X-ray structure. While all three simulations extensively sample conformational space near the [Fe3S4]+(open, substrate-free) IspH crystal structure for the first ∼20 ns of the simulation, one simulation diverges from this experimental reference, implying that other distinctive, low-energy conformational states exist for substrate-free IspH.
Using Schrodinger's Glide program [35]–[37], we dock HMBPP to the unique iron site in IspH. Docked poses are filtered applying knowledge from experiment that the terminal alkoxide/alcohol group of HMBPP directly chelates the apical Fe site [26], [28], [38], [39]. The docked pose used in our MD studies is found by constraining the position of the terminal alkoxide moiety to within a 2.5 Å radius of the apical iron. While the orientation of the PPi moiety in our docked pose differs from the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure (3KE8) [28], it is worth mentioning that the cyclic structure of HMBPP observed in the crystal structure likely results from “induced fit” effects, with polar and charged groups closing around the PPi moiety. Given these effects are absent from our docking procedure, we use the Glide geometry as a starting point for elucidating how open, substrate-free IspH responds to the formation of an encounter complex with HMBPP bound to its unliganded Fe.
Similar to the [Fe4S4]2+(open, substrate-free) simulations, three independent, 100 ns aMD simulations of HMBPP docked into the open, [Fe4S4]2+-IspH structure (henceforth referred to as [Fe4S4]2+/HMBPP(open, docked)) are aligned to the [Fe3S4]+(open, substrate-free) IspH crystal structure, with the RMSD of all backbone atoms to the crystal structure given in Figure 3b. Both seeds one and three (Figure 3b, black and blue, respectively) approach an RMSD of ∼8–10 Å, with respect to the crystal structure. This jump occurs rapidly for seed three (in the first 20 ns of simulation), while seed one only appears to approach this level in the last 10 ns of simulation. This shift from the [Fe3S4]+(open, substrate-free) IspH crystal structure results from the closing of D2 and D3 around the docked HMBPP, matching the conformation assumed by the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure (Figure S1).
To gain insight into the dominant conformations sampled by these [Fe4S4]2+/HMBPP(open, docked) simulations, we cluster the frames of each trajectory according to pairwise RMSD comparing Cα atoms, as described in the Methods. The dominant cluster (58%) corresponds to an open conformation, similar to the [Fe3S4]+(open, substrate-free) IspH crystal structure [28]. The second most populated cluster (18%) contains closed structures resembling the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure (Figure 4). When considering the structures in this closed cluster, it is notable that the ligand does not form a ring structure consistent with its pose in the crystal structure. Nevertheless, the closing of the D2 and D3 domains around the substrate is consistent with the [Fe4S4]2+(closed, HMBPP-bound) experimental reference [28]. A more detailed inspection of the HMBPP environment in a representative structure from this closed cluster reveals the three key active site histidines, as well as the conserved Thr-165, Thr-166, Glu-126, Ser-221, Asn-223 and Ser-265 forming contacts with HMBPP that appear identical to those seen in the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure (Figure 4b,c). While Glu-126 and Thr-167 are co-localized with the iron-sulfur cluster in the active site of [Fe4S4]2+(open, substrate-free) IspH (Figure 4d), the other contacts mentioned are unique to substrate-bound IspH, as seen in the [Fe4S4]2+(closed, HMBPP-bound) crystal structure (Figure 4b). These findings demonstrate that aMD simulations have effectively captured the closing of loops from D2 and D3 around HMBPP—confirming earlier hypotheses for how conformational change occurs upon substrate binding [28].
Inconsistent with the RMSD results for seed three, both seeds one and two (black and red, Figure 3b) are trapped in a basin near the [Fe3S4]+(open, substrate-free) IspH crystal structure for a majority of their respective simulations. These differing trajectories arise, in part, because the residues in D3 that coordinate the PPi of HMBPP in seeds one and two do not coordinate the bound HMBPP. For instance, the side chain of Ser-265 in seed two does not extend inward toward the bound substrate, instead interacting with loop residues at the interfaces of D3 with D1 (Phe-12, Asn-43 and Thr-266). The local conformations of these residues are more consistent with those observed in [Fe4S4]2+(open, substrate-free) simulations of IspH. Coupled with the observed closing event in seed three, these findings demonstrate the presence of a barrier between the open and closed states, requiring the intramolecular interactions present in the substrate-free state to break in order to form interactions with bound HMBPP.
The second most populated cluster from [Fe4S4]2+/HMBPP(open, docked) IspH simulations, which corresponds to the most populated closed conformation, is used as a starting point for three additional 100 ns aMD simulations (henceforth referred to as [Fe4S4]2+/HMBPP(closed) simulations). Plots of the computed RMSD with respect to the [Fe3S4]+(open, substrate-free) IspH crystal structure for these simulations are marked by their lack of change, not deviating more than ∼3 Å from the closed conformations sampled in [Fe4S4]2+/HMBPP(open, docked) simulations (Figure 3c, Figure S1). Similar to what is seen in the [Fe4S4]2+/HMBPP(open, docked) aMD simulations, we note that HMBPP never fully reaches its ring conformation seen crystallographically [28]. From these simulations, it is evident that substrate-bound IspH, once folded around HMBPP, has less conformational space accessible to it and does not access open states.
In constructing principal component (PC) space from all [Fe4S4]2+(open, substrate-free) and [Fe4S4]2+/HMBPP(open, docked) simulations, as described in the Methods, we observe that the first two principal components account for 83% of the variance. Using Bio3D [40], the motions that correspond to movement along PC1 and PC2 are visualized (Figure S2) and are shown to correspond to opening and closing motions achieved through the hinge-like properties of the loops that connect D3 to D1 and D2 and D2 to D1 and D3, as suggested by Groll and co-workers [28].
All simulations ([Fe4S4]2+(open,substrate-free), [Fe4S4]2+/HMBPP(open,docked), and [Fe4S4]2+/HMBPP(closed)), as well as the coordinates from the [Fe3S4]+(open, substrate-free) and the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structures, are projected onto the PC space to assess how the simulations sample configuration space with respect to the crystal structures within this coordinate system (Figure 5). Viewing these projections, it is clear that the [Fe4S4]2+(open,substrate-free) simulations (Figure 5a) sample significantly greater conformational space than the [Fe4S4]2+/HMBPP(open,docked) and [Fe4S4]2+/HMBPP(closed) simulations (Figure 5b,c). While other local minima are present, the [Fe4S4]2+(open,substrate-free) simulations sample energy wells near both the open (PDB ID: 3DNF) and closed (PDB ID: 3KE8) crystal structures along PC1 but do not overlap with the latter, HMBPP-bound crystal structure. This finding suggests that the precise closing motions that accompany ligand binding are absent without HMBPP bound to IspH, despite the intrinsic ability of [Fe4S4]2+(open,substrate-free) IspH to sample alternative closed states.
Volume analysis of the states sampled in the [Fe4S4]2+(open,substrate-free) IspH simulations demonstrates the extent to which various open and closed states are sampled within this PC framework. Using the Pocket Volume MEasurer (POVME) program [41], the volumes of representative structures from the clusters generated from [Fe4S4]2+(open,substrate-free) aMD trajectories are obtained and given in Figure 5a. Using this algorithm, it is notable that the [Fe3S4]+(open, substrate-free) IspH crystal structure [23] has a binding pocket volume of 451 Å3, whereas the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure [28] has a volume of 6 Å3 (71 Å3, in the absence of HMBPP). Movement along PC1 generally accompanies a decrease in binding pocket size in the [Fe4S4]2+(open,substrate-free) aMD simulations (from 612 Å3 at the most negative values of PC1 to 319 Å3 at the most positive values, Figure 5a). The characteristics of these different pockets are probed later in this report.
Projection of the [Fe4S4]2+/HMBPP(open,docked) simulations onto PC space reveals a single, clear pathway for the transition between open and closed states (Figure 5b). Three minima are apparent in the projections, one centered near the substrate-free crystal structure and two near the HMBPP-bound crystal structures that differ slightly in the specific contacts made between the protein and ligand. The extent to which bound-HMBPP restricts IspH dynamics is highlighted from the projection of the [Fe4S4]2+/HMBPP(closed) simulations onto PC space. When simulated from a closed conformation, it is clear that bound-HMBPP effectively locks the protein in a closed conformation, unable to access open states—evident by a well present only around the closed, HMBPP-bound IspH crystal structure (Figure 5c).
Combining the three trajectories for each individual system simulated, we performed a root-mean-square fluctuation (RMSF) analysis to quantify the extent to which each residue fluctuates in the different systems (Figure 6). In the case of the [Fe4S4]2+(open,substrate-free) simulations (Figure 6, black curve), the fluctuations in D2 and D3 are slightly greater than what is seen in the [Fe4S4]2+/HMBPP(open,docked) simulations. These fluctuations are abolished when the simulations are started from a closed conformation with HMBPP bound ([Fe4S4]2+/HMBPP(closed) simulations).
Changes in various peptide dihedral angles (phi, psi and chi) typically accompany global changes in protein conformation [42], [43]. In other words, certain dihedral angles may select for specific conformations in proteins [42], [43]. Recently, McClendon et al. contributed a method that quantifies differences in probability distributions of protein dihedral angles between a reference and altered state of a protein by using an expansion of the Kullback-Leibler (KL) Divergence [42]. This application assigns a value for the “mutual divergence” of each residue—a measure of the extent to which the distributions of dihedral angles differ between the two states. Using the [Fe4S4]2+(open,substrate-free) simulation as a reference, we compute “mutual divergence” values upon substrate binding to IspH using the MutInf suite of programs [42], [44] in an attempt to isolate local changes in protein structure that give rise to globally different conformational ensembles between the [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) aMD simulations.
A visual representation of “mutual divergence” values is provided in Figure 7a, with the highest scoring residues shown in Table 1. We present these data together with measures of sequence conservation, computed as Shannon entropy [40], [45], [46], as both these metrics are suggested to highlight residues of functional importance [42], [45]. The link between sequence conservation and functionality is obvious—residues that are highly conserved are usually conserved for some purpose, e.g. to bestow certain structural features to a protein or to participate in catalysis. Similarly, residues whose conformations change dramatically upon some natural perturbation to the system, ligand binding in our case, are likely responsible for the functionality of that protein. Consequently, we propose that residues that both are highly conserved and display high “mutual divergence” upon ligand binding are critical to the structure and function of IspH.
Interestingly, five residues displaying higher levels of “mutual divergence” (Phe-63, Lys-64, Glu-65, Gly-66 and Asp-67) are located in a loop region in D1 and are neither conserved nor directly interacting with bound-HMBPP (Figure 7, Table 1). Arg-72 and His-74 are positioned at the opposite end of this loop region and form hydrogen bonds with the PPi tail of HMBPP in the [Fe4S4]2+/HMBPP(closed) simulations. From these observations, it can be reasoned that the conformations of Arg-72 and His-74, altered upon HMBPP binding, in turn disrupt the conformations of the residues at the end of the loop.
Most other residues with high “mutual divergence” can be characterized by one of two distinct environments in the protein: either (a) coordinating HMBPP when it is bound (e.g. His-42, His-124, Asn-223 and Ser-265); or (b) structurally flanking the thiolates that anchor the [Fe4S4]2+ cluster to the protein (as is the case for Phe-12, which is adjacent to Cys-13).
High mutual divergence is seen for residues that occupy the first coordination shell of HMBPP when it is bound. These residues assume different conformations based upon whether they are coordinating the substrate. For instance, both His-42 and Arg-72 from D1, His-124 from D2 and both Asn-223 and Ser-265 from D3 all assume different main and side chain dihedral angle distributions in the [Fe4S4]2+/HMBPP(closed) state compared to the [Fe4S4]2+(open,substrate-free) state. These differences derive from the reorientation of these residues about the PPi of HMBPP in order to participate in hydrogen bonds or salt bridges.
The other class of residues with high “mutual divergence” reside adjacent to the thiolates tethered to the [Fe4S4]2+ cluster. Phe-12 exemplifies this finding, in that it maintains altered φ/ψ angle distributions, contingent on whether HMBPP is bound (Figure S3). The case of Phe-12 suggests similar behavior may exist in other thiolate-adjacent residues. In inspecting the dihedral angle distributions of Thr-95 and Asn-194 in the [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) simulations (both having more modest “mutual divergence” scores of 0.77 and 0.32, respectively; Figure S3), it is evident that while the φ/ψ angle distributions are virtually identical for Thr-95, Asn-194 samples different distributions in HMBPP-free and bound states, much like Phe-12. Unlike Phe-12, however, the φ/ψ angles of Asn-194 are unimodal in the closed simulations, indicating that closed conformations require that Asn-194 maintain certain backbone dihedral angles. Indeed, when the [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) simulations are combined and clustered together into open and closed conformations, it is clear that Asn-194 samples entirely different psi angles, contingent on whether D2 and D3 are open or closed (Figure 7b).
Whereas the psi angle for Asn-194 in open states contributes to the residue's disordered secondary structure, as computed by STRIDE calculations [47], Asn-194 in all closed states is strictly α-helical with a mean psi angle of −34°. It is clear from these distributions that dihedral angles near −34° select for the closed conformations of IspH and contribute to the helicity of Asn-194, unseen in the open conformations that only exist in the ensemble of states sampled in [Fe4S4]2+(open,substrate-free) simulations.
Moving from the dihedral angle to the global structure of D3, it is evident that the helicity of Asn-194 is achieved via cranking motions that pull the helix, comprised of residues 195 to 207 and anchored by Asn-194, behind the [4Fe-4S] cluster in all closed states. This “crank” motion effectively compresses the D3 domain and also draws inward the loops needed to corral HMBPP into a closed active site. In contrast, Asn-194 samples states with no ordered secondary structure in [Fe4S4]2+(open,substrate-free) IspH simulations, while extended in the open conformation.
Clustering of the [Fe4S4]2+(open,substrate-free) IspH aMD simulations reveals dominant structures with substrate pockets of differing volumes and chemical environments. Using representative structures from each of the clusters, we investigate the druggability of these different pockets by performing solvent mapping with the FTMAP program [48].
Taking the fragment positions as they are docked by FTMAP into each representative structure from the [Fe4S4]2+(open,substrate-free) simulations, we synthesize information regarding where the docked fragments congregate by generating a probe occupancy map for IspH. Probe occupancy is highest at the pocket corresponding to the substrate-binding site (Figure 8a, Figure S4). In the more voluminous clusters, as well as the most dominant cluster, probes expand beyond the HMBPP-binding site at the iron, into a crevice between D1 and D3 (Figure 8a, Figure S4). This finding suggests that inhibitors capable of occupying this expanded pocket while locking the protein in a state that is more open with respect to the [Fe3S4]+(open, substrate-free) IspH crystal structure may provide a feasible route toward novel inhibitor design.
An unanticipated finding from solvent mapping concerns the side of IspH opposite the substrate-binding pocket. When the protein opens fully, as seen in the [Fe4S4]2+(open,substrate-free) simulations, the hinge-like quality of the interface between D3 and D1/D2 hyperextends, creating a druggable pocket found opposite the side of the HMBPP-binding site (Figure 8b, black rectangle; Figure S4). When the hinge is opened, this pocket occupies a POVME-measured volume of 330–500 Å3 and accommodates a variety of polar and nonpolar probes. This result, stemming from the opening motions intrinsic to substrate-free, [Fe4S4]2+ IspH, may provide an allosteric target for inhibitor design.
Application of the aMD method to sample conformational space in both [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) states of IspH increases our understanding of how HMBPP binding affects IspH structure and dynamics, as well as highlights alternative routes for the design of novel IspH inhibitors.
In regard to IspH dynamics, our [Fe4S4]2+/HMBPP(open,docked) aMD simulations are able to capture the closing event that accompanies ligand binding in two out of three simulations. In these simulations, residues in D1 that are needed to coordinate HMBPP (His-42 and His-74) are already properly positioned to interact with the pyrophosphate tail of HMBPP, whereas residues in D2 and D3 that coordinate HMBPP require domain motions to bring them in proximity of the substrate. Once D2 and D3 close around HMBPP, it is apparent from our PCA that IspH is unable to reopen, with the fluctuations of residues from D2 and D3 largely suppressed as these domains engage in multiple electrostatic and hydrogen bond interactions with HMBPP (e.g. His-124, and Ser-226). These observations underscore the suggestions by others that both electron addition to the substrate and changes in active site and substrate titration states are necessary, not only for catalysis, but also to alter the electrostatics in the active site to enable IspH opening and release of the catalytic product, IPP or DMAPP [24].
In contrast with the [Fe4S4]2+/HMBPP(open,docked) and [Fe4S4]2+/HMBPP(closed) systems, [Fe4S4]2+(open,substrate-free) IspH is much more flexible and thus able to access both closed states and conformations that open beyond what is seen in the [Fe3S4]+(open, substrate-free) IspH crystal structure. When closed in our simulations, projections of [Fe4S4]2+(open,substrate-free) IspH onto PC space show no overlap with the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure, indicating that substrate binding allows IspH to sample a closed state that is inaccessible in the absence of HMBPP.
In our simulations of the substrate-free state, IspH accesses both open and closed conformations. Since closed states preexist in the substrate-free ensemble, it is tempting to suggest that conformational selection (CS) [49] is responsible for ligand recognition in IspH. Following the logic of Sullivan and Holyoak [50], however, induced fit (IF) likely better describes the conformational changes occurring upon ligand binding since HMBPP cannot actually bind to the closed state that preexists in substrate-free IspH (due to occlusion of the active site by D2 and D3) [51]. We propose that ligand binding may still be described as a combination of CS and IF, where the ligand initially selects open conformations for formation of an encounter complex. Once initially bound, HMBPP induces closure of D2 and D3 via motions that are also intrinsic to IspH in the absence of ligand. This ligand recognition mechanism, drawing from both CS and IF, is not unique to IspH, but rather gives further support to the suggestions of others that ligand binding can contain elements of both CS and IF [52]–[54].
Returning to the structures observed in the [Fe4S4]2+(open,substrate-free) IspH aMD simulations, it is interesting that the active site volume is subject to significant fluctuations—largely due to the flexibility of the loop regions connecting D3 to D1 and D2 and, to a lesser extent, D2 to D1 and D3. These fluctuations are expected, as HMBPP likely binds initially to an enlarged binding pocket that may accommodate the expansive hydration shell expected for pyrophosphate-containing molecules [55] that HMBPP carries from solution into an encounter complex with IspH. The larger pocket stemming from the super-open state seen in the [Fe4S4]2+(open,substrate-free) simulations would allow for this initial complex to form. Given the presence of these larger pockets in our simulations and this mechanistic rationale, it is reasonable to hypothesize that a variety of differently sized ligands may also be accommodated in the binding pocket.
Combining these volume data for the [Fe4S4]2+(open,substrate-free) state with the results from our KL divergence analysis and FTMAP solvent mapping of IspH, we can build on the work of others [29]–[34] in suggesting a novel framework for future IspH inhibitor design. HMBPP binding to IspH can be regarded the first step in the catalytic process vital to most microbes for production of IPP and DMAPP. Preventing this binding event is thus the goal of competitive inhibitor development.
From our KL divergence analysis, we find that in addition to conserved residues that coordinate HMBPP upon its binding, residues that are adjacent to thiolate residues achieve high “mutual divergence” scores due to their distinct dihedral distributions when IspH is open and closed. Given its position adjacent to the fully conserved Cys-193, Asn-194 likely coordinates the hinge motions of D3 that give way to the necessary closing events that accompany HMBPP binding. Preventing the closing of the D3 hinge and, consequently, locking the Asn-194 backbone dihedrals in their disordered, open conformations may provide a novel mode of inhibiting IspH.
From our aMD simulations, two differing mechanisms for disrupting the hinge motions of D3 are apparent. The first targets the outward motion of D3 from the HMBPP binding site that creates an enlarged cavity that extends from the active site to the interface between D3 and D1 (Figure 8a). Either design of larger competitive inhibitors that interact with the apical iron and the D3/D1 interface or design of ligands that interact allosterically with the D3/D1 interface could successfully exploit the enlarged pocket on the active site side of IspH. Alternatively, the presence of an allosteric pocket opposite the side of the HMBPP binding site may be targeted for inhibitor design (Figure 8b, Figure S4). Both these proposed sites for inhibitor design are “hot spots” found by solvent probes with FTMAP. Noting that probe occupancy correlates well with sequence conservation as measured by Shannon entropy (r = 0.49) provides further support for these suggested modes of inhibition.
Given the documented difficulties of rational drug design for metalloproteins, notably from a computational perspective [56], allosteric sites that do not require a detailed description of metal binding (e.g. orbital interactions, polarization and charge transfer) are highly desirable if existent. Furthermore, it has been shown that perturbations to allosteric networks in redox-active metalloproteins may affect the redox potential of these proteins and, consequently, alter their activities [57]. These factors motivate us to include the different pockets revealed by aMD simulations, particularly those that may provide routes to noncompetitive inhibition, in future computer aided drug design workflows.
Using aMD simulations, we are able to capture the closing event that accompanies the binding of HMBPP to IspH when starting from the substrate-free crystal structure. Drawing from PCA and visual analyses of the different trajectories considered, we propose that ligand binding occurs via a combination of induced fit and conformational selection. We note that a single dihedral angle, the ψ angle in Asn-194, selects for either open or closed conformations of IspH, the latter being achieved via a crank motion that draws D3 inward to corral the active site. Furthermore, our aMD simulations reveal both an expanded active site pocket encompassing a crevice between D1 and D3, as well as an allosteric pocket between D1 and D3 on the side opposite the substrate binding pocket that may be utilized for the design of novel IspH inhibitors.
Since the questions under consideration in this study begin with open, substrate-free IspH protein, we use the [Fe3S4]+(open, substrate-free) IspH crystal structure from Rekittke, et al (PDB ID: 3DNF) as a starting point [23]. Applying insight from the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure from Gräwert, et al. (PDB ID: 3KE8) [28], we model the apical iron into the cluster by superposition.
Using the Amsterdam Density Functional program [58], a model [Fe4S4]2+ cluster is geometry optimized using broken symmetry density functional theory (BS-DFT) [59], [60] at the OLYP/TZP level of theory [61], [62]. With the Gaussian 09 suite of programs [63], we optimize the geometry of HMBPP and compute the electrostatic potentials of both geometry optimized HMBPP and the model [Fe4S4]2+ cluster using MK radii [64] at the HF/6-31G(d) level of theory. The antechamber program [65] in the AmberTools 13 suite of programs [66] is then used to apply the restrained electrostatic potential (RESP) procedure to derive point charges for use in MD simulations. In the case of the [Fe4S4]2+ cluster, parameters for nonbonded terms are taken from the AMBER GAFF force field [67], and bonds and angles between atoms are implicitly accounted for by harmonic restraints applied to these terms, using parameters from the [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structure [28]. For HMBPP, all force field parameters are taken from the AMBER GAFF force field [67]. All charge and nonbonded parameters, as well as, a more detailed discussion of the ligand parameterization process, are included in the Supporting Information.
Hydrogens are added using PDB2PQR [68], [69], with protonation states assigned using the PROpKa program [70]. In our setup, His-42 and His-124 are set to their imidazolium states, and Glu-126 is protonated. Following hydrogen addition, the protein systems are minimized for 2000 steps in the gas phase using the sander module in AMBER12 [66], to remove problematic steric clashes. The systems are solvated in a box of TIP3P waters [71] that extends 12 Å beyond the closest solute atom, with counterions added to enforce electroneutrality. Non-water bonds to hydrogen atoms are constrained using the SHAKE algorithm [72], while the O-H bonds in water are constrained using the SETTLE algorithm [73]. All protein force field parameters are taken from the AMBER ff99SB force field [74], while the ligand parameters referred to above are taken from the AMBER GAFF force field [67]. Subsequent 2000 step minimizations are performed (a) to relax the water with protein fixed by positional constraints, (b) to relax the protein with all waters constrained, and (c) relax the whole system. Following this minimization protocol, all systems are equilibrated at constant pressure and temperature (NPT) conditions for 1 ns, with the protein fixed by positional constraints. The pressure is regulated using the Berendsen barostat [75] with isotropic position scaling (ntp = 1) and a pressure relaxation time of 2.0 ps, while a Langevin thermostat [76] with collision frequency of 2.0 ps−1 is used to increase the temperature of the system from 0 to 300K. The protein constraints are then lifted and a subsequent 2 ns NPT equilibration is performed at 300K to verify the density of the system is reasonable and stable. The last equilibration step is performed at constant volume and temperature (NVT) for 5 ns at 300K to prepare the system for production MD simulations. All dynamics are conducted using the pmemd.cuda engine [66], [77], with Particle Mesh Ewald summations used for computing long-range electrostatic interactions and short-range nonbonded interactions truncated beyond a cutoff of 10 Å [78], [79].
Given current computational power, most MD simulations are limited to sampling timescales on the order of 10–1000 ns. Since many biomolecular processes, including, for example, protein folding, ligand binding, and cis/trans isomerization events, may occur on the order of milliseconds to days, enhanced sampling techniques that facilitate traversing of configuration space efficiently are often implemented to provide information about the relevant conformations of biomolecules [80], [81]. Accelerated molecular dynamics (aMD) simulations promote enhanced sampling of systems without the need for defining a reaction coordinate. In aMD simulations, when the potential energy of the system, V(r), is below a threshold energy level, E, a boost energy, ΔV(r), is applied to encourage exploration of other areas of phase space (Eq. 1). The parameter α modulates the aggressiveness of this boost by altering the depth of the wells in the modified potential.(1)
We employ the dual-boost implementation of aMD to boost both dihedral and total potential energy force field terms to promote side chain dihedral angle rotations and diffusive transitions, respectively [82], [83]. We set the parameters E and α for our systems by defining these variables for the dihedral and total potential energy components with respect to the number of residues in the system, Nres, and the number of atoms in the system, Natoms, respectively (Eq. 2–5):(2)(3)(4)(5)
Subsequent reweighting of the trajectory frames from the aMD simulations using a tenth-order Maclaurin series expansion allows us to extract canonical ensemble averages of the system (further details included in Text S2). Recently, both these methodologies for obtaining aMD parameters and reweighting aMD results were successfully applied to bovine pancreatic trypsin inhibitor (BPTI) to properly obtain the relative populations of relevant, low-lying energetic states [84]. For some semblance of statistics, 3×100 ns aMD simulations are performed on all systems explored in this study.
RMSD, RMSF, clustering, and dihedral angle analyses are all performed using the AmberTools 12 suite of programs [66]. Alignment procedures implemented in the RMSD and tRMSF calculations are performed with respect to the [Fe3S4]+(open, substrate-free) IspH crystal structure (PDB ID: 3DNF [23]), aligning to the backbone atoms of D1, as this domain is the most rigid in all simulations. Clustering analyses for each of the simulations use pairwise RMSD computed for Cα atoms between frames to divide the cumulative trajectories for each system simulated into eight clusters using the average-linkage algorithm [85].
Principal component analysis (PCA) reduces atomic fluctuations in the various trajectories into vectors that represent the dominant correlated motions present in the simulations [86], [87]. Since we want our PCA to assess how well the different simulations sample conformational space with respect to the [Fe3S4]+(open, substrate-free) and [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structures (PDB ID: 3DNF and 3KE8, respectively), we first align the two crystal structures using the STructural Alignment of Multiple Proteins (STAMP) procedure [88], as implemented in the VMD MultiSeq plugin [89], [90]. The indices of aligned residues are then used in subsequent PCA.
Principal component (PC) space is constructed from the three [Fe4S4]2+(open,substrate-free) simulations and three [Fe4S4]2+/HMBPP(open,docked) simulations. The trajectories for each set of simulations ([Fe4S4]2+(open,substrate-free), [Fe4S4]2+/tHMBPP(open,docked), [Fe4S4]2+/HMBPP(closed)) are then projected onto the first and second principal components. Additionally, [Fe3S4]+(open, substrate-free) and [Fe4S4]2+(closed, HMBPP-bound) IspH crystal structures are projected onto PC space to assess overlap between the simulations and these structures along the PC1 and PC2 coordinates. The modes that correspond to PC1 and PC2 are visualized using the Bio3D suite of programs [40].
We quantify differences in IspH structure upon ligand binding by applying the Kullback-Leibler (KL) Divergence expansion, also referred to as relative entropy, to assess differences in the distributions of φ, ψ, and χ dihedral angles in [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) ensembles generated by aMD simulations. To obtain the KL divergence for each residue, we first split the 3×100 ns sets of simulations for the [Fe4S4]2+(open,substrate-free) and [Fe4S4]2+/HMBPP(closed) systems into 6 sets of 50 ns to provide statistical robustness to the calculations. The MutInf program [42], [44] processes the dihedral angle distributions for each of these 50 ns blocks as computed by the g_torsion program from the GROMACS suite of programs [91], computing the KL Divergence for a specific dihedral angle using Eq. 6:(6)
In this equation, pi refers to the probability that a particular dihedral angle from the [Fe4S4]2+/HMBPP(closed) simulations falls into a specific range of torsional space, which has been divided into 12° bins. The term pi* is the corresponding probability that the same dihedral angle from the [Fe4S4]2+(open,substrate-free) simulation falls into the same bin. Combining the KL terms for each of the dihedral angles (φ, ψ, and χ's) of a given residue provides a value for the KL divergence of a specific residue:(7)
This value for the KL divergence of a given residue provides a measure of the difference between the dihedral angle probability distribution functions of the [Fe4S4]2+/HMBPP(closed) simulations with respect to the [Fe4S4]2+(open,substrate-free) reference simulations.
Using the Bio3D suite of programs, we compute the Shannon entropy [45], [46] according to equation 3 for all residues in the A. aeolicus IspH structure with a 22-letter alphabet, where the 20 amino acids are included, as well as a gap character ‘-’ and a mask character ‘X’ [40].(8)
After normalizing the Shannon entropy score, residues that are fully conserved assume the value 1, while residues with no conservation have a Shannon entropy of 0.
We employ FTMAP [48] to allow many drug-like, organic fragments to bind to representative structures from the dominant clusters from [Fe4S4]2+(open,substrate-free) aMD simulations. FTMAP utilizes a fast Fourier transform (FFT) algorithm to allow the organic probes to sample many positions along the protein surface. Prevalence of fragment hits along the protein surface signifies “hot spots” that correspond to potentially druggable pockets [48].
We measure the ability of residues in substrate-free IspH to bind FTMAP probes by first defining binding of the residue by the probe as existent if the distance between their respective heavy atoms is less than 5 Å. We then combine these binding results across all dominant clusters from the [Fe4S4]2+(open,substrate-free) aMD simulations and count the number of probes that bind each residue. Normalization of these data indicates the relative propensity of each residue to bind drug-like molecules [92].
|
10.1371/journal.pgen.1007409 | TGFβ signaling limits lineage plasticity in prostate cancer | Although treatment options for localized prostate cancer (CaP) are initially effective, the five-year survival for metastatic CaP is below 30%. Mutation or deletion of the PTEN tumor suppressor is a frequent event in metastatic CaP, and inactivation of the transforming growth factor (TGF) ß signaling pathway is associated with more advanced disease. We previously demonstrated that mouse models of CaP based on inactivation of Pten and the TGFß type II receptor (Tgfbr2) rapidly become invasive and metastatic. Here we show that mouse prostate tumors lacking Pten and Tgfbr2 have higher expression of stem cell markers and genes indicative of basal epithelial cells, and that basal cell proliferation is increased compared to Pten mutants. To better model the primarily luminal phenotype of human CaP we mutated Pten and Tgfbr2 specifically in luminal cells, and found that these tumors also progress to invasive and metastatic cancer. Accompanying the transition to invasive cancer we observed de-differentiation of luminal tumor cells to an intermediate cell type with both basal and luminal markers, as well as differentiation to basal cells. Proliferation rates in these de-differentiated cells were lower than in either basal or luminal cells. However, de-differentiated cells account for the majority of cells in micro-metastases consistent with a preferential contribution to metastasis. We suggest that active TGFß signaling limits lineage plasticity in prostate luminal cells, and that de-differentiation of luminal tumor cells can drive progression to metastatic disease.
| Prostate cancer is among the leading causes of cancer deaths in men. While treatments for localized disease are quite effective, once the cancer metastasizes five-year survival rates drop to below 30%. The transforming growth factor (TGF) ß pathway is frequently inactivated in prostate cancer, and reduced expression of TGFß pathway components is associated with more advanced disease. Mouse models have shown that deletion of TGFß pathway components, in the background of a tumor initiating mutation, accelerates progression to invasive and metastatic prostate cancer. We analyzed the outcome of combining a deletion of the TGFß type 2 receptor (Tgfbr2) with deletion of the Pten tumor suppressor gene, which initiates tumorigenesis in the mouse prostate. Deletion of Tgfbr2 results in increased expression of markers of stem cell function and basal cell markers, and increased basal cell proliferation. Since human prostate cancer has a primarily luminal cell phenotype, we deleted Pten and Tgfbr2 specifically in luminal cells. This results in progression to invasive and metastatic cancer that is not seen with deletion of Pten alone. Analysis of the cell types in these tumors reveals that Tgfbr2 mutant luminal tumor cells differentiate to an intermediate cell type that has both basal and luminal characteristics, and also to basal cells. Analysis of early metastatic lesions suggests that the de-differentiated intermediate cells may be the drivers of metastasis.
| Prostate cancer (CaP) is the second-leading cause of cancer deaths in men [1], with more than 161,000 cases and nearly 27,000 deaths predicted in the US in 2017 (http://seer.cancer.gov/statfacts/html/prost.html). Although five-year survival rates for patients with localized disease are high, for patients with distant metastases five-year survival is below 30%. Patients initially respond favorably to therapies based on androgen depletion reflecting the androgen-dependence of these tumors. However, anti-androgen therapy becomes ineffective as prostate cancers progress to a castration resistant state [2].
Transforming growth factor (TGF) ß family ligands assemble a complex of type I and type II receptors, resulting in type I receptor activation [3–5]. The activated type I receptor phosphorylates Smad proteins, primarily Smad2 and Smad3 for TGFß. Phosphorylated Smads bind Smad4 and accumulate in the nucleus where they regulate target gene expression [6]. In many cell types, including epithelial cells, TGFβ signaling via Smad2 and Smad3 promotes a G1 cell cycle arrest preventing uncontrolled cell proliferation. Thus, TGFß signaling frequently plays a tumor suppressive role [7, 8]. The TGFβ signaling pathway is disrupted by mutation or loss of expression of pathway components in many human cancers, including CaP [9–11]. Reduced expression of the TGFβ type I and type II receptors (encoded by the TGFBR1 and TGFBR2 genes) is associated with increased Gleason score and decreased survival, and reduced SMAD4 expression is also found in advanced human CaP [9, 12–14].
Inactivation of the PTEN tumor suppressor is found in more than 30% of primary human prostate tumors and in approximately 60% of CaP metastases [15–18]. In mice prostate specific deletion of the Pten tumor suppressor is one of the more robust models of prostate cancer. All mice develop prostate intraepithelial neoplasia (PIN) by six weeks of age that rapidly progresses to high grade PIN (HGPIN). In this model, HGPIN eventually develops to locally invasive cancer, but this occurs relatively slowly, and metastases are rarely seen [19–23]. Combining additional mutations with Pten deletion accelerates the progression from HGPIN to invasive poorly differentiated adenocarcinoma (PDA) and results in metastasis. Deletion of either Smad4 or Tgfbr2, to inactivate TGFß signaling in prostate epithelial cells, drives invasion and metastasis in the background of a Pten deletion [20, 24]. Combining other oncogenic hits with a Pten deletion has also been shown to increase invasion and metastasis. For example expression of a mutant Kras together with Pten deletion resulted in metastasis to lung and liver [25]. Although inactivating mutations in the APC tumor suppressor gene are less frequent in human CaP than in colon cancer, for example, nuclear β-catenin is observed in the majority of cases of advanced human CaP suggesting that this pathway is frequently de-regulated [26]. More recent genomic analyses of human prostate cancer suggest that APC mutations are found at a higher frequency than originally thought, with these being primarily nonsense and frame-shift mutations [27]. In mouse models, deletion of the Apc gene in mouse prostate epithelium results in HGPIN [28], which rapidly progresses to invasive cancer when combined with loss of the Tgfbr2 gene [19].
The two major epithelial cell types within the prostate are luminal and basal cells, and there is evidence for stem cell pools within both populations. Lineage tracing in mouse prostate suggests that multipotent progenitors present within the basal cell population may give rise to unipotent basal and luminal progenitors during post-natal development [29]. More recent work with organoids suggests that the luminal population also contains stem cells capable of regenerating both basal and luminal cell types, albeit at a much lower frequency than basal cells [30]. The majority of cells in human prostate cancer have a luminal phenotype, although rare basaloid carcinomas are seen [31]. This is consistent with the idea that luminal cells are the cell of origin for most of human CaP. Prostate luminal cells have been shown to have the potential to be cancer initiating cells, and it has been suggested that castration resistant cells within this population may be the tumor initiating cells [32–34]. However, a number of studies have also suggested that basal cells can be a cell of origin for prostate cancer [32, 35, 36], and a basal-like stem cell gene expression signature is associated with aggressive cancer [37]. Despite the differing requirements for the androgen receptor (AR) in basal and luminal cell survival, loss of Pten can overcome the effect of AR deletion, allowing luminal-like tumors to develop from both cell types [38]. Deletion of Pten may limit androgen-responsive gene expression by modulating AR function, decreasing the ability of the AR to promote differentiation [39]. Recent work has suggested that lineage plasticity in human CaP may allow for a fraction of tumor cells to escape the effects of anti-androgen therapies, allowing the tumor to recur as a therapy resistant cancer despite initial regression of the primary tumor [40]. In mouse prostate, the deletion of Rb1 together with Pten results in increased lineage plasticity and metastasis with a gene expression signature reminiscent of human neuroendocrine CaP [41]. This may be one mechanism to explain the eventual failure of androgen ablation, suggesting that a better understanding of linage plasticity, rather than the initial cell of origin, is of importance.
Disruption of TGFß signaling increases proliferation and overcomes senescence in prostate tumors initiated by loss of either Pten or Apc, and accelerates progression to a widespread invasive phenotype [19, 20, 24]. To further understand the cellular changes induced by loss of Tgfbr2 expression in mouse prostate tumors we used transcriptional profiling of these two tumor models. This analysis reveals increased expression of genes associated with stem-like properties in these highly proliferative tumors. In addition we observed increased numbers of basal cells, which make up much of the invasive cancer, and we show that loss of Tgfbr2 primarily affects basal cell proliferation. However, deletion of Tgfbr2 and Pten specifically in luminal cells resulted in metastatic cancer and extensive de-differentiation to an intermediate cell type expressing both basal and luminal markers. Analysis of early lung metastatic lesions indicates that these de-differentiated prostatic cells may preferentially generate metastases, suggesting that increased lineage plasticity in the absence of TGFß signaling may contribute to an aggressive disease phenotype.
We have previously shown that deletion of Tgfbr2 from mouse prostate epithelium accelerates tumor progression in the background of either a Pten or Apc null mutation [19, 20]. To examine the status of the TGFß signaling pathway in human CaP we analyzed a panel of high grade (primarily Gleason score 8–10) human tumors by IHC for SMAD4 and active phosphorylated SMAD2 (pSMAD2). We chose these as pSMAD2 provides a readout for receptor activity (the combination of both Type I and Type II), and SMAD4 is required to complex with the receptor activated SMAD. More than half of the 38 tumors analyzed had reduced expression of SMAD4 and a slightly higher number had low pSMAD2 signal (S1 Fig). These proportions are similar to that seen for activation of AKT in this set of samples, and is consistent with previous reports showing reduced expression of both Type I and Type II receptors, and of SMAD4 in human CaP [9–14, 24].
To begin to examine how loss of TGFß signaling drives prostate tumor progression we performed transcriptome profiling on normal wild type prostate and on tumors from mice with prostate-specific deletion of Apc, or Pten, with or without deletion of Tgfbr2. For the Apc mutants, we isolated tumors at 36 weeks of age, when they had extensive adenosquamous HGPIN, and at 20–24 weeks from the Apc;Tgfbr2 mutants when they first showed adverse signs of tumor burden. For the Pten model, we analyzed two different ages each for both single (Pten) and double (Pten;Tgfbr2) mutants. Tumors were isolated from both groups at 8 weeks of age when the predominant phenotype in both is HGPIN, and at 22 weeks of age from the Pten single mutants, which still have primarily HGPIN. Since double mutants do not generally survive beyond ~15 weeks, we isolated tumors at 11–14 weeks when they first showed signs of excess tumor burden. We anticipated that this combination would provide a comparison of progression in the more relevant Pten based model as well as comparison to a second model in which the tumor-initiating event is different. As shown in Fig 1A, principle component analysis (PCA) separated the samples into three broad groups, with the wild type samples clustering tightly together. The other two groups primarily appeared to cluster by phenotype, with less separation between the Pten and Apc based tumors. Thus, the eight-week Pten;Tgfbr2 double mutants clustered with the 8- and 22-week Pten single mutants, all of which have HGPIN as the primary phenotype. Unsupervised hierarchical clustering based on the 1000 most variable genes revealed a similar pattern (Fig 1B), with the older Pten;Tgfbr2 and Apc;Tgfbr2 tumors grouped together, although there are clearly some differences between them. Again, eight-week tumors, whether Pten single mutant or Pten;Tgfbr2 double do not separate into distinct groups and also appear similar to the older Pten single mutants. To examine which genes are differentially expressed among the different groups, we performed all pairwise comparisons between the seven groups of samples, and identified genes as differentially expressed if there was a log2-fold difference of greater than +/-1.0, with an adjusted p-value of < 0.0001. From these comparisons, we first focused on differences between each group of tumors and the wild type prostates. Despite the relatively stringent cut-off these comparisons revealed large numbers of genes that were differentially expressed between wild type and each tumor group (S1A and S1B Table). For example, 4079 genes passed this cut-off when comparing the older Pten;Tgfbr2 tumors to wild type, with somewhat lower numbers from the comparisons of other groups to the wild type (Fig 1C and 1D). Comparison of the 8-week Pten;Tgfbr2 tumors and Pten single mutants at both ages revealed largely overlapping gene expression changes, although changes in these samples were quite variable, possibly due to the mix of normal and tumor tissue at this early age. In the older Pten;Tgfbr2 tumors a large fraction of the changes in gene expression were not seen in these other three groups (Fig 1C). Given the lack of difference between the two 8-week groups, and the variability among these early tumors we focused on the older Pten single and double mutants, and compared them to the Apc and Apc;Tgfbr2 mutant tumors. Almost 900 genes were differentially expressed in all four tumor types compared to wild type, with a slightly larger number shared only between the two double mutant tumors (Fig 1D). GO analysis revealed that two of the top three enriched biological process terms in the genes changing in only the two double mutants were epithelial to mesenchymal transition (EMT) and cell cycle, consistent with the double mutants being more proliferative and invasive (Fig 1E, S1C Table). GSEA analysis comparing wild type to 22-week Pten and to 11-week Pten;Tgfbr2 mutants, and comparing these two groups of tumors to each other showed increasing enrichment for E2F targets with progression to the more aggressive tumors, but suggested that the EMT signature was already present in the Pten single mutant tumors (Fig 1F).
The previous analyses suggest that with the transition from HGPIN to invasive cancer there is an enrichment for gene changes associated with cell cycle progression and an underlying enrichment for EMT in the HGPIN tumors. Comparison of increases in gene expression relative to the wild type samples revealed a similar pattern to that seen with all gene expression changes, and enrichment for GO terms associated with cell cycle, EMT and regulation of transcription (Fig 2A and 2B and S1C Table). In contrast, many fewer genes were consistently decreased in expression compared to the wild type, and GO analysis did not reveal significant enrichment of any terms among these groups (Fig 2C). Analysis of ENCODE and ChEA transcription factor binding data suggested an enrichment for genes bound by transcription factors associated with stem cell function, such as SOX2, KLF4, NANOG and SALL4 among genes with increased expression in the tumors compared to wild type (Fig 2B). Scanning the RNA-seq data we noticed that a number of markers associated with stem cell-like function in cancer were increased in the double mutant tumors compared to wild type and to the single mutants (Fig 2D). To validate these changes, we focused on the wild type, Pten single and Pten;Tgfbr2 double mutant tumors, since the Pten null model is more relevant to human CaP. Analysis of a distinct set of tumors of these genotypes by qRT-PCR demonstrated increased expression of a panel of cancer stem cell markers in the Pten single mutants compared to wild type, with further increases in the double mutant tumors (Fig 2E). Several gene sets that are associated with more aggressive tumors or with stem-like features of human tumors have been identified. For example, the CIN70 gene set, which is associated with more aggressive tumors that have higher levels of chromosome instability includes a number of genes with mitotic functions, including FOXM1 [42]. This gene set may also be an indication of highly proliferative tumors. GSEA analysis with the CIN70 set revealed increasing enrichment in Pten null and Pten;Tgfbr2 double null tumors (Fig 2F). Since FOXM1 targets were also enriched in the tumor samples based on ENCODE ChIP-seq data (Fig 2B), we also compared to a FOXM1 target gene set identified by ChIP-seq [43]. This gene set showed enrichment in the tumor samples, and further enrichment comparing double null to single null tumors (S2 Fig). Similar patterns were seen with other data-sets, including an embryonic stem cell (ESC) associated signature identified in a pan cancer analysis as being enriched in aggressive tumors [44] (S2 Fig). Although there is some overlap amongst the genes in these three data-sets, there are also large numbers of genes that are distinct to each, particularly when comparing the ESC and FOXM1 data-sets (S2B Fig). To confirm the changes seen in the RNA-seq data, we again performed qRT-PCR on wild type prostates and Pten single and double null tumors, probing a selection of the genes present in these three data sets. The majority of genes tested were increased in the Pten single null tumors and most increased when comparing double to single mutant tumors (Fig 2G). In this analysis we examined both Foxm1 and Mybl2, as these two transcription factors have been shown to co-regulate a number of the mitotic target genes that are present in these data sets. Both showed increased expression in the double mutant tumors, and similar patterns were seen with known target genes, such as Ccnb1. Together these analyses suggest that loss of TGFß signaling in the background of a Pten null prostate tumor further activates pro-proliferative and stem-like gene expression programs, which may already be initiated by the loss of Pten.
Among the most highly expressed and significantly up-regulated genes in the 11-week old Pten;Tgfbr2 tumors were Krt5 and Krt14, two basal cell specific keratin genes. In addition, expression of the basal cell-enriched p53 paralog, Trp63, was significantly increased in Pten;Tgfbr2 tumors. Comparing expression of three genes each that are representative of basal, luminal or neuroendocrine cells showed clear increases in expression of the basal genes in the double null tumors (Fig 3A). Luminal specific genes were generally lower in the Apc-based tumors. There was some apparent increase in neuroendocrine genes in the 8-week tumors, but it should be noted that expression of these genes is very low and quite variable, and the changes did not reach statistical significance. Increased expression of Trp63 and the two basal keratins in Pten;Tgfbr2 tumors was validated by qRT-PCR, and for at least some luminal-enriched genes we observed a decrease in expression (Fig 3B). The basal cell population has been thought to contain stem cells that can regenerate the entire prostate, although recent evidence suggests that such stem cells may also reside in the luminal population [29, 30]. Analysis of luminal and basal/progenitor-like cell populations isolated from normal mouse prostate, based on expression of fluorescent marker transgenes, identified gene expression signatures associated with each [45]. Comparison of our RNA-seq data to these two data-sets revealed enrichment of the basal/progenitor gene set in the Pten;Tgfbr2 tumors and an enrichment of the luminal signature in Pten single null tumors when compared to the double nulls (Fig 3C). Normal prostate ducts in the mouse are primarily composed of a single layer of luminal epithelial cells with an incomplete basal layer surrounding it. Immunofluorescent detection of basal (Krt5) and luminal (Krt8) keratins showed that this organization is not altered in the absence of Tgfbr2 and, as previously reported, there is no significant phenotype associated with loss of Tgfbr2 alone (Fig 3D) [19, 20]. Where HGPIN was observed in the Pten single mutant tumors, the ducts become filled with luminal cells, a pattern that was also seen in HGPIN ducts in the double mutant tumors. However, in the Pten;Tgfbr2 tumors, a large proportion of the invasive cancer consisted of cells that were strongly positive for Krt5, with Krt8 cells intermixed (Fig 3D). To better characterize the cell types present in the Pten;Tgfbr2 tumors we examined expression of additional markers. The androgen receptor (AR) was highly expressed in luminal cells in intact ducts that were also strongly Krt8 positive. AR was present at a lower level in the Krt5 positive invasive cancer, and was clearly absent from stromal cells (Fig 3E). Trp63 expression was largely absent from cells that stained strongly for Krt8, but was clearly present in Krt5 positive cells, consistent with these being basal cells (Fig 3F). Thus deletion of Tgfbr2 in the context of a tumor initiating mutation results in an increase in the basal cell population together with a rapid transition to invasive cancer.
To better assess the increase in basal cell numbers in the Pten;Tgfbr2 tumors we stained a series of Pten and Pten;Tgfbr2 null samples for both Krt5 and Krt8 and quantified the proportion of cells in each. As shown in Fig 4A, in Pten null tumors HGPIN was composed of more than 90% luminal cells at all ages examined. Even in invasive cancers in older Pten null tumors, most of the invasive cells were luminal (Fig 4A and S3 Fig). Similarly, in Pten;Tgfbr2 tumors HGPIN was primarily luminal, with no significant difference in the proportion of basal cells compared to HGPIN in the Pten null. In contrast, in invasive regions in the Pten;Tgfbr2 tumors there were more than twice as many basal as luminal cells (Fig 4A). While performing this analysis we also examined the number of cells that were positive for both basal and luminal keratins. These cells were extremely rare in HGPIN in Pten;Tgfbr2 tumors, but were slightly increased in the invasive cancer (Fig 4B). However, it should be noted that even in the invasive cancer these dual positive cells were still only a very small fraction of the total.
Given the increased frequency of basal cells, we tested whether this was due to a specific increase in basal cell proliferation. We examined a panel of tumors for expression of both Krt5 and Krt8, together with Ki67 to identify actively proliferating cells within each population. In Pten;Tgfbr2 tumors, the invasive cancer was highly proliferative and was largely Krt5-high and Krt8-low, whereas fewer cells stained positive for Ki67 in intact HGPIN ducts that were luminal (Fig 4C). Closer examination of regions of invasive cancer in which both basal and luminal cells were present suggested that within the invasive cancer more Krt5-high cells were proliferating than Krt8-high cells (Fig 4D). When we quantified this across both genotypes and a range of ages and phenotypes, there was little difference between proliferative rates in basal and luminal cells in HGPIN, except in early Pten;Tgfbr2 tumors, where there was an increased rate in the basal cells compared to the luminal, although these numbers were quite variable (Fig 4E). The most striking difference was the high proliferation rate of basal cells in invasive cancer in the Pten;Tgfbr2 tumors compared to those in the Pten or in intact HGPIN ducts in the same animals, consistent with a preferential effect of TGFß signaling on basal cell proliferation. It is also possible that luminal to basal de-differentiation contributes to the increase in basal cells in the invasive tumors, although these data clearly show a role for increased basal cell proliferation.
As discussed before, the majority of human CaP exhibits a luminal phenotype. However, there has been some interest in basal and luminal subtyping of human CaP by the identification of gene expression signatures that correlate with disease progression. Analysis of human CaP data using a 50 gene set that had been shown to distinguish the four types of human breast cancer, identified three human CaP subtypes that matched the luminal A, luminal B and basal subtypes from this gene set [46, 47]. Comparison of our RNA-seq data with this gene set suggested a preferential enrichment for increased expression of genes associated with the luminal B subtype, rather than the basal signature, in the double null tumors (S4A Fig). Comparison to a different 37 gene signature, that again identified two luminal subtypes and one basal human CaP subtype [48], suggested increased expression of the genes indicative of the more aggressive of the two luminal signatures in the Pten;Tgfbr2 double null tumors (S4B and S4C Fig). Thus, although there is a significant basal cell component to these tumors, they appear to contain links to gene expression signatures indicative of aggressive luminal tumors.
Human prostate cancer cells typically have a luminal phenotype, whereas in the PbCre4 model the Cre is expressed in both basal and luminal cells and deletion of Tgfbr2 appears to activate basal cell proliferation. Therefore, we examined the effect of deleting Pten and Tgfbr2 specifically in luminal cells. To do this we combined the conditional alleles of Pten and Tgfbr2 with a lineage tracing marker (mTmG) and a tamoxifen inducible Krt8-Cre transgene [49, 50]. This combination allows for deletion specifically in luminal epithelial cells, and an ability to permanently track which cells have undergone Cre activation by examining expression of GFP following excision of the STOP cassette. To induce recombination driven by the Krt8-CreERT2 we initiated tamoxifen treatment by gavage at four weeks of age and examined tissues by IF detection of the GFP lineage tracing allele to determine whether there was significant recombination in the prostate. Co-staining for GFP together with Krt8 and Krt5 revealed a clear overlap between the GFP signal and Krt8, whereas basal cells lacked detectable GFP expression (Fig 5A). From examination of 59 GFP positive normal ducts and 115 ducts with focal PIN from six mice at 2–6 weeks post-tamoxifen we did not identify any clearly GFP positive basal cells. In some cases where GFP positive luminal cells were directly adjacent to a basal cell it may be difficult to exclude partial co-localization of the GFP and Krt5 signal. However, we never observed GFP signal close to a basal cell without adjacent luminal cells that were clearly GFP positive (see Fig 5A). Furthermore, by confocal analysis we were unable to detect any GFP signal that was clearly present in a basal (Krt5 positive) cell (Fig 5B and S5A Fig). Co-staining of prostates six weeks after tamoxifen for GFP and pAkt showed that Akt was activated in GFP positive cells, consistent with loss of Pten in these cells (S5B Fig). Similarly, Tgfbr2 expression was reduced in GFP positive cells compared to adjacent GFP negative cells (S5C Fig). Together with previous analyses that have used this transgene to drive luminal specific deletions [29], this suggests that if deletion occurs in cells other than luminal cells it is extremely rare.
To examine the initial phenotype resulting from deletion of Pten and Tgfbr2 in luminal cells we examined GFP expression at 5–6 weeks post-tamoxifen treatment. At this stage we observed a range of phenotypes from normal tissue to HGPIN, with all HGPIN ducts containing GFP positive cells. The proportion of GFP positive cells in ducts with HGPIN was significantly higher than in non-phenotypic ducts, with the mean proportion of GFP positive cells per duct being 25% at this stage, rising to almost 70% for HGPIN ducts (S6A Fig). Among 143 ducts examined 106 had GFP positive cells, with none of the GFP negative ducts exhibiting any phenotype, whereas, all HGPIN ducts were GFP positive (S6B and S6C Fig). In ducts with HGPIN, the GFP positive cells were strongly positive for Krt8, and basal cells at the edges of the ducts did not have GFP signal (Fig 5C). We next followed twelve Pten;Tgfbr2 Krt8-CreERT2 mice after tamoxifen treatment beginning at 4 weeks of age to examine survival of these mice. Of the twelve, eleven had to be euthanized for tumor burden by 20 weeks post-tamoxifen treatment. All eleven had large prostate tumors with extensive invasive adenocarcinoma, and were euthanized for bladder obstruction, with a median survival time of about 97 days post-tamoxifen treatment (or about 125 days of age) (Fig 5D and 5E). The final mouse was euthanized at 24 weeks after starting treatment and found to have HGPIN, but no invasive cancer. Examination of individual lobes suggested that the ventral prostate was most severely affected, with the anterior having the mildest phenotype (S6D Fig). We also tested different dosing schedules for the tamoxifen, which resulted in similar terminal phenotypes, but with much more variable lag times and lower penetrance (S6E and S6F Fig). As a comparison we treated mice with only conditional Pten alleles using the more robust treatment protocol and found that these mice had extensive HGPIN, but none had progressed to invasive cancer even by 40 weeks after treatment. Taken together, these data suggest that deletion of Pten and Tgfbr2 together specifically in luminal cells results in rapid progression to invasive cancer that is not seen with deletion of Pten alone.
We next examined the cell types present in HGPIN and invasive cancer from the Krt8-Cre driven Pten;Tgfbr2 tumors. As shown in Fig 5C, prostate ducts with HGPIN from mice at 4–8 weeks after tamoxifen induction were exclusively Krt8 positive, with rare basal (Krt5 positive) cells that lacked GFP signal at the edges. However, when we examined invasive cancer from mice that had been euthanized for tumor burden, there was a considerable overlap between Krt5 and GFP signals (Fig 6A). More surprisingly, much of the Krt5 positive invasive cancer was also positive for Krt8, suggesting co-expression of both basal and luminal keratins. Closer examination of more isolated cells, where apparently overlapping keratin signals were unable to come from adjacent cells showed that both Krt5 and Krt8 were indeed expressed within the same cell, and that both signals were surrounded by the membrane targeted GFP recombination marker (Fig 6B). In advanced invasive cancer (all of which was GFP positive) we observed three distinct classes of keratin staining: Cells with high Krt8 and very low or absent Krt5 signal, as would be predicted from the Krt8-Cre driver, cells with high to intermediate levels of both Krt5 and Krt8, and those that had high Krt5 with little or no Krt8. Quantification of the number of basal (Krt5 high;Krt8 low/absent), luminal (Krt5 low/absent;Krt8 high) and dual positive cells from multiple tumors showed a significant increase in basal and dual positive cells in the invasive cancer compared to HGPIN, with a concomitant decrease in luminal cells (Fig 6C). Examination of invasive cancers from the Krt8-Cre model did not reveal any overt squamous differentiation, as seen in Apc;Tgfbr2 null mouse prostate tumors, for example (S7A–S7C Fig) [19]. Gene expression analysis of PbCre Pten;Tgfbr2 tumors indicated extremely low expression of neuroendocrine markers, including synaptophysin. Similarly, analysis of synaptophysin in Krt8-Cre invasive tumors suggests that neuroendocrine differentiation in this model is extremely rare (S7D Fig). Thus it appears that luminal differentiation to dual positive and basal cells is the main effect of Tgfbr2 deletion in this model. We next examined proliferation by staining for Ki67 together with Krt5 and Krt8. As in the PbCre4 model, the majority of proliferating cells were basal rather than luminal in invasive cancer (Fig 6D and 6E). Proliferation in luminal cells increased from HGPIN to invasive cancer, with basal cell proliferation being even higher (Fig 6D). However, the proliferation rate of dual positive cells was not increased relative to luminal cells and was significantly lower than in basal cells, suggesting that the increase in their numbers is not due to rapid proliferation of rare dual positive cells present in HGPIN or early invasive tumors.
In the lungs of six of seven Krt8-CreERT2 Pten;Tgfbr2 mice examined we identified metastases, which were positive for GFP. In larger metastases, we observed a mix of basal, luminal and dual positive cells that was similar to that seen in the primary tumors, as evidenced by staining for Krt5 and Krt8 (Fig 7A). Since the Krt8-CreERT2 driver is not specific to the prostate we also examined potential lung metastases for AR and Trp63. As shown in Fig 7B, GFP positive cells were positive for AR, with higher AR expression generally seen in Krt8+ GFP+ cells. The majority of GFP+ cells clearly expressed AR, with higher levels in Krt8+ cells, whereas no AR expression was detected in adjacent lung tissue (S8 Fig). A proportion of GFP positive cells within the lung were also Trp63 positive, and this signal was primarily distinct from Krt8+ cells consistent with these being basal cells (Fig 7C). This marker analysis indicates a similar expression pattern in lung tumors to that seen in the primary invasive prostate cancer. Thus metastasis from prostate is the most likely explanation for these GFP+ AR+ tumors in lung, although we cannot definitively rule out that some lung tumors are not derived from prostate metastases. While examining the cell types present in GFP positive lesions in the lungs of these mice, we noticed that most cells in lung micro-metastases were strongly positive for both Krt5 and Krt8, consistent with these smaller metastases being primarily composed of dual positive cells (Fig 7D). We previously reported frequent micro-metastases to the lungs of mice with prostate specific Pten and Tgfbr2 deletion using the PbCre4 driver [20]. In the primary tumors of these animals dual positive cells were very rare compared to the relatively high frequency seen in the Krt8-Cre model (see Fig 4B). Given the different balance of cell types in these two models, we examined the cell types present in lung micro-metastases from the PbCre4 tumors. As shown in Fig 7E, lung metastases frequently contained a large proportion of cells with high levels of both basal and luminal keratins. When we quantified this across 42 independent lung micro-metastases, we found that almost 40% of the cells examined were dual positive, with a similar proportion being basal, and only 20% luminal (Fig 7F). More than 90% of the metastases examined contained dual positive cells, with the remaining three metastases composed entirely of basal cells (Fig 7F). However, as we only examined a single section for each, it remains possible that other cell types were missed. Since the frequency of dual positive cells in the primary tumors in the PbCre4 model is very low, these data suggest that micro-metastases are either preferentially derived from this rare population, or that the cells that generate metastases have increased lineage plasticity once they seed the lung.
Given that de-differentiated dual positive cells in these Pten;Tgfbr2 CaP models appear to drive invasion and metastasis, we examined whether cells with expression of both basal and luminal keratins were present in human CaP. We examined a panel of human CaP samples by IF for both Krt5 and Krt8. Of the 138 informative samples included on three tissue microarrays (21% Gleason 5–6, 68% Gleason 7–8, 11% Gleason 9–10), only one sample had tumor cells that were clearly dual positive (Fig 8A). Strikingly, this was the only prostate sample with marked atrophic changes consistent with prior androgen ablation therapy. Although it has not been common practice to undergo prostatectomy after androgen deprivation therapy (ADT), we obtained samples from ten additional patients who had had radical prostatectomy after some form of ADT. However, the types of drugs administered and their dosages and durations were quite variable. The range of beginning ADT to time of radical prostatectomy was between 2 weeks and 3 years (median: 19 months). Of these ten, two more contained dual positive cancer cells (Fig 8B). Although these dual positive cells were quite rare, the fact that we identified a small number of human cancers with tumor cells co-expressing both basal and luminal keratins is consistent with the idea that this is a potentially important cell type in human CaP. This notion is also supported by recent work showing increased lineage plasticity in Pten null tumors lacking both Rb1 and Trp53 [41].
Here we show that in the background of a Pten null mutation in mouse prostate, deletion of Tgfbr2 results in increased basal cell proliferation as tumors become invasive. Analysis of micrometastases from these mice suggests that cells expressing both luminal and basal keratins are more metastatic. Deletion of Pten and Tgfbr2 specifically from luminal cells also causes a rapid onset of invasive and metastatic cancer that is accompanied by a transition to a less differentiated cell type. We suggest a model in which TGFß signaling maintains a differentiated cell phenotype, and that loss of signaling may allow for de-differentiation to a more invasive and metastatic cell type.
Disruption of TGFß signaling either by Smad4 or Tgfbr2 deletion in the background of a prostate specific tumor initiating mutation results in rapid progression to invasive and metastatic cancer in mouse models [20, 24]. Transcriptome analysis suggested that this was accompanied by increased expression of basal cell specific genes, including Trp63 and basal keratins. This increase appears to be due to increased basal cell numbers in these tumors following Tgfbr2 deletion, and we show that a major effect of TGFß signaling in this context is to limit proliferation of basal cells. Given the results from the luminal-specific deletion model, some de-differentiation of luminal to basal cells may also contribute, although this is likely a minor component in the PbCre4 driven tumors. Previous work has also suggested a role for TGFß signaling, in conjunction with the Notch pathway, in limiting basal cell proliferation [51]. Analysis of gene expression data suggests that in comparison to the Pten null tumors, disruption of TGFß signaling primarily affects expression of gene sets associated with highly proliferative, aggressive cancers. Although gene sets associated with EMT or invasion are enriched in the tumors compared to wild type prostate, it appears that such invasive gene expression signatures are already enriched by deletion of Pten alone. Although deletion of Pten from mouse prostate results in invasive cancer, it does so relatively slowly. Regions of micro-invasion may be detectable early, but large locally invasive cancers are not generally detected in the Pten model until 30–50 weeks of age, and metastases are extremely rare [19, 20]. Thus, the combination of a local micro-invasive phenotype initiated by Pten loss, together with increased proliferation due to loss of Tgfbr2 drives rapid progression to invasive cancer. However, this model is complicated by the loss of Pten and Tgfbr2 from both basal and luminal cells, together with the dramatic effect of the Tgfbr2 deletion specifically on basal cell proliferation.
Human CaP primarily consists of malignant luminal epithelial cells, with other subtypes, including basaloid and neuroendocrine cancers being much less frequent [31]. Recent evidence has renewed interest in understanding the neuroendocrine phenotype, as it may be part of the mechanism by which tumors escape the effects of anti-androgen therapy [40, 41]. Despite the predominance of the luminal phenotype in human CaP, there is considerable evidence that both basal and luminal cells can be the cell of origin for cancer [32–36]. To more accurately model tumors in which a single cell type is the target for oncogenic transformation, we used a luminal-specific Cre driver [50]. Deletion of Pten using a tamoxifen inducible Krt8-Cre transgene that is active specifically in luminal cells resulted in HGPIN, consistent with previous reports [32]. When combined with a conditional Tgfbr2 allele we observed a rapid progression to invasive and metastatic cancer, which was not seen in Pten single mutants, even at advanced age. Given that the primary effects of Tgfbr2 deletion using the PbCre4 model appeared to be in basal cells, the rapidity with which these tumors progressed was somewhat unexpected. The PbCre4 transgene begins to drive recombination at around 4 weeks of age [52], and extensive phenotypes were clearly detectable by 8 weeks. In our previous work, Pten;Tgfbr2 double mutants had a median survival of around 12–13 weeks [20]. Using the Krt8-CreERT2 driver here, and initiating tamoxifen treatment at 4 weeks after birth, median Pten;Tgfbr2 double mutant survival was 17–18 weeks of age, or 13–14 weeks after initiating tamoxifen treatment. This represents an approximately 50% increase in lifespan that may be due to the fact that only a fraction of luminal cells undergo recombination, initially resulting in smaller tumors. Although, it is possible that differences in strain background might also contribute to this difference in survival. However, this clearly shows that disrupting TGFß signaling specifically in luminal cells can drive progression to invasive cancer. Other tamoxifen dosing regimens resulted in similar phenotypes with reduced penetrance and longer lag times, although we did not attempt to induce recombination in very old animals.
The Krt8-CreERT2 driver used here demonstrates recombination that is highly specific to luminal cells in the prostate [29]. We searched extensively for basal cells in which recombination of the lineage tracing reporter could be detected early after tamoxifen treatment and did not find any evidence for recombination in Krt5 positive basal cells. Despite this, it is clearly possible that rare basal or uncommitted prostate epithelial cells may undergo recombination driven by this transgene. However, given our inability to identify such cells and the rapid onset of the phenotypes observed here we do not think that such inappropriate recombination contributes significantly to the phenotypes observed. Since basal cell proliferation is so rapid in the absence of Tgfbr2, if rare basal cells underwent recombination we would expect to identify small focal regions of basal cells within the tumors, without the relatively high proportion of cells expressing both basal and luminal keratins. It should be noted that these dual positive cells have a relatively low proliferation rate, certainly much less than that of basal cells and even lower than Pten;Tgfbr2 null luminal cells. If these cells arise from rare recombination events in uncommitted progenitors, or rare dual positive cells, it seems unlikely that they would become such a large proportion of the tumor given that their proliferation rate is lower than that of other cell types. Thus, the generation of dual positive cells in these tumors most likely occurs by differentiation of either luminal or basal cells to a more intermediate cell type. Since these dual positive cells have a lower proliferative rate than either basal or luminal cells, it is tempting to speculate that this may be linked to their apparently higher invasive ability. The fact that we see both dual positive and basal cells at high frequency in invasive cancers, while HGPIN is exclusively luminal suggests that both dual positive and basal cells arise from luminal cells lacking Pten and Tgfbr2. The relatively low proliferation rate of dual positive cells suggests that de-differentiation to this cell type must be a relatively frequent event in this tumor model. This suggests that TGFß signaling in the prostate acts to maintain the differentiated phenotype of luminal cells, and that loss of TGFß signaling in CaP may contribute to lineage plasticity. Interestingly, recent work has shown that TGFß signaling may also limit de-differentiation in intestinal tumors [53], suggesting that this may be a more common role of TGFß signaling in cancer progression.
Recent work has suggested that lineage plasticity may contribute to CaP recurrence following anti-androgen therapy [40, 41]. One possibility is that relatively rare multipotent cells may be able to survive androgen ablation and allow subsequent relapse even after the majority of the tumor, which has a primarily luminal phenotype, has regressed. Thus, therapeutic intervention might select for pre-existing cells that are more able to withstand the treatment and retain the ability to differentiate to luminal or other cell types. In addition, there has been interest in so called castration resistant Nkx3-1 expressing cells (CARNs). These cells may represent a similar population that is derived from luminal cells, and can be effectively selected for by androgen deprivation and restoration in mouse models [33, 34]. If this is a mechanism by which CaP can recur, then it would be of significant interest to understand what molecular pathways promote the existence or maintenance of such de-differentiated cells within the tumor. Our data suggest that TGFß signaling acts to maintain the differentiated state of luminal tumor cells and that loss of this signaling pathway allows de-differentiation to an intermediate cell type, which can then further differentiate to a basal cell phenotype. While the final differentiation to basal cells as seen in our model may not be the usual situation in human CaP, the generation of de-differentiated intermediate cells might provide insight into how these cells arise in human CaP. In this context, it is worth noting that in the extensively basal Pten;Tgfbr2 null invasive tumors from the PbCre4 model, there is enrichment for gene expression signatures that are indicative of aggressive luminal cancers.
The rapid basal cell proliferation in both the PbCre4 and Krt8-CreERT2 models places some limitation on them as far as being representative of human CaP, since basaloid carcinoma of the prostate is quite rare [31]. As with many mouse models of human cancer, one potential confounding issue is the complete inactivation of a particular gene (Tgfbr2 in this case). In contrast, in the human disease changes in expression or activity may be more graded, resulting in reduced activity rather than complete loss of function. Although recurrent inactivating mutations or deletions of the major components of the TGFß signaling pathways are not found in human CaP, there is evidence for reduced expression of both TGFBR1 and TGFBR2 in more advanced CaP [12–14], and reduced SMAD4 and TGFBR2 expression due to promoter methylation [9, 11]. Low SMAD4 expression was found to be part of a four gene signature that was prognostic for CaP recurrence and metastasis [24]. Our analysis of SMAD4 and active phosphorylated SMAD2 in human CaP also supports the idea that pathway activity may be dampened rather than completely abrogated in human CaP. While the mouse models used here are not ideal for addressing all questions regarding advanced human CaP, they may uncover important features of earlier stages of disease progression and metastasis. In the Krt8-CreERT2 model luminal tumor cells lacking Pten and Tgfbr2 de-differentiate to dual positive cells, a process that may be of interest for understanding human CaP progression. While loss of TGFß signaling may not be the only way to drive this de-differentiation, understanding this transition could contribute to the development of better therapeutic approaches. Recent work with mouse models and xenografts has suggested that androgen ablation selects for a more de-differentiated cell type with increased lineage plasticity in CaP [40, 41, 54]. This lineage plasticity may underlie the ability of human CaP to escape anti-androgen or anti-AR therapies, ultimately resulting in therapy failure. Analysis of human CaP phenotypes from before and after the introduction of newer anti-AR therapies, such as enzalutamide and abiratirone, suggests an increase in the frequency of AR negative tumors after anti-AR therapy [55]. The majority of these AR negative recurrent tumors lacked neuroendocrine features, and this transition could be mimicked by shutting off AR expression in the human LNCaP cell line. Interestingly, AR negative LNCaP cells gained expression of some basal markers, while retaining expression luminal-specific genes [55]. These studies clearly support the notion that current therapies for human CaP may fail in part due to the presence of de-differentiated tumor cells that do not rely on androgens for survival. These cells may initially be rare, or only exist in readily detectable numbers transiently following therapy. Once the recurrent disease has progressed to a more advanced stage such intermediate cells will likely have differentiated, either to a neuroendocrine phenotype [40, 41, 54], or to androgen-independent non-neuroendocrine tumors [55]. While the Krt8-CreERT2 model used here results in large numbers of de-differentiated dual positive cells, one limitation is that these tumors are overtaken by basal cells, possibly derived from the dual positive population. We do not know if interfering with androgen signaling affects the generation of dual positive cells in this model, but this may be of interest given the possibility that anti-androgen therapy selects for de-differentiated cells in human CaP.
A major question raised by our work is how frequently dual positive cells are found in human CaP samples, and what role they might play in tumor progression. Our analysis of human tumors is consistent with the idea that if dual positive cells are present, they are extremely rare in tumors prior to ADT. Indeed, we only identified dual positive cells in tumors that had undergone ADT. The frequency of post-ADT tumors with dual positive cells was more than 25% (3 of 11 samples), despite the heterogenous modes of ADT, and the variable time after treatment that the samples were isolated. Although this is based on a very small sample size, the identification of any tumors with dual positive cells, together with the evidence for increased lineage plasticity in post-ADT tumors, is quite intriguing. This population may represent an initially very rare cell type that is selected for by ADT, but is present only transiently prior to transition to a more differentiated cell type in the recurrent tumor. Clearly, it will be important to examine additional post-ADT tumors for the presence of dual positive cells to better determine how frequently they are present, and to further interrogate the molecular features of such cells. With a larger cohort of tumors bearing dual positive cells, it might be possible to identify pathways that promote this cell type in human CaP. Together, our data and that from other mouse CaP models, suggest that luminal cell differentiation to an intermediate cell type may be important for driving invasion, metastasis and disease recurrence. Intermediate cells may be selected for by androgen depletion, or may be promoted by other mechanisms, such as loss of Trp53 and Rb1, or by reduced activity of the TGFß pathway [40, 41, 54]. In human tumors, reduced Trp53, Rb1 or TGFß signaling might facilitate the selective effects of ADT on cells with increased linage plasticity.
Our analysis of metastases in the Krt8-CreERT2 model suggests that larger lung tumors resemble the primary prostate tumors, expressing the AR at varying levels and Trp63 in a proportion of the cells. This is clearly consistent with the notion that these are indeed prostate-derived metastases, although we cannot definitively rule out other origins. In addition, these lung metastases have a mixture of basal, luminal and dual positive cells as seen in the primary tumors. However, our examination of smaller metastases in the PbCre4 and Krt8-CreERT2 models suggests that a larger proportion of the cells within these micro-metastases are dual positive. This is particularly striking in the PbCre4 model, in which lung metastases must be from prostate. In this model a very low proportion of the primary tumor is dual positive. In contrast, more than one third of cells in the micro-metastases are dual positive and almost all micro-metastases identified contained dual positive cells. This suggests that these cells may be more metastatic, or that cells that are competent for metastasis are also able to de-differentiate, further emphasizing the potential importance of such highly plastic cells to disease progression. In this context, it is interesting to note that the widespread appearance of dual positive cells in the Krt8-CreERT2 model roughly coincides with the onset of extensive locally invasive cancer. Even in a single tumor, ducts with HGPIN are purely luminal whereas adjacent regions of invasive cancer frequently contain a mix of luminal, basal and dual positive cells. One question this raises is why, if TGFß limits plasticity, the cells within a relatively intact duct do not undergo de-differentiation. One possibility is that the initial invasion out of the duct exposes the cells to other signals that in the absence of a luminal cell TGFß response are able to drive de-differentiation.
In summary, we show that loss of TGFß signaling specifically from Pten null luminal prostate cells allows for rapid progression to an invasive and metastatic phenotype that is accompanied by de-differentiation. We suggest that this may be analogous to the proposed lineage plasticity seen in human CaP following androgen ablation or in other mouse models in which Pten and Rb1 have been deleted. Our analysis of human samples suggests that dual positive cells can be found in human CaP, potentially only after ADT. Importantly, by combining luminal specific deletion, lineage tracing and extensive analysis of basal and luminal markers, we provide evidence for a transition from luminal tumor cells to a de-differentiated invasive cell type that may be important for metastatic spread.
All animal procedures were approved by the Animal Care and Use Committee of the University of Virginia, which is fully accredited by the AAALAC. The use of archival human specimens was approved by the University of Virginia IRB (HSR# 13310; Identification of biomarkers in diseased human specimens). Under this protocol no patient consent was required, as samples were archival leftovers from diagnostic samples with patient identifiers not released to the principal investigator.
The loxP flanked Pten, Apc and Tgfbr2 alleles and the PbCre4 allele have been described previously [52, 56–58]. Tgfbr2 and Apc mice, and the PbCre4 transgenics were obtained from the NCI. The Krt8-CreERT2 line was from Jax (#017947 [50]). Conditional alleles with loxP flanked exons, which when recombined result in null alleles, and are referred to here as ‘r’ for recombined (null), or ‘f’ for the conditional loxP flanked (equivalent to wild type). Mice with a lineage tracing reporter (mTmG; #007576) were obtained from Jax [49]. All mouse lines were maintained on a mixed C57BL/6J x FVB strain background, as previously described [19, 20]. Tamoxifen treatment was by gavage, using 200mg/kg tamoxifen in corn oil daily for five days, or two sets of five days separated by a two day break. Genomic DNA for PCR genotype analysis was purified from ear punch, at post-natal day 21 (P21), by HotShot [59], and genotypes were determined by PCR. TRAMP tumor samples were as in [60].
RNA was isolated and purified using Absolutely RNA kit (Agilent) and quality checked by Bioanalyzer. For RNA isolation ventral lobes were used in all cases where the lobes were still distinguishable. For larger invasive tumors, a portion of the tumor was taken, although in most cases the lobes were not distinguishable. cDNA was generated using Superscript III (Invitrogen), and analyzed in triplicate by real time PCR using a BioRad MyIQ cycler and Sensimix Plus SYBRgreen plus FITC mix (Bioline), with intron spanning primer pairs, selected using Primer3 (http://frodo.wi.mit.edu/). Expression was normalized to Rpl4 and Cyclophilin using the delta Ct method.
Poly-A RNA-seq libraries generated with Illumina barcodes were sequenced (Illumina HiSeq at HudsonAlpha) to a target depth of ~ 25M paired end 50bp reads per sample as previously described [61]. Raw FASTQ sequencing reads were chastity filtered and reads were assessed for quality using FastQC. Reads were splice-aware aligned to the reference genome using STAR [62], and reads overlapping gene regions were counted using featureCounts [63]. The DESeq2 Bioconductor package [64] in the R statistical computing environment (http://www.R-project.org/) was used for normalizing count data, performing exploratory data analysis, estimating dispersion, and fitting a negative binomial model for each gene comparing the expression from tumors to wild type samples. Benjamini-Hochberg FDR was used to correct p-values for multiple testing. A cut-off of +/- 1.0 log2 and an adjusted P-value of < 0.0001 was considered significant. GO analysis was performed using either DAVID (https://david.ncifcrf.gov/) [65, 66] or ENRICHR (http://amp.pharm.mssm.edu/Enrichr/) [67, 68], heat maps were generated using ClustVis (http://biit.cs.ut.ee/clustvis/) [69] and gene set enrichment was performed using GSEA software from the Broad Institute [70, 71]. RNA-seq data is deposited at GEO, under the accession number: GSE108017.
Tissues were fixed in zinc-formalin, paraffin-embedded and sectioned at 5 microns, and were stained with Hematoxylin and Eosin (H&E), or prepared for immunostaining as previously described [61, 72]. Images were captured with 10, 20 or 40x objectives, using a Nikon Eclipse NI-U with a DS-QI1 or DS-Ri1 camera and NIS Elements software, and adjusted in Adobe Photoshop. Antibodies were as follows: Rabbit and chicken anti-Krt5, mouse anti-Krt8 (Covance), rabbit anti-Krt8 (Abcam), rabbit anti-Ki-67 (Abcam), rabbit anti-AR (AR-441; Abcam), mouse anti-p63 (Biocare medical), rabbit anti-Syp (Thermo Fisher Scientific), rabbit anti-Krt10 (Covance), rabbit anti-Tgfbr2 (Novus), mouse anti-pAkt (Cell Signaling), rabbit anti-pSmad2 (Millipore), rabbit anti-Smad4 (Millipore) and chicken anti-GFP (Abcam). Alexafluor 488, 546 and 647 secondary antibodies were from Invitrogen. DNA was stained with Hoechst 33342. Confocal images were captured using a Zeiss LSM710 Multiphoton Confocal microscope and adjusted in Adobe Photoshop. Two tissue microarrays (TMA) contained four 0.6mm tissue cores from each of 93 cancers from patients who had radical prostatectomy. A third TMA contained four 0.6mm cores from each of 45 cancers from TURP samples. TMAs were examined by IHC, as in [20], or by IF for Krt5 and Krt8. Following IF and imaging, the same slide was stained with H&E.
|
10.1371/journal.pbio.1001900 | Calcineurin Mediates Synaptic Scaling Via Synaptic Trafficking of Ca2+-Permeable AMPA Receptors | Homeostatic synaptic plasticity is a negative-feedback mechanism for compensating excessive excitation or inhibition of neuronal activity. When neuronal activity is chronically suppressed, neurons increase synaptic strength across all affected synapses via synaptic scaling. One mechanism for this change is alteration of synaptic AMPA receptor (AMPAR) accumulation. Although decreased intracellular Ca2+ levels caused by chronic inhibition of neuronal activity are believed to be an important trigger of synaptic scaling, the mechanism of Ca2+-mediated AMPAR-dependent synaptic scaling is not yet understood. Here, we use dissociated mouse cortical neurons and employ Ca2+ imaging, electrophysiological, cell biological, and biochemical approaches to describe a novel mechanism in which homeostasis of Ca2+ signaling modulates activity deprivation-induced synaptic scaling by three steps: (1) suppression of neuronal activity decreases somatic Ca2+ signals; (2) reduced activity of calcineurin, a Ca2+-dependent serine/threonine phosphatase, increases synaptic expression of Ca2+-permeable AMPARs (CPARs) by stabilizing GluA1 phosphorylation; and (3) Ca2+ influx via CPARs restores CREB phosphorylation as a homeostatic response by Ca2+-induced Ca2+ release from the ER. Therefore, we suggest that synaptic scaling not only maintains neuronal stability by increasing postsynaptic strength but also maintains nuclear Ca2+ signaling by synaptic expression of CPARs and ER Ca2+ propagation.
| Synaptic scaling is a form of homeostatic plasticity that normalizes the strength of synapses (the structure that allows nerve cells to communicate) and is triggered by chronic inhibition of neuronal activity. Although extensive studies have been conducted, the molecular mechanism of this synaptic adaptation is not understood. Using cultured cortical neurons, we show that chronic inhibition of neuronal activity reduces calcium influx into neurons, which, in turn, decreases the activity of the calcium-dependent phosphatase calcineurin. These changes lead to an increase in GluA1-containing, calcium-permeable AMPA receptors, which mediate communication at the synapse. Newly inserted calcium-permeable AMPA receptors restore calcium currents, which enhance synaptic strength and recover calcium signaling. We also show that inhibition or activation of calcineurin activity is sufficient to induce or block synaptic scaling, respectively, suggesting that calcineurin is an important mediator of homeostatic synaptic plasticity. Taken together, our findings show that synaptic scaling is a homeostatic process that not only enhances synaptic transmission but also maintains calcium signaling in neurons under activity deprivation.
| Synaptic scaling, a form of homeostatic synaptic plasticity, is a negative feedback process that stabilizes neuronal activity in response to changes in synaptic strength by altering various aspects of neuronal function [1]. It has been implicated in neurodevelopment and in neurological disorders [2]–[5]. One of the mechanisms underlying synaptic scaling is the regulation of synaptic strength through control of delivery or retention of AMPARs at synapses [1]. During homeostatic adaptation, synaptic AMPARs are increased or reduced in response to activity deprivation or overexcitation, respectively, by altering AMPAR synaptic insertion and internalization [6]. Synaptic adaptation can be global and multiplicative, which is important for preserving the relative strength differences between synapses. Because each synapse strength is multiplied or divided by the same factor, each synaptic strength is increased or decreased in proportion to its initial strength [7]. Synaptic scaling is also induced by synapse-specific processes, providing local control of synaptic strength [8]. Numerous treatments that induce homeostatic regulation but differ in their experimental conditions have been reported. Nonetheless, the homeostatic plasticity mechanism is still not well understood. Here, we describe a novel mechanism in which activity deprivation induces synaptic scaling by a calcineurin-mediated process.
AMPARs are the major excitatory postsynaptic glutamate receptor in the central nervous system and consist of four subunits (GluA1–4) [9]. There are two general types of AMPARs formed through combination of these subunits, Ca2+-impermeable GluA2-containing and Ca2+-permeable, GluA2-lacking/GluA1-containing receptors [10]. Ca2+-permeable AMPARs (CPARs) are generally sensitive to polyamine block, although there is a third class of AMPARs that are Ca2+-permeable but insensitive to polyamines [11]. The GluA1 and GluA2 AMPAR subunits can assemble channels with markedly different electrophysiological and trafficking properties [10],[11] and both GluA1 and GluA2 can contribute to homeostatic synaptic plasticity [1],[12],[13]. Phosphorylation of GluA1 within its intracellular carboxyl-terminal domain can regulate AMPAR membrane trafficking and channel open probability [14]. Phosphorylation of serine 845 in GluA1 [pGluA1(S845)] is important for activity-dependent trafficking of GluA1-containing AMPARs, and cAMP-dependent protein kinase A (PKA) and cGMP-dependent protein kinase II (cGKII) can mediate this phosphorylation [14],[15]. The Ca2+/calmodulin-dependent protein phosphatase, calcineurin, dephosphorylates pGluA1(S845), which enables GluA1-containing AMPARs to be endocytosed from the plasma membrane during long-term depression [16],[17]. Therefore, activity-dependent GluA1 phosphorylation can play critical roles in GluA1 synaptic trafficking and forming CPARs in synapses.
The most studied experimental system for synaptic scaling is the inhibition of neuronal activity by TTX (tetrodotoxin), which blocks sodium channels and thereby inhibits action potentials. TTX-dependent chronic inhibition of action potentials results in an increase in the strength of synaptic transmission as a compensatory process that can be measured by increases in AMPAR-mediated miniature excitatory postsynaptic currents (mEPSCs) [18]. Recent studies suggest that TTX reduces somatic Ca2+ influx and inhibits activation of Ca2+/calmodulin-dependent protein kinase IV (CaMKIV), which promotes synaptic scaling [19]. CaMKs are important for Ca2+-dependent synaptic plasticity [20], and inhibition of Ca2+ influx is sufficient to induce AMPAR-mediated synaptic scaling [21],[22]. This suggests that reduction of CaMK activation and downstream signaling by activity deprivation-induced inhibition of Ca2+ influx can play a critical role in AMPAR-dependent homeostatic scaling, yet there is no complete molecular mechanism linking activity-dependent Ca2+ signals and homeostatic regulation of AMPARs.
Here, we focus on the role of synaptic Ca2+ and calcineurin in synaptic scaling. We show that activity suppression reduces Ca2+ influx in neurons, which in turn decreases the activity of calcineurin. This stabilizes pGluA1(S845), which increases synaptic CPARs. This increases synaptic strength as a compensatory response to activity deprivation and restores synapse-to-nucleus Ca2+ signaling via ER Ca2+ wave propagation. Thus, we conclude that synaptic scaling via calcineurin and CPARs provides a means to maintain not only synaptic activity but also Ca2+ signaling as a homeostatic response.
To confirm activity-dependent homeostatic scaling, we studied spontaneous synaptic transmission by measuring mEPSCs in DIV14–17 cultured mouse cortical neurons (Figure 1a) and found that treatment for 48 h with 2 µM TTX significantly increased average mEPSC amplitude (no TTX, 19.68±0.99 pA and 48 TTX, 28.01±1.12 pA, p<.0001) (Figure 1b) consistent with the previous finding [23], whereas mEPSC frequency was not altered (Figure 1c). Importantly, cumulative probability distributions of the mEPSC amplitude were uniformly increased by TTX treatment, and the increase in the amplitude with TTX treatment was multiplicative (Figure 1e). There was a significant decrease in mEPSC decay time (peak to 10%) with TTX treatment (no TTX, 2.66±0.15 ms and 48 TTX, 1.98±0.05 ms, p = .0008) (Figure 1d). Because CPARs show a shorter decay time [21],[24], we used 20 µM naspm (1-naphthyl acetyl spermine) or 5 µM PhTX (philanthotoxin-74), blockers of CPARs, to determine if CPARs were responsible for the TTX-mediated increase of the amplitude (Figure 1a). Consistent with previous findings [21],[23],[25], naspm and PhTX treatment significantly reduced the TTX-induced increase in amplitude (48 TTX, 28.01±1.12 pA; 48 TTX+naspm, 21.31±0.44 pA, p = .0002; and 48 TTX+PhTX, 18.50±0.58 pA, p<.0001) (Figure 1b), but frequency was not affected (Figure 1c). Naspm and PhTX also significantly increased decay time (48 TTX, 1.98±0.05 ms; 48 TTX+naspm, 2.47±0.15 ms, p = .0186; and 48 TTX+PhTX, 2.38±0.07 ms, p = .0453) (Figure 1d) as found previously [21]. CPAR inhibitors had no effects on mEPSCs of neurons in the absence of TTX treatment, suggesting that CPARs made no contribution under the basal condition (Figure 1). Thus, TTX treatment induced CPAR-mediated multiplicative synaptic scaling.
Because pGluA1(S845) is required not only for homeostatic scaling in the visual cortex [26] but also for maintaining CPARs on the synaptic membrane [27], we measured the effects of TTX on pGluA1(S845) levels by purifying synaptosomes from TTX-treated neurons and measuring protein and phosphorylation levels of AMPAR subunits. TTX treatment significantly increased pGluA1(S845) (p = .024), whereas total GluA1 and GluA2/3 levels were not changed (Figure 2a). We further determined that surface GluA1 levels were increased (p = .0127) after TTX treatment, whereas surface GluA2/3 was not altered (Figure 2b). We next analyzed mutant GluA1 (GluA1 S845A, unable to be phosphorylated on serine 845) using GluA1 S845A knock-in mice [28] and found that TTX treatment was unable to induce synaptic scaling in neurons from the mutant mouse (Figure 2c). This suggested that TTX treatment enhanced GluA1 surface trafficking by increasing pGluA1(S845). This newly trafficked GluA1 could be in the form of CPARs that promote synaptic scaling.
Increasing pGluA1(S845) can be achieved either by enhancing kinase activity or by decreasing phosphatase activity. A-kinase anchoring protein (AKAP) and SAP97 form a protein complex with GluA1 that tethers PKA and calcineurin, which regulate channel functions, respectively, through GluA1 phosphorylation and dephosphorylation [29],[30]. Therefore, reduction of calcineurin activity is a candidate for mediating an increase of pGluA1(S845) in response to the TTX-induced reduction of Ca2+ influx. We found that calcineurin protein levels were significantly decreased (p = .0414) in synaptosomes following TTX treatment (Figure 2a). To measure in vivo calcineurin activity directly, we used a fluorescence resonance energy transfer (FRET)-based calcineurin activity sensor that utilizes a calcineurin activity-dependent molecular switch based on the N-terminal regulatory domain of nuclear factor of activated T cells (NFAT) as a specific substrate, which was inserted between CFP and YFP [31]. Inhibition of calcineurin activity by 12 h treatment with 5 µM FK506, which forms a drug-immunophilin complex that is a highly specific inhibitor for calcineurin [32], significantly decreased FRET activity (assayed by measuring the emission ratio) as compared with that under the basal condition (no TTX, 1.45±0.02 and FK506, 1.06±0.01, p<.0001), which confirmed that the reporter detected calcineurin activity (Figure 3). Calcineurin activity was significantly decreased after 24 h TTX treatment and further reduced after 48 h TTX treatment, whereas 12 h TTX had no effect on the emission ratio (12 TTX, 1.43±0.02; 24 TTX, 1.33±0.02, p<.0001; and 48 TTX, 1.20±0.01, p<.0001) (Figure 3). This suggested that chronic inhibition of neuronal activity decreased calcineurin activity in a time-dependent manner and lowered synaptic calcineurin levels.
Calcineurin inhibition affects both mEPSC frequency and amplitude [33],[34] and stabilizes pGluA1(S845) [35], and reduction of cytoplasmic Ca2+ lowers calcineurin activity, followed by enhancement of GluA1-containing AMPAR-mediated transmission [36]. To determine whether inhibition of calcineurin was sufficient for inducing a pharmacologic form of synaptic scaling in the absence of TTX treatment, we next blocked calcineurin activity by 12 h treatment with 5 µM FK506 and measured mEPSCs (Figure 4a). FK506 treatment significantly increased mEPSC amplitude compared with DMSO treatment (DMSO, 20.38±1.08 pA and FK506, 28.54±1.41 pA, p<.0001) (Figure 4b). Consistent with a previous study showing that inhibition of calcineurin increases mEPSC frequency [33],[34] through calcineurin modulation of presynaptic activity [37], we found increased mEPSC frequency in FK506-treated neurons (DMSO, 4.21±0.37 Hz and FK506, 9.88±0.27 Hz, p<.0001) (Figure 4c). Furthermore, the mEPSC decay time in FK506-treated neurons was significantly faster (DMSO, 2.51±0.10 ms and FK506, 1.96±0.10 ms, p = .0047), suggesting that CPARs mediated the scaling induced by FK506 (Figure 4d). Moreover, cumulative probability distributions were uniformly shifted by FK506, and the increase in the amplitude was multiplicative (Figure 4e). We confirmed CPAR-mediated scaling in FK506-treated neurons by adding naspm or PhTX (Figure 4a), which caused a significant reduction of mEPSC amplitude (FK506, 28.54±1.41 pA; FK506+naspm, 20.31±1.14 pA, p<.0001; and FK506+PhTX, 19.33±0.76 pA, p<.0001), whereas no effect was observed following naspm or PhTX treatment of DMSO-treated neurons (Figure 4b). There were no significant changes in mEPSC frequency after napsm or PhTX treatment of either DMSO or FK506-treated neurons (Figure 4c). Moreover, the FK506-induced change in decay time was reversed by naspm and PhTX only for the FK506-treated neurons (FK506, 1.96±0.10 ms; FK506+naspm, 2.44±0.20 ms, p = .0152; and FK506+PhTX, 2.70±0.15 ms, p = .0003) (Figure 4d). Similar to the effects of 48 h TTX treatment, FK506 treatment increased pGluA1(S845) (p = .0474) in synaptosomes, whereas total GluA2/3 and GluA1 levels were not altered (Figure 5a). Surface GluA1 was significantly elevated with FK506 treatment (p<.0001) (Figure 5b). Moreover, FK506 treatment significantly reduced calcineurin levels in synaptosomes (p = .0299) (Figure 5a). These results indicated that inhibition of calcineurin by FK506 was sufficient to induce synaptic trafficking of CPARs by increasing pGluA1(S845) and that FK506 could produce a pharmacologic form of synaptic scaling without TTX-mediated activity deprivation.
To test whether persistent calcineurin activity could block TTX-mediated synaptic scaling, we generated a constitutively active calcineurin mutant, which has Ca2+-independent, constitutive phosphatase activity, by deleting the calcineurin autoinhibitory domain (CaN-ΔAI) [38]. As expected, when we cotransfected HEK293 cells with GluA1 and CaN-ΔAI, pGluA1(S845) levels were significantly lower (p = .0198) than in cells transfected with GluA1 alone (Ctrl) (Figure 6a). Although CaN-ΔAI decreased pGluA1(S845), surface GluA1 levels remained unaffected in cultured neurons (Figure S1a). When CaN-ΔAI was cotransfected with GFP into neurons and we measured mEPSCs after 48 h TTX treatment, we found that TTX was unable to induce synaptic scaling in the presence of CaN-ΔAI (Figure 6b–e). However, TTX treatment of neurons expressing GFP alone induced a typical CPAR-mediated synaptic scaling (Figure 6b–e) as seen previously, with increased mEPSC amplitude (no TTX, 12.98±0.47 pA and 48 TTX, 18.21±1.52 pA, p<.0001) (Figure 6c) and decreased decay time (no TTX, 4.37±0.31 ms and 48 TTX, 3.05±0.31 pA, p = .0072) (Figure 6e), whereas frequency of mEPSCs was not altered (Figure 6d). This suggested that a gain-of-function calcineurin mutant could inhibit TTX-induced synaptic scaling.
Ca2+ signals are thought to be important for synaptic scaling, which suggests that a reduction of Ca2+ influx may be a critical trigger for synaptic scaling [1],[19],[21],[22]. Furthermore, lowering cytoplasmic Ca2+ levels has been reported to enhance GluA1-containing AMPAR-mediated transmission [36]. We investigated Ca2+ activity in cultured neurons transfected with GCaMP5, a genetically encoded Ca2+ indicator [39] (Figure 7a). We found active spontaneous Ca2+ transients in neurons without TTX treatment (Figure 7b–c). To determine the effects of action potentials and mEPSC activity on Ca2+ transients, we first added TTX at the time of imaging and found that acute TTX treatment completely blocked Ca2+ activity (p<.0001) (Figure 7a–c). Furthermore, naspm treatment of neurons in the absence of TTX treatment had no significant effect on Ca2+ transients (Figure 7a–c), suggesting that action potentials play a critical role in generating the Ca2+ activity observed under these conditions, and that this activity is not dependent on CPARs. In contrast, following 48 h treatment with TTX, about 50% of the Ca2+ signal was restored (p = .0002), and this restored activity observed in the presence of TTX was significantly reduced by naspm (p = .0079) (Figure 7a–c). This suggested that TTX-induced scaling provided a mechanism for maintaining Ca2+ activity that was dependent in part upon the synaptic expression of CPARs.
We next investigated effects of calcineurin inhibition on Ca2+ signals (Figure 8a). Neurons with 12 h DMSO treatment displayed normal Ca2+ activity, and acute TTX treatment completely inhibited the activity (p<.0001) (Figure 8a–c). Neurons treated for 12 h with FK506 showed active spontaneous Ca2+ transients comparable to those in neurons without TTX treatment (Figure 8a–c). Conversely, when TTX was acutely added at the time of imaging to neurons that had been treated for 12 h with FK506, TTX was unable to block the Ca2+ signals completely (p = .005) (Figure 8a–c). To determine whether this Ca2+ signal activity was mediated by CPARs, naspm was added to neurons at the time of recording (Figure 8a). Naspm significantly reduced the activity (p = .0069), indicating it was from CPARs (Figure 8a–c). Furthermore, we confirmed that CaN-ΔAI blocked synaptic scaling-mediated recovery of Ca2+ signals (p<.0001) (Figure S1b), consistent with the finding that CaN-ΔAI inhibited TTX-induced synaptic scaling (Figure 6c). This suggested that calcineurin activity is important for both synaptic scaling and Ca2+ homeostasis mediated by CPARs, which partially restored Ca2+ signaling. Taken together, these results demonstrate that CPAR/calcineurin-dependent synaptic scaling provides a mechanism for homeostasis of Ca2+ signals in part as a homeostatic response to activity deprivation-induced inhibition of Ca2+ activity.
Both extracellular and intracellular sources of Ca2+ are used by neurons [40]. Although Ca2+ influx from extracellular sources is mediated by various Ca2+ channels including NMDA receptors (NMDARs) at synapses and voltage-gated Ca2+ channels in the plasma membrane, inositol 1,4,5-trisphosphate receptors (IP3Rs) and ryanodine receptors (RyRs) in the ER are responsible for intracellular Ca2+ release [40]. We first investigated which Ca2+ sources were responsible for GCaMP5-positive Ca2+ signals (Figure 9a). To address this question, we blocked each Ca2+ channel and measured spontaneous Ca2+ signals without drug pretreatment (Figure 9a–b). When we acutely treated neurons with 10 µM nifedipine, an L-type Ca2+ channel blocker, spontaneous Ca2+ signals were unaltered, but the NMDAR antagonist, 50 µM APV, significantly reduced Ca2+ activity (p<.0001), suggesting that GCaMP5 detected Ca2+ signals including those from NMDARs but not from L-type Ca2+ channels (Figure 9a–b). We next depleted Ca2+ from the ER by inhibiting sarco/endoplasmic reticulum Ca2+-ATPase using 1 µM thapsigargin and found that thapsigargin treatment completely inhibited Ca2+ activity (p<.0001) (Figure 9a–b). Moreover, blocking both IP3Rs and RyRs by 50 µM 2APB and 25 µM dantrolene significantly lowered Ca2+ signals (p<.0001), suggesting that GCaMP5 detected Ca2+ released from the ER, possibly dependent on the activity of NMDARs (Figure 9a–b). This further suggested that GCaMP5-positive Ca2+ signals restored by synaptic scaling were mediated by ER Ca2+ release.
NMDAR-mediated synaptic Ca2+ influx evokes Ca2+ signals in the nucleus via Ca2+ wave propagation through the ER [40]. This Ca2+ signaling is essential for synaptic plasticity and regulates gene expression through CREB in addition to local signaling in synapses [40]. Because an NMDAR antagonist blocks CREB activation [40] and CPARs also regulate ER Ca2+ release [41], we hypothesized that CPARs replace the role of NMDARs in synapse-to-nucleus Ca2+ signaling via the ER Ca2+ release when neuronal activity is chronically suppressed by TTX. Consistent with previous findings [42],[43], we found that CREB activity (assayed by measuring phosphorylation at serine 133 of CREB) was reduced with 6 h treatment of 2 µM TTX (p = .0004) or 1 µM thapsigargin (p = .0019), confirming that CREB activity was dependent on both neuronal activity and ER Ca2+ (Figure 9c). However, after synaptic scaling was induced by 48 h treatment with TTX, CREB activity was significantly increased, suggesting that ER Ca2+ signals restored by synaptic scaling provided a means to maintain CREB phosphorylation in the nucleus (Figure 9c). Treatment with 20 µM naspm for 6 h significantly reduced the CREB phosphorylation seen in neurons pretreated with TTX for 48 h (p<.0001), suggesting CPARs were responsible for homeostasis of CREB phosphorylation (Figure 9c). Taken together, this work shows that when neuronal activity is suppressed by TTX, synaptic scaling maintains basal CREB activity via synapse-to-nucleus Ca2+ signals by expression of CPARs at synapses and by ER Ca2+ waves.
We demonstrate a novel Ca2+ homeostasis-dependent mechanism of synaptic scaling mediated by calcineurin and CPARs. Based on our findings, we propose the following model. Under basal conditions, action potentials provide synaptic Ca2+ signals via NMDARs, followed by Ca2+-induced Ca2+ release from the ER, leading to nuclear Ca2+ signals that maintain CREB-mediated transcriptional activity. In addition, synaptic Ca2+ influx activates calcineurin, which removes GluA1 from the synaptic membrane by dephosphorylating pGluA1(S845), providing a balance between GluA1 insertion by kinases and removal by phosphatases in synapses (Figure 10). However, under the condition of activity deprivation, NMDAR-mediated synaptic Ca2+ influx is inhibited, leading to inactivation of calcineurin. This induces synaptic expression of CPARs via stabilization of pGluA1(S845), thereby enhancing synaptic strength and promoting synaptic Ca2+ influx via CPARs instead of NMDARs (Figure 10). This restores Ca2+ signals and CREB phosphorylation and activation. We thus suggest that synaptic scaling not only maintains neuronal activity by increasing CPAR-dependent postsynaptic strength but also maintains CREB activation by synapse-to-nucleus Ca2+ signaling.
Although it has been shown that postsynaptic AMPARs play a critical role in homeostatic synaptic plasticity, there is no generally agreed mechanism for synaptic scaling, possibly due to the fact that multiple experimental conditions have been investigated [13]. Many studies conducted in several experimental models support a role for this plasticity mediated by GluA1-containing AMPARs. For example, various protocols have been used to inhibit neuronal activity and induce synaptic scaling in cultured neurons, such as inhibition of action potentials by TTX [23], AMPARs by NBQX [21], L-type Ca2+ channels by nifedipine [21], or NMDARs and action potentials together by APV and TTX [22]. Regardless of inhibition protocols, each treatment induced CPAR-dependent synaptic scaling. Furthermore, visual deprivation in the cortex is sufficient for inducing CPAR-dependent homeostatic synaptic plasticity in vivo [26]. Nonetheless, the cellular mechanism by which neurons detect activity deprivation and what is the molecular readout of this signal that regulates postsynaptic AMPARs for synaptic scaling has not yet been identified. A recent study by Gainey et al. employing GluA2 knockdown reports that GluA2 is required for homeostatic synaptic plasticity [44]. It is possible that the increased expression of synaptic CPARs that occurs in the GluA2 knockdown prior to addition of TTX increases synaptic Ca2+ fluxes that prevent further CPAR synaptic trafficking required for synaptic scaling. In contrast, the GluA2 knockout exhibits normal synaptic scaling after chronic TTX treatment [45]. Significantly, in the knockout of GluA2, GluA1 levels at synapses are lower than in the wild type [46], a change which would generate significantly lower Ca2+ flux than the knockdown, which in turn would make synaptic scaling possible. Given these considerations, the experimental findings of the others are consistent with the current work.
Ca2+ influx in response to synaptic stimulation or action potentials plays an important role in regulating various neuronal functions including releasing neurotransmitter, modulating ion channels, and promoting synaptic plasticity and gene expression [47],[48]. Somatic Ca2+ levels are thought to be an important activity sensor in homeostatic synaptic plasticity [1],[19],[49]. Downstream effectors of Ca2+ signaling including CaMKs and adenylyl cyclases can be molecular readouts of the Ca2+ influx [13]. Because chronic neuronal inactivation reduces Ca2+ influx and downregulates adenylnly cyclases [50], cAMP-dependent PKA activity is unlikely elevated by TTX to increase phosphorylation of S845 of GluA1, although it needs further investigation. Calcineurin is the only Ca2+/calmodulin-activated phosphatase in the brain, and it is a major regulator of several key proteins mediating synaptic transmission and neuronal excitability in both pre- and postsynaptic areas [51]. Due to the fact that calcineurin inhibition promotes an increase in both mEPSC frequency and amplitude, it has been proposed to have a role in homeostatic synaptic plasticity [33]. Moreover, lowering basal Ca2+ levels has been shown to strengthen AMPAR-mediated transmission, which is dependent on GluA1 and calcineurin [36]. A computational modeling study predicts that calcineurin can be active at moderate Ca2+ concentrations, whereas the activity of PKA requires high Ca2+ levels [52]. It is thus possible that calcineurin can remain active in the short term, even after action potentials and synaptic Ca2+ influx are abolished by TTX. Consistent with this study, we found that calcineurin activity was not reduced immediately after TTX treatment, and the reduction was found after a 24 h treatment with TTX (Figure 3). This persistence of activity may explain why a significant length of time of application of TTX is required to express synaptic scaling. It has been shown that calcineurin inhibition increases pGluA1(S845) and selectively increases synaptic expression of CPARs [17]. Further investigation is required to determine how calcineurin inhibition selectively increases CPARs, given that it could potentially affect both GluA1 homomeric and GluA1/2 heteromeric AMPARs. We also showed that CaN-ΔAI was sufficient for reducing pGluA1(S845) levels, although surface GluA1 levels were not altered, which accounts for normal mEPSCs (Figure 6 and Figure S1a). Although it is not clear how normal levels of surface GluA1 are maintained when S845 phosphorylation is decreased, this is not surprising because several lines of studies already show that (1) there is normal synaptic transmission in the hippocampus of CaN-ΔAI overexpressed transgenic mice [53]; (2) in GluA1 S845A mutant mice, surface GluA1 levels are not affected [28]; and (3) GluA1 S845A mice also display normal mEPSCs in the amygdala [54]. Taken together, phosphorylation of GluA1 to yield pGluA1(S845) may not be critical for maintaining basal synaptic transmission, but can be important for activity-dependent plasticity, such as long-term potentiation or synaptic scaling.
It has been shown that lowering calcineurin activity with cyclosporin A, another calcineurin inhibitor, decreases not only enzymatic activity but also calcineurin protein levels [55], consistent with our findings (Figures 2a and 5a), suggesting that inactive calcineurin may be degraded. Neuronal activity regulates synaptic proteins and signaling through the ubiquitin-proteasome system, providing a mechanism that links activity and protein turnover [56]. Ca2+ entry is an important process regulated by neuronal activity that promotes a decrease of protein ubiquitination in synapses, which depends on calcineurin activity [57]. Calcineurin also can be ubiquitinated and undergo proteolysis in cardiomyocytes [58]. Thus, it is possible that chronic inhibition of neuronal activity decreases calcineurin activity by lowering Ca2+ influx, which promotes increased protein ubiquitination, including ubiquitination of calcineurin itself, followed by proteasome-mediated degradation, although this requires further investigation.
During activity deprivation, synaptic Ca2+ influx is reduced, possibly followed by inhibition of downstream Ca2+ signaling [1],[6],[12]. Synaptic scaling may provide a mechanism to overcome these problems. GluA1-containing CPARs are an attractive candidate for restoration of Ca2+ activity during synaptic scaling because unlike GluA2-containing, Ca2+-impermeable AMPARs, they not only have larger single channel conductance but also are Ca2+-permeable [59]. Based on our findings, we suggest that during homeostatic synaptic scaling, CPARs are stabilized in synapses and conduct Ca2+, which increases synaptic strength and also partially restores suppressed synaptic Ca2+ signals as shown by the finding that GCaMP5 predominantly detected Ca2+ release from the ER (Figures 7–9). In addition, cytosolic Ca2+ levels may reflect neuronal activity on a cell-wide basis, permitting Ca2+-dependent mechanisms to control all synapses, a feature of multiplicative homeostatic synaptic plasticity. Therefore, recruitment of CPARs may provide feedback regulation to maintain neuronal activity and Ca2+ signaling during synaptic scaling.
CaMKs are important for Ca2+-dependent synaptic plasticity, and reduction of CaMKIV activation is sufficient for inducting synaptic scaling without TTX treatment [19],[20]. CaMKIV-mediated activation of the CREB transcription factor is important for synaptic plasticity and learning [40]. Thus, it is possible that inhibition of CaMKIV activity reduces CREB activation and promotes synaptic scaling. Because the neuronal CREB transcription factor regulates various signaling pathways including those for learning, addiction, and pain [40], homeostasis of basal CREB activity can be important for brain function and may be maintained by synaptic scaling. Thus, synaptic scaling can provide a mechanism for maintaining basal levels of CREB-mediated transcriptional activity via synaptic Ca2+ and CaMKIV when neuronal activity is suppressed. Our data support this idea by showing that (1) TTX inhibited somatic Ca2+ signals, (2) TTX treatment reduced CREB phosphorylation, and (3) synaptic scaling restored Ca2+ signaling and CREB activation.
In summary, we conclude that synaptic scaling not only maintains neuronal stability by increasing CPAR-dependent postsynaptic strength but also maintains basal CREB transcriptional activity through nuclear Ca2+ signaling as a homeostatic response to suppression of neuronal activity.
Cortical primary neurons were prepared by a modification of the previously described method [60]. Neurons were isolated from embryonic day 17–18 C57Bl6 or GluA1 S845A mouse embryonic brain tissues. All animal studies were performed with an approved protocol from New York University Langone Medical Center's Institutional Animal Care and Use Committee. Neurons were plated on poly-L-lysine–coated 15 cm dishes for biochemical experiments, size 12 mm cover glasses for electrophysiology and FRET assay, or glass-bottom dishes for Ca2+ imaging. Cells were grown in Neurobasal medium with B27 and 0.5 mM Glutamax (Life Technologies). For neuronal transfection, DIV4 neurons were transfected with Lipofectamine 2000 (Life Technologies) according to the manufacturer's protocol, and analysis was performed at DIV14–17.
Constitutively active calcineurin mutant (CaN-ΔAI) was generated according to the previous study [38]. A stop codon was introduced at lysine 399 of wild-type murine calcineurin alpha (Addgene, 17871) to produce CaN-ΔAI that lacked the calmodulin-binding and autoinhibitory domains, leading to Ca2+-independent, constitutive phosphatase activity [38]. CaN-ΔAI was cotransfected with GluA1 into HEK293 cells using Lipofectamine 2000 (Life Technologies) to confirm expression and pGluA1(S845) levels by immunoblots. For surface GluA1 leveling, CaN-ΔAI was cotransfected with mCherry into neurons, and surface GluA1 was determined by incubation of a GluA1 antibody (Calbiochem, PC246) under the nonpermeable condition. The Alexa Fluor-488 secondary antibody (Molecular Probes, A-11008) was used to visualize surface GluA1, and proximal dendrites (<100 µm from the cell body) were captured using Zeiss Axiovert 200 m. Images were analyzed by the ImageJ software.
Miniature EPSCs were measured in cortical neurons cultured from C57Bl6 or GluA1 S845A embryos at DIV14–17 as described previously [60]. Neurons were voltage clamped with the whole cell ruptured path technique during the recording. The bath solution contained (in mM) 119 NaCl, 5 KCl, 2.5 CaCl2, 1.5 MgCl2, 30 glucose 20 HEPES (Life Technologies), and 0.001 glycine (Sigma), pH 7.4. Patch electrodes (5–8 MΩ) were filled with (in mM) 120 K-gluconate (Sigma), 9 NaCl, 1 MgCl2, 10 HEPES, 0.2 EGTA (Sigma), 2 Mg-ATP (Sigma), and 0.2 GTP (Sigma). We added 1 µM TTX (Tocris Biosciences) and 10 µM bicuculline (Tocris Biosciences) to the bath to inhibit action potentials and miniature inhibitory postsynaptic currents, respectively. mEPSCs were recorded at −60 mV with a Warner amplifier (PC-501A) and filtered at 1 kHz. Recordings were digitized (Digidata 1440, Molecular Devices) and analyzed using the Mini Analysis software (Synaptosoft). The access resistance (Ra<25 MΩ) was monitored during recording to eliminate artifacts. Events whose amplitude was less than 7.5 pA were rejected. To induce synaptic scaling, neurons were pretreated with 2 µM TTX for 48 h or 5 µM FK506 or DMSO for 12 hrs. We added 20 µM naspm (1-naphthylacetyl spermine trihydrochloride, Tocris Biosciences) or 5 µM philanthotoxin-74 (Tocris Biosciences) to suppress CPAR-mediated transmission in the bath solution. For the CaN-ΔAI experiment, GFP was cotransfected with CaN-ΔAI to visualize transfected neurons, and mEPSCs were analyzed at DIV14–17.
Synaptosomal fractions from DIV14 primary cortical neurons were prepared as described previously [60],[61]. Surface biotinylation was performed according to the previous study [60]. Equal amounts of protein were loaded on 10% SDS-PAGE gel and transferred to the nitrocellulose membrane. Membranes were blotted with GluA1 (Millipore, 1∶5,000), GluA2/3 (Millipore, 1∶500), pGluA1(S845) (Millipore, 1∶1,000), calcineurin (Millipore, 1∶1,000 or Santa Cruz Biotechnology, 1∶500), actin (Sigma, 1∶5,000), pCREB (Cell Signaling, 1∶500), and CREB (Cell Signaling, 1∶1,000) antibodies and developed with ECL (Perkin Elmer). Synaptosomes were isolated from at least three independent cultures, and immunoblots were least duplicated for quantitative analysis.
Neurons were transfected with the calcineurin activity biosensor, and FRET activity was measured at DIV14 according to a modification of the previously described method [31]. Neurons were pretreated with 2 µM TTX for 12, 24, or 48 h, or 5 µM FK506 for 12 h, and fixed with 4% paraformaldehyde. Images were captured by using Applied Precision PersonalDV live-cell imaging system in the Microscopy Core of New York University Langone Medical Center. The following formula was used to calculate the emission ratio:Pseudocolor images of the emission ratio were generated by using the ImageJ software, as previously reported [62].
DIV4 neurons were transfected with GCaMP5 (Addgene, 31788). Neurons were grown for 10–12 d after transfection in Neurobasal medium without phenol red and supplemented with B27 and 0.5 mM Glutamax. Glass-bottom dishes were mounted on a temperature-controlled stage on Zeiss Axiovert 200M and maintained at 37°C and 5% CO2 using a Zeiss stage incubator model S with CTI, digital temperature, and humidity controller. The imaging was captured for periods of 0.5 to 1.0 s depending on the intensity of the fluorescence signal using a 63× oil-immersion objective. One hundred images were obtained with a 1-s interval, and Ca2+ activity in the cell body (excluding dendrites) was analyzed using the ImageJ software. F0 was determined as the minimum value during the imaging. Total Ca2+ activity was obtained by combining 100 values of ΔF/F0 = (Ft−F0)/F0 in each image, and values of ΔF/F0<0.3 were rejected due to bleaching.
Neurons pretreated with 2 µM TTX for 6 or 48 h or 1 µM thapsigargin for 6 h were lysed with a nuclear preparation buffer A (10 mM Tris-HCl, pH 7.9, 1.5 mM MgCl2, 10 mM KCl, and 0.25% NP40). Nuclear fraction was collected by centrifugation, resuspended in a nuclear preparation buffer B (20 mM Tris-HCl, pH 7.9, 1.5 mM MgCl2, 420 mM KCl, 0.2 mM EDTA, and 20% glycerol), and analyzed by immunoblots.
Most statistical comparisons were analyzed with the GraphPad Prism6 software. Unpaired two-tailed Student's t tests were used in single comparisons. For multiple comparisons, we used one-way analysis of variance (ANOVA) followed by Fisher's Least Significant Difference (LSD) test to determine statistical significance. The Kolmogorov-Smirnov (K-S) test (http://www.physics.csbsju.edu/stats/KS-test.html) was used for comparisons of cumulative probabilities. Results were represented as mean ± s.e.m., and a p value<.05 was considered statistically significant.
|
10.1371/journal.ppat.1002503 | Immune Subversion and Quorum-Sensing Shape the Variation in Infectious Dose among Bacterial Pathogens | Many studies have been devoted to understand the mechanisms used by pathogenic bacteria to exploit human hosts. These mechanisms are very diverse in the detail, but share commonalities whose quantification should enlighten the evolution of virulence from both a molecular and an ecological perspective. We mined the literature for experimental data on infectious dose of bacterial pathogens in humans (ID50) and also for traits with which ID50 might be associated. These compilations were checked and complemented with genome analyses. We observed that ID50 varies in a continuous way by over 10 orders of magnitude. Low ID50 values are very strongly associated with the capacity of the bacteria to kill professional phagocytes or to survive in the intracellular milieu of these cells. Inversely, high ID50 values are associated with motile and fast-growing bacteria that use quorum-sensing based regulation of virulence factors expression. Infectious dose is not associated with genome size and shows insignificant phylogenetic inertia, in line with frequent virulence shifts associated with the horizontal gene transfer of a small number of virulence factors. Contrary to previous proposals, infectious dose shows little dependence on contact-dependent secretion systems and on the natural route of exposure. When all variables are combined, immune subversion and quorum-sensing are sufficient to explain two thirds of the variance in infectious dose. Our results show the key role of immune subversion in effective human infection by small bacterial populations. They also suggest that cooperative processes might be important for successful infection by bacteria with high ID50. Our results suggest that trade-offs between selection for population growth-related traits and selection for the ability to subvert the immune system shape bacterial infectiousness. Understanding these trade-offs provides guidelines to study the evolution of virulence and in particular the micro-evolutionary paths of emerging pathogens.
| Every pathogen is unique and uses distinctive combinations of specific mechanisms to exploit the human host. Yet, several common themes in the ways pathogens use these mechanisms can be found among distantly related bacteria. The understanding of these common themes provides useful concepts and uncovers important principles in pathogenesis. Here, we have made a cross-species analysis of traits thought to be relevant for virulence of bacterial pathogens. We have found that the infectious dose of pathogens is much lower when they are able to kill professional phagocytes of the immune system or to survive in the intracellular milieu of these cells. On the other hand, bacteria requiring higher infectious dose are more likely to be motile, fast-growing and regulate the expression of virulence factors when the population quorum is high enough to be effective in starting an infection. This suggests that infectious dose results from a trade-off between selection for fast coordinated growth and the ability to subvert the immune system. This trade-off may underlie other traits such as the ability of a pathogen to live outside the association from a host. Understanding the patterns shaping infectious dose will facilitate the prediction of evolutionary paths of emerging pathogens.
| Bacteria are a significant part of the human body, often establishing commensal or mutualistic interactions with it [1]. Yet, some species, or some strains within species, have a significant negative impact on the host while exploiting its resources. Such antagonistic associations lead to co-evolution between the two sides, often in the form of an arms race [2]. Pathogenic bacteria aim at exploiting the host, which usually involves eluding its defenses. The known mechanisms for immune evasion are varied and sophisticated, but nevertheless a few common themes emerge that are shared between phylogenetically distant bacteria (reviewed in [3]–[6]). Passive mechanisms of immune evasion include invading immune-privileged locations, antigenic variation, development of quiescent states, and modification of the cellular envelope. Protection from the immune system is even more effective when bacteria are able to subvert the immune system. Bacteria that are able to kill professional phagocytes or to survive/replicate in the intracellular milieu of these cells are mentioned throughout this text as being able to “kill or subvert professional phagocytes” or just able to do “immune subversion” (see also Discussion for possible limitations and extensions of this definition). Such bacteria may not search to escape the immune response but rather to stimulate it. This can be advantageous if the bacteria can grow inside phagocytes or if the immune response competitively disadvantages neighboring bacteria [7], [8]. Effective subversion of immune cells involves a variety of mechanisms including induction of stress response to escape reactive oxygen species, subversion of signaling pathways, inhibition of fusion between phagosomes and lysosomes, escape into cytoplasm, production of toxin-killing phagocytes and manipulation of apoptosis [9], [10]. Knowledge of the mechanisms involved in subverting the immune system is important to effectively control virulence in clinical settings, but precise identification of the common themes behind them is essential to understand the evolution of virulence and its mechanisms [11]–[13].
Commonalities among strategies used by bacteria to subvert the host's immune system suggest the existence of trade-offs shaping the pathogens' life-histories [14], [15]. Notably, investment in increasing transmission between hosts often leads to increased virulence at the cost of faster clearance (or host death) [16], [17]. Investment in the control of the host immune system most often comes at the cost of less efficient extraction of host resources for growth and transmission [18]. As a result, the forms of the interaction of the pathogen with the immune system, i.e. its efficiency, its cost and its mechanisms, are key parameters shaping the evolutionary dynamics of the pathogen-host association. One influential categorization of virulence themes separates frontal-attack from stealth pathogens, in an analogy with classical military strategies [19]. Frontal-attack pathogens would have limited capacity to deal with the immune system, especially its adaptive component, producing acute diseases by growing fast and by secreting extracellular toxins that contribute to overwhelm momentarily the host defenses. Stealth pathogens, on the other hand, would manipulate efficiently the immune system achieving more stable associations where slow growth is compensated by increased persistence. Naturally, there is a gradient between stealth and frontal-attack strategies resulting from the use of a multitude of combinations of a variety of mechanisms.
The positioning of a bacterial pathogen in terms of virulence strategies results from the diverse life-history trade-offs shaping its evolution. To understand these trade-offs one must be able to categorize virulence mechanisms and virulence strategies. This requires empirical analysis of the association between their infectivity and other physiological and genetic traits such as the mechanisms of immune subversion, effector secretion, growth rates, motility or genome size. The mechanisms used to manipulate host immune responses are expected to be key determinants of the size of the infectious dose required to start an infection. Accordingly, it has been proposed that bacteria coding for secretion systems injecting effectors directly into host cells using type 3 or type 4 secretion systems (resp. T3SS and T4SS) are associated with low infectious dose [20]. Direct secretion allows local manipulation of eukaryotic cells by a small number of bacteria. On the other hand, extracellular effectors provided by other secretion systems would have a global action requiring many cooperating bacterial cells to secrete enough molecules to be effective [20]. Since the size of the bacterial population is determinant in the success of infection, many pathogens show quorum-sensing regulated expression of virulence factors such as toxins and adhesins [21], [22]. In the same line, fast growth is expected to contribute much more to the success of frontal-attack than that of stealth pathogens. Additionally, bacterial motility facilitates dispersion and colonization [23]–[25] and allows counteracting peristalsis and mucus flow [26]. But motility appendages also pose problems: motility facilitates phagocytosis [27], flagella are costly [28], poorly adapted to intracellular environments, where bacteria tend to use actin-based motility [29], and flagellin-mediated activation of dendritic cells is rapid and highly deleterious to bacterial survival [30]. This suggests that flagella loss might be adaptive in pathogens with little need of motility, but strongly selecting for immune evasion. All these hypotheses are consistent with theoretical arguments and can be illustrated with examples. Yet, we know of no comparative study making a statistical assessment of the association of these variables with infectious dose.
It has been argued that to establish a theory for the evolution of virulence aiming at explaining differences and similarities between pathogens one has to bridge the gap between evolutionary ecology concepts and the mechanistic processes of virulence [11], [14], [15], [31]. In this study we only analyze bacterial pathogens. Bacteria use very diverse mechanisms of pathogenicity for which there is a large accumulated body of literature. Instead of attempting to analyze directly the trade-offs shaping the virulence strategies, which is difficult even for a single pathogen, we will use a statistical approach, based on the comparative method, to assess the associations between eight variables thought to have key roles in virulence strategies. (i) Infectious dose. (ii) The ability of the bacteria to kill professional phagocytes or to survive in the intracellular milieu of these cells. (iii) The minimum generation time, i.e. the lineage's ability to grow very fast whenever conditions are favorable. (iv) The use of quorum-sensing to regulate the expression of virulence factors. (v) Bacterial motility. (vi) The use of secretion systems able to deliver protein effectors directly into eukaryotic cells. (vii) The genome size, which correlates to the size of the functional repertoire of a bacterium. (viii) The route of exposure. Our immediate objective was to clarify the factors relevant for the understanding of variation in ID50 among bacteria. Our final goal was to complement current approaches, mostly focused on how a population of parasites evolves in relation to virulence, by way of an analysis focusing on the common trends observed in a very diverse panel of bacterial pathogens.
ID50 measures the minimum size of a population of infectious agents required to start an infection with 50% probability. We computed log-transformed average values of ID50 for all 48 bacterial pathotypes for which we could collect experimental values in the literature (Figure 1, Table S1 in Text S1). Infectious dose in humans may significantly differ from that of animal models. Within humans, ID50 varies with the host health state, genetic background and the route of exposure [32] (see Discussion). Therefore, we restricted our analysis to values obtained from experimental infections of healthy human individuals using natural routes of exposure (see Text S1 for details). The only exception concerns the use of ID50 data from experiments in rhesus monkeys for Helicobacter pylori [33]. Exclusion of this species has no effect on the conclusions of this work. The use of ID50 obtained from human hosts prevents spurious interpretations when animal models do not adequately represent the human host. We observe a wide range of ID50 values in our dataset (Figure 1), from 3 bacterial cells in Orientia tsutsugamushi [34] to more than 1010 in Gardnerella vaginalis [35]. We find no clear bimodal distribution separating high and low ID50 values. Bimodality of ID50 values could have arisen from perfectly dichotomic virulence strategies, e.g. stealth versus frontal [19]. Instead, our data is consistent with a continuum of ID50 values, suggesting that more complex trade-offs shape the virulence strategies of bacterial pathogens.
Genome size in bacteria is directly proportional to the number of encoded genes and therefore corresponds to its functional potential [36], [37]. In our dataset, the smallest genomes are found among obligatory pathogens whereas larger genomes are associated with facultative pathogens [38]. This could impact infectious dose. However, we found no significant correlation between genome size and ID50 (Spearman rho = 0.18, p = 0.23), or between the number of genes in genomes and ID50 (Spearman rho = 0.20, p = 0.19). This is in line with low infectious dose in bacteria with genome sizes as diverse as the Rickettsia (∼1 Mb) and Burkholderia pseudomallei (7 Mb). We will therefore ignore the variable genome size in the subsequent analyses.
We have also analyzed the association between ID50 and the route of exposure. We classed these routes into three categories: ingestion, inhalation or other routes (including intravenous and urogenital) (Table S1 in Text S1). There is a weakly significant difference in terms of ID50 between these groups (p = 0.024, Wilcoxon test; p = 0.04, ANOVA), such that bacteria that are ingested tend to have a higher ID50 (median 1 000 000 vs 1000 for inhaled and 40 for other routes). This difference is weak. Analysis using Tukey-Kramer HSD tests do not show significant differences in ID50 between pairs of classes (p>0.05), and pairwise t-tests are at the edge of statistical significance (p = 0.03). We have not found enough data to compare the ID50 of different routes of exposure for the same pathotype. Most bacteria have one single most frequent route of exposure, the one for which we collected the experimental data. As a consequence, the values of ID50 following unnatural routes are difficult to interpret in an evolutionary context. The higher ID50 among ingested bacteria, relative to the other routes is not very surprising given the effects of stomach acidity on bacterial viability [39].
Closely related strains of diverse pathotypes show large differences in ID50. An extreme case is provided by Escherichia coli pathotypes, including the Shigella, where ID50 varies between 10 and over 109 cells (Table S1 in Text S1). The correlation between the value of a trait and the organismal phylogenetic relatedness is given by its phylogenetic inertia. High inertia reflects important vertical inheritance, in which case closely related organisms resemble due to their recent common ancestry [40]. We assessed the impact of phylogenetic inertia on the ID50 values, i.e. we assessed if a significant fraction of the variance in ID50 values can be explained by evolutionary relatedness. For this, we built the phylogenetic tree linking the 48 pathotypes using the 16S rRNA subunit (see Materials and Methods) (Figure 2). High phylogenetic inertia should lead to clusters of ID50 in the phylogenetic tree. Instead, using both the Blomberg's K index [41] and Pagel's Lambda index [42] we find no significant association between variation in ID50 and phylogenetic relatedness (K = 0.000394, p = 0.33; Lambda = 8×10−05, p = 1), indicating no need to correct ID50 statistical analyses for phylogenetic structure. These results suggest that adaptation, e.g. by horizontal gene transfer and pathoadaptive deletions, can quickly erase the signs of vertical inheritance in ID50.
We found published experimental data indicating that 34 of the 48 pathotypes are able to kill professional phagocytes or to survive in the intracellular milieu of these cells (Table S2 in Text S1). These bacteria are expected to surmount more efficiently the immune response. We found a striking association between this trait and infectious dose (Figure 1). The set of bacteria with this trait has a median infectious dose of 250 cells versus about 30 000 000 cells for the remaining dataset (p<0.0001, Wilcoxon test) (Figure 3). This trait alone explains 56% of all variance in ID50 (R2 = 0.562, p<0.0001, ANOVA). There is a clear separation of the two groups of bacteria at the ID50 threshold of ∼106 bacterial cells. The two outliers are Neisseria meningitidis and Clostridium perfringens, having respectively unexpected low and high ID50 (Figure 3). Several concordant experimental works indicate that C. perfringens has ID50 values higher than 108 cells (see Table S1 in Text S1) in spite of its production of a toxin capable of killing macrophages [43]. Relative to other species in this study, Neisseria cells are specifically protected from complement by capsules of sialic acid and by the action of surface glycolipid lipopolysaccharides [44]. Also, N. meningitidis is capable of evading the immune system by antigenic variation [45]. Hence, the unexpectedly low ID50 value of N. meningitidis might result from its efficient use of other means of escaping the immune system.
We then tested the robustness of the association between ID50 and immune subversion. First, we removed Treponema pallidum, for which literature is somewhat equivocal on its ability to kill or subvert professional phagocytes (see comments in Table S2 in Text S1). Removing this species made no qualitative difference (R2 = 0.558, p<0.0001, ANOVA). We then further removed from the analysis the two pathotypes able to kill professional phagocytes but for which evidence of intracellular survival is lacking (Table S2 in Text S1), obtaining similar conclusions (R2 = 0.560, p<0.0001, ANOVA). Overall, these results are in excellent agreement with the hypothesis that the capacity of a bacterium to manipulate the immune system is a key determinant of the population size required for infection.
Bacteria often use secreted proteins to subvert or kill professional phagocytes [3]. In particular, it has been suggested that direct delivery of effectors into the eukaryotic cell allows bacteria to lower their ID50 [20]. This hypothesis predicts more frequent presence of T3SS and/or T4SS in low ID50 bacteria. It does not presume of the eventual presence or absence of other secretion systems. We therefore identified non-flagellar T3SS and non-conjugative T4SS in the genomes of these pathotypes (see Materials and Methods, Table S3 in Text S1, Table S4 in Text S1). This list includes all experimentally validated systems in the pathotypes, most of which were shown experimentally to be implicated in pathogenesis. Our analysis shows weakly significant lower ID50 for bacteria using T3SS or T4SS in pathogenesis (resp. medians of 150 and 80 000, p = 0.034, Wilcoxon test, p = 0.072, ANOVA). We tested the robustness of this analysis in several independent ways. Firstly, we excluded bacteria with cellular envelopes for which no direct delivery secretion systems are known (firmicutes, tenericutes, actinobacteria). Differences among the remaining (diderm) bacteria having and lacking T3SS/T4SS in terms of ID50 are not significant (p = 0.09, Wilcoxon test, p = 0.2, ANOVA). Secondly, we included type 6 secretion systems (T6SS) in the analysis, some of which are also capable of delivering effectors into eukaryotic cells [46], [47]. Bacteria with genomes encoding at least one of the three secretion systems (T3SS, T4SS or T6SS) are not significantly different from the others in terms of ID50 (p = 0.42, Wilcoxon test, p = 0.64, ANOVA). Thirdly, we removed from the analysis the three pathotypes lacking evidence of being able to survive inside professional phagocytes (Table S2 in Text S1) and this rendered the tests scarcely more significant (p = 0.03, Wilcoxon test, p = 0.04, ANOVA). Finally, bacteria with genomes encoding T3SS or T4SS have a slightly higher probability of being able to kill or subvert professional phagocytes (p = 0.054 and p = 0.026, Fisher's exact test, resp. for all genomes and for diderms only). We conclude that the evidence is weak for an association between the use of direct delivery protein secretion systems and low ID50 values.
The information for motility was taken from the literature [48] and checked by genome analysis (Table S4 in Text S1). Motile bacteria are associated with higher than expected values of ID50 (resp. medians 300 000 and about 150 for non-motile bacteria, p = 0.013, Wilcoxon test and p = 0.02, ANOVA, Figure 4). Motile bacteria are also less often able to kill or subvert professional phagocytes (p = 0.030, Fisher's exact test). Redoing the analysis by marking Yersinia pseudotuberculosis and Listeria monocytogenes as non-motile (Table S5 in Text S1), since they down-regulate motility when expressing virulence factors [49], [50], makes no qualitative difference (p<0.03, both for Wilcoxon test and ANOVA). Detailed statistical analysis of the effect of the differences between mechanisms of motility on ID50 was not possible because all but two bacteria in our dataset (Mycoplasma pneumoniae and Haemophilus ducreyi) use the same mechanisms of motility (flagella). Exclusion of these two bacteria did not significantly affect the association between motility and ID50 (p = 0.012, Wilcoxon test and p = 0.019, ANOVA) or the ability to kill or subvert professional phagocytes (p = 0.036, Fisher's exact test).
It has been proposed that frontal-attack pathogens compensate their high infectious dose and lower capacity to subvert the immune system by growing quickly. Fast coordinated growth could overwhelm the immune system and allow the pathogen's propagation before clearance [19], [20]. We thus tested the hypothesis that high ID50 pathogens tend to grow quickly under favorable conditions and to use quorum-sensing to regulate the expression of virulence factors. Indeed, the lowest experimentally determined minimal generation times [51](Table S5 in Text S1) are found among the pathogens with highest ID50 (Spearman's rho = −0.41, p<0.01, Figure 4). On the other hand, low ID50 bacteria include fast and slow growers. This is not unexpected since fast growers can grow slowly by way of genetic regulation whereas slow growers, by definition, never grow fast. This suggests that the capacity to grow fast is important for pathogens with high ID50 but less relevant for the others.
We then identified from the literature the bacteria with experimental evidence of using quorum-sensing to regulate the expression of virulence factors (Table S6 in Text S1). Such pathogens do have a significantly higher median infectious dose than the other bacteria (resp. 1 000 000 and 45, p<0.0005, both for Wilcoxon test and ANOVA). As expected, there is also a significant association between quorum-sensing and minimal generation times (p<0.005, both for Wilcoxon test and ANOVA) and between both variables and the ability of a pathogen to kill or subvert professional phagocytes (p<0.05, for both variables). Thus, motility, fast growth and quorum-sensing based regulation of virulence factors are associated with bacteria with high infectious dose.
Many of the variables studied in this work are associated with ID50 and with each other. To disentangle their respective effects we made a forward stepwise regression analysis of the effect of these variables on ID50 (same results for mixed and backward stepwise regressions). The ability of a pathogen to kill or subvert professional phagocytes is the major explanatory variable of infectious dose (R2 = 0.562, p<0.0001), followed by quorum-sensing based regulation of virulence factors (+10%, cumulated R2 = 0.657, p<0.001). This shows that ∼66% of the variance in the ID50 values of our dataset can be explained by just two variables. The other variables have no further significant contribution to the regression (using both F-test or Akaike's criterion [52], Table 1). Hence, immune subversion and quorum-sensing are sufficient to explain all variance that can be explained by our set of variables.
We used methodological approaches commonly used in evolutionary ecology and comparative genomics to quantify the association of certain virulence-related traits with infectious dose. Such empirical approaches may help bridge the gap between theory in evolutionary ecology and knowledge of the molecular mechanisms of virulence [3], [19], [31]. To test quantitatively these ideas we put together variables expected to be both relevant for the mechanisms of virulence and quantifiable in very different organisms. Biological processes linked with immune defense are complex and the precise mechanisms used by pathogens to confront it can be very diverse. As a result, finding variables that can highlight commonalities among pathogens is challenging. Here, we have focused our attention on one set of traits associated with immune subversion: the pathogen's ability to kill and/or survive and/or replicate in professional phagocytes. These traits have been assessed extensively among pathogens and are important for virulence strategies. Our simplified approach remains meaningful, since we were able to explain a significant fraction of the variance of infectious dose. Future work should aim at further including strategies such as phagocytosis avoidance, interference with opsonization, antigenic variation, induction of immunopathology or taking refuge inside non-immune cells [3], [10], [53], for which systematic cross-species data is not yet available.
The robustness of our findings depends on the quality of the underlying data. Some variables, e.g. genome size, motility, secretion systems, were quantified on the basis of extensive genome data and were in excellent concordance with the literature. Minimum generation times, quorum-sensing regulated expression of virulence factors and the capacity of bacteria to resist to professional phagocytes are all extensively studied in the literature. Naturally, we cannot exclude the possibility that future studies may lead to revision of some of these numbers. To assess the robustness of our findings we have made a number of supplementary analyses where we exclude more doubtful data points. These analyses invariably confirmed the main conclusions of this work. One further source of complications in this type of analyses results from the variation of traits within pathotypes. Lack of published data forced us in a few cases to use more than one strain of a given pathotype to gather the information on the full set of traits used in the statistical analyses. Variability between strains can add noise to the data and decrease association measures. Nevertheless, the traits with lower association measures in this study, motility, genome size, secretion systems, could be analyzed using data from both literature and genome data. The two sources of data were in general concordant suggesting the robustness of our analyses. Also, we found little variation in these traits among the genomes of the strains of the same pathotype. Estimation of accurate ID50 values for bacterial pathogens is more difficult because of a number of factors including variation in the route of exposure, host health condition and genetic background [32], [54]. To reduce these sources of variation we averaged over multiple references of ID50 and used experimental values obtained from natural exposure routes in healthy humans (see Text S1). It has been suggested that ID50 does not adequately characterize the relative hazard of pathogenic organisms in humans [55]. Our results argue strongly against this perspective. If variation in ID50 were biologically irrelevant, or its measurement too noisy, then one could not have explained 66% of its variance with two pertinent biological variables. The existence of quorum-sensing based regulation of virulence shows in itself the relevance of infectious dose for virulence strategies. For obvious ethical and technical reasons, new ID50 values in humans have become almost impossible to obtain. Therefore, it will be important to study the relation between ID50 data in humans, as used in our work, and data obtained from model organisms. This would allow extending the dataset of ID50 and broaden the scope of these analyses.
Motility is less frequently found among bacteria with low ID50 and among bacteria able to kill or subvert professional phagocytes. This is in agreement with a series of observations. Listeria monocytogenes and Yersinia enterocolitica switch-off motility when up-regulating virulence factors [49], [50]. Loss of flagella is associated with the emergence of Bacillus anthracis lineages within the B. cereus clade [56], and with immune evasion in Pseudomonas aeruginosa [57]. These results point to a trade-off between motility and infectiousness, whereupon low ID50 bacteria tend to be non-motile, presumably as a result of investment on immune evasion and the use of actin-based motility in intracellular environments [29]. On the other hand, high ID50 bacteria are motile, possibly because this facilitates grouping, chemotaxis, transmission and life outside the host.
We found little supporting evidence for the proposal that secretion systems capable of direct delivery of protein effectors into eukaryotic cells are important determinants of low infectious dose [20]. Protein secretion by T3SS and T4SS plays essential roles in ecological interactions of pathogens and mutualists and is widespread among proteobacterial pathogens [58]–[60]. However, many bacteria lacking such systems have low ID50 and efficiently subvert macrophages, e.g. Mycobacterium tuberculosis [61]. This does not necessarily contradict the idea that virulence strategies are affected by the range of secreted proteins leading to local versus global effects [20]. In fact, protein secretion in very viscous environments or inside professional phagocytes is effectively local, independently of being done by T3SS/T4SS or by secretion systems targeting proteins to the extracellular environment. While the categorization of global and local effectors might still be pertinent, our data suggests it is not directly traceable from the identification of the respective secretion system. Furthermore, many bacteria use multiple means of secretion. For example, H. pylori blocks antigen-dependent proliferation of T-cells and suppresses B-cell apoptosis [62]–[64], which depend respectively on the secretion of VacA into the extracellular space [65] and CagA into the eukaryotic cell by a T4SS [66]. Thus, whether the secretome acts locally or globally might be less dependent on the nature of the secretion system and more on the bacterial micro-environment and the coordinated action of the different effectors.
Selection for quorum-sensing based regulation of the expression of virulence factors among high ID50 pathogens is expected since these bacteria can only effectively infect their hosts when their quorum is above a certain threshold. Quorum-sensing may be required to the establishment of cooperative processes, e.g. productive extracellular secretion of virulence factors [67]. However, such cooperative behavior can be prone to exploitation, especially in the case of multiple infections [68], if cooperating pathotypes can be outgrown by non-cooperating pathotypes with shorter minimal generation times [69]. This might explain why high ID50 bacteria grow fast. Interestingly, we find that high ID50 bacteria are more likely to enter the human body by ingestion. Since the oral cavity and the gut are the most diverse environments of the human body, bacteria in these environments are likely to face social dilemmas more frequently, and this could place particularly strong selection on growth rates, motility and cooperation by quorum-sensing. Conversely, bacteria that are introduced by insect vectors in the body experience little competition from other pathogens, and we find that they tend to grow slowly and have low ID50. Fast growth among high infectious dose bacteria also creates the demographic conditions allowing the population to counteract rapid immune clearance, as proposed in the frontal-attack model [19]. On the other hand, investment in immune evasion carries a metabolic cost that might implicate slow growth among stealth pathogens [70]. Furthermore, persistence might be favored under slow growth [71]. This suggests a trade-off between growth rate and immune subversion that could drive the evolution of frontal-attack or stealth strategies.
The ability of bacteria to kill or subvert professional phagocytes has by far the largest explanatory power over ID50 variation in our analysis. The lack of correlation between these two traits and genome size suggests that none of it requires a large catalogue of genetic information. The lack of significant phylogenetic inertia in ID50 clearly shows the potential for rapid evolution of infectious dose, e.g. by transfer or deletion of virulence factors [72], [73]. The observed large range of ID50 values is likely to result from a conjugation of factors. Immune subversion results from mechanisms with different costs and efficacies. These mechanisms are involved in trade-offs with traits that are adaptive in non-antagonistic associations. On this line of thought, one is inclined to speculate that the development of subversion skills in obligatory pathogens allows infection at low infectious dose, at the cost of poor growth outside of the host. Facultative pathogens would require higher infectious dose because they tend to remain less competent at immune subversion. Bacteria within these latter clades often receive by horizontal transfer genes rendering them more competent at immune subversion. Occasionally, some of these lineages emerge as specialists that are apt to immune subversion. This will lead or be accompanied with the evolution of lower ID50. Examples of such lineages include the Shigella, Yersinia pestis and Bacillus anthracis. Such adaptive shifts lead to lower growth rates and sexual isolation [74]. These costs lead to further selection for increased competence at immune subversion because such lineages are less competitive in the original environment.
Data was retrieved from RefSeq GenBank Genomes (ftp://ftp.ncbi.nih.gov/genomes/Bacteria). Annotations were retrieved from the GenBank files and pseudogenes were ignored. In most cases there was more than one genome for each pathotype and in those cases we analyzed the available genomes (Table S3 in Text S1). In general the different genomes for a given pathotype produced concordant results in which case we picked one genome randomly for the phylogenetic analysis. However, in the few cases where different strains led to different results, we just analyzed the strains used to produce the infectious dose data. When even such information was unavailable we used the type-strain. The accession numbers of the reference strain are indicated in Table S3 in Text S1.
For each system we initially obtained a set of proteins known to be part of the core of the system (see Supplementary Material for details). Our references for core proteins were [75], [76] for T3SS [77], [78], for T4SS [79], for T6SS and [76] for flagella. For these core proteins, we performed all against all blast searches and psi-blast searches among genomes. The hits were then clustered by sequence similarity with MCL [80]. The resulting protein families were manually curated. We then constructed multiple sequence alignments of the families using MUSCLE [81] and subsequently manually edited the alignments using SEAVIEW [82]. For each of the protein families, we built sequence profiles with HMMER 3.0 [83]. We then used the same program to perform searches with our profiles in the genomic sequences. Sets of hits with c-val <10−3 and aligning at least 50% of the protein profile for a given genome were further checked for co-localization of the respective genes within a replicon. These clusters were then analyzed to pinpoint the relevant systems. T3SS and flagella were separated based on the identification of one gene specific of non-flagella T3SS (the secretin) and 3 genes specific of flagella [76]. T4SS associated with protein transport were identified apart from the conjugation systems by verifying the lack of a nearby relaxase in the genome [77], [78]. We then checked extensively that our list included all systems for which experimental evidence could be found in the literature for the pathotypes of interest, which was the case (see Supplementary Material).
We extracted one 16S sequence from one representative genome of each pathotype. The sequences were aligned with MUSCLE [81] and the alignment manually corrected. We built a phylogenetic tree using PhyML with the GTR+Γ(4) model [84]. We then applied two methods to evaluate the phylogenetic signal in the ID50 data: Blomberg's K statistic [41] and Pagel's Lambda [42]. K, Lambda and the respective p-values were computed in R using the ape package [85] and the phylosig function (http://anolis.oeb.harvard.edu/~liam/R-phylogenetics/phylosig/v0.2/phylosig.R).
Data on infectious dose, motility, quorum sensing and bacterial interactions with the immune system were retrieved from the literature. For motility and most traits we first followed Bergey's Manual of Systematic Bacteriology [48]. For infectious dose we started by using a few reference works [32], [34], [86]–[88] as well as the reference websites of the FDA (http://www.fda.gov/Food/FoodSafety/FoodborneIllness/FoodborneIllnessFoodbornePathogensNaturalToxins/BadBugBook/), the CDC (http://www.wadsworth.org/testing/biodefense/) and the Public Health Agency of Canada (http://www.phac-aspc.gc.ca/lab-bio/res/psds-ftss/). These were then complemented with direct searches of primary literature (see Table S1 in Text S1). All these searches were done using PubMed, the Web of Science and Google Scholar with appropriate keywords. Minimal generation time data was retrieved from the supplementary material of [51], with some new data retrieved from the primary literature. All data tables and bibliographic references are published in Supplementary Material (see Tables S1 to S6 in Text S1).
|
10.1371/journal.pntd.0005484 | An integrated overview of the midgut bacterial flora composition of Phlebotomus perniciosus, a vector of zoonotic visceral leishmaniasis in the Western Mediterranean Basin | The Leishmania developmental life cycle within its sand fly vector occurs exclusively in the lumen of the insect’s digestive tract in the presence of symbiotic bacteria. The composition of the gut microbiota and the factors that influence its composition are currently poorly understood. A set of factors, including the host and its environment, may influence this composition. It has been demonstrated that the insect gut microbiota influences the development of several human pathogens, such as Plasmodium falciparum. For sand flies and Leishmania, understanding the interactions between the parasite and the microbial environment of the vector midgut can provide new tools to control Leishmania transmission.
The midguts of female Phlebotomus perniciosus from laboratory colonies or from the field were collected during the months of July, September and October 2011 and dissected. The midguts were analyzed by culture-dependent and culture-independent methods. A total of 441 and 115 cultivable isolates were assigned to 30 and 11 phylotypes from field-collected and colonized P. perniciosus, respectively. Analysis of monthly variations in microbiota composition shows a species diversity decline in October, which is to the end of the Leishmania infantum transmission period. In parallel, a compilation and a meta-analysis of all available data concerning the microbiota of two Psychodidae genera, namely Phlebotomus and Lutzomyia, was performed and compared to P. perniciosus, data obtained herein. This integrated analysis did not reveal any substantial divergences between Old and New world sand flies with regards to the midgut bacterial phyla and genera diversity. But clearly, most bacterial species (>76%) are sparsely distributed between Phlebotominae species.
Our results pinpoint the need for a more exhaustive understanding of the bacterial richness and abundance at the species level in Phlebotominae sand flies in order to capture the role of midgut bacteria during Leishmania development and transmission. The occurrence of Bacillus subtilis in P. perniciosus and at least two other sand fly species studied so far suggests that this bacterial species is a potential candidate for paratransgenic or biolological approaches for the control of sand fly populations in order to prevent Leishmania transmission.
| The use of conventional microbiological methods gave us the opportunity to investigate the richness of symbiotic bacteria that inhabit the gut of P. perniciosus during its main period of activity. Our results were subsequently analyzed in the framework of what has been done on sand flies microbiota in order to validate our results and to address the question of the definition of the core bacterial microbiota of sand flies. A meta-analysis on the respective gut microbiota of Old and New World sand flies shows that the majority of bacterial species is observed only in one host whereas less than 8% are shared by more than two hosts. Our results pinpoint the need for a more exhaustive understanding of the microbiota composition and dynamic in phlebotominae, with the aim to implement new biological approaches for the control of sand fly populations in order to prevent Leishmania transmission.
| Sand flies are vectors of various pathogens, including arboviruses and bacteria, but are best known as the principal vectors of Leishmania, the etiological agent of leishmaniasis, a neglected tropical disease with clinical symptoms varying in form from cutaneous to visceral [1,2]. According to the most recent reports, leishmaniasis affects nearly 12 million people located in tropical, subtropical, and Mediterranean regions [3,4] with an estimated 350 million people at risk [5]. Among all vector-borne diseases, visceral leishmaniasis (VL) is the second leading cause of death after malaria, with an annual incidence of 500,000 cases and 60,000 deaths each year [3,4]. To date, no effective vaccine is available against leishmaniasis, and treatments mainly rely on chemotherapy using pentavalent drugs. Currently, the effectiveness of the treatment varies because of adverse side effects on patients and the emergence of parasite drug resistance [6,7].
Phlebotomus and Lutzomyia are the main sand fly genera involved in the transmission of Leishmania sp. in both the Old and the New World [2, 8–10]. Sand flies become infected when they blood feed on an infected host. Ingested amastigote parasites undergo a complex developmental cycle within the sand fly and are limited to the midgut of the insect [11]. Thus, the midgut of the vector is the first point of contact between ingested parasites and the apical surface of the intestinal epithelial cells of the vector. Bacteria have been isolated from the midgut of P. papatasi, a vector of Leishmania major, the etiologic agent of zoonotic cutaneous leishmaniasis (ZCL) [12], and studies have suggested a role for these bacteria in the immune response and homeostasis [12–15]. Female sand flies feed on blood for egg laying. In addition to blood, they take sugar meals derived from a number of different sources, including leaves, fruit, and aphid honeydew. Such food sources offer many opportunities to ingest microorganisms [16–18]. The microbiota found in sand fly guts could mirror their diets.
In low- and middle-income countries, such as Tunisia, large vector eradication programs are challenging owing to limited resources. New approaches to control vector transmission of Leishmania infantum are of major interest. These programs are needed to control the transmission of L. infantum in Tunisia. Paratransgenesis has been suggested as a feasible strategy for controlling the transmission of pathogens by arthropod vectors. This approach consists of the use of genetically altered symbiotic bacteria that secrete effector molecules that kill the infectious agents. Since these bacteria should co-localize with the pathogen and be transmitted vertically to the next generation, they are introduced into vectors to block pathogen transmission [19–20]. This "Trojan-Horse" approach was initially developed to interfere with the transmission of Trypanosoma cruzi by its triatomine vector [19]. Among possible bacterial species that could be considered as candidates for the development of a paratransgenic approach, Bacillus pumilus and Bacillus flexus were identified as the most frequent cultivable bacteria identified in the midgut of P. papatasi field-collected from Tunisia, Turkey, and India [21]. In addition, Bacillus subtilis isolated from Phlebotomus argentipes is currently being considered as a possible candidate for paratransgenesis aimed at preventing Leishmania donovani transmission [22,23].
In North Africa, Phlebotomus perniciosus is the main vector of L. infantum, the etiologic agent of zoonotic visceral leishmaniasis (ZVL) [24]. We sought to develop a paratransgenic platform to control the transmission of L. infantum by P. perniciosus. Here, we assessed the richness of bacterial species of laboratory-reared and field-collected sand flies. We investigated the monthly variations of the bacterial diversity carried by sand flies in an endemic area of ZVL in Tunisia, during the period of Leishmania infantum transmission. We analyzed these new data within the context of previously published studies on the microbiota of sand flies.
Sand flies collection: Laboratory-reared P. perniciosus (Tunisian strain) was obtained from a colony maintained at the Vector Ecology Laboratory of Pasteur Institute of Tunis [25]. Phlebotomus perniciosus individuals were also collected in a sheep shelter in the village of Utique located in Northern Tunisia (37°08’N, 7°74’E), with the owner consent, by using CDC traps. Sand fly trapping was performed from dusk to dawn one night per month, from July to October 2011. This period corresponds to the period of main activity of P. perniciosus in Tunisia [26]. Field-collected sand flies were brought alive to the laboratory. However, as it is difficult to determine the age of field-collected sand flies, we arbitrarily attribute the day of their sampling as the day one. All field-collected sand flies were dissected within three days after collection. Laboratory-reared sand flies were dissected three-to-seven days after their emergence. Prior to dissection, each sand fly was rinsed in 70% ethanol for 3 minutes, followed by three successive rinsings in sterile PBS. Sand flies were then dissected on ice under stereo-microscope, in order to remove the midgut for bacterial identification and the genitalia for morphological identification to species level [26,27]. Only P. perniciosus females were used.
Gut dissection: Each sand fly gut was individually placed in 1.5 ml microcentrifuge tubes containing 200 μl of sterile PBS (pH 7.3), homogenized with a disposable pestle, and diluted from 10−1 to 10−10 in 200 μl PBS. Each homogenate was plated onto individual 1.5% agar plates with TSA (Trypticase Soy Agar), PCA (Plate Count Agar), YMA (Yeast Mannitol Agar) or Luedemann medium and incubated at 30°C for 2 to 4 days in aerobic conditions. Individual colonies were selected and used for further identification.
Chromosomal DNA extraction was performed as previously described [28]. After overnight incubation at 30°C in TSA, PCA, Luedemann or YMA medium, colonies were suspended in 500 μl of TE buffer (10 mM Tris-HCl, 0.1 mM EDTA, pH 8) to which 20 μl of lysozyme (35 mg/ml) was added and incubated at 37°C for 30 min. Then, 40 μl of sodium dodecyl sulfate (SDS 10%) and 5 μl of freshly prepared proteinase K (10 mg/ml) were added, and the solution was incubated at 30°C for 30 min. The solution was homogenized after the addition of 100 μl of 5 M NaCl and 80 μl of CTAB/NaCl (10%/0.7 M) and incubated at 65°C for 10 min. DNA was purified by the addition of phenol-chloroform-isoamyl alcohol (25:24:1, pH 8.0), followed by chloroform-isoamyl alcohol (24:1) and then precipitated by the addition of 0.6 volumes of isopropanol. DNA pellets were washed with 200 μl of 70% ethanol and dried at 37°C before being resuspended in TE buffer (10 mM Tris-HCl, 0.1 mM EDTA, pH 8) and stored at -20°C. Total DNA extraction for the Denaturing Gradient Gel Electrophoresis (DGGE) analysis was conducted on whole midguts dissected from sand flies using the same total DNA extraction protocol described above [28].
Fig 1 summarizes the procedure used for the isolation and identification of bacterial species. A total of 180 field-collected and 35 colonized P. perniciosus females were processed. From field-collected sand flies, 135 guts were used for culture-dependent identification and 45 guts were analyzed by DGGE, a culture-independent method. The 35 samples from colonized P. perniciosus were processed only for culture-dependent identification.
The length and sequences polymorphisms of the Intergenic Transcribed Spacers (ITS), located between the 16S and 23S rRNA, is quite often due to the presence of tRNA genes. PCR amplification of the 16S-23S intergenic transcribed spacer regions between the rRNA genes (ITS) was performed for screening the bacterial phylotype diversity [29–31]. The universal primers, ITSF (5’-GTCGTAACAAGGTAGCCGTA-3’) and ITSR (5’-CAAGGCATCCACCGT-3’), are complementary to nucleotide (nt) positions 1423–1443 of the 16S rDNA and nt positions 38–23 of the 23S rDNA of Escherichia coli, respectively [30]. Each reaction tube contains 1X PCR buffer (Invitrogen), 2 mM MgCl2, 0.2 mM deoxynucleoside triphosphate mix, 0.1 μM of each primer, 0.5 U of Taq polymerase (Invitrogen) and 400 ng of DNA extracted from single colonies. The total volume was adjusted to 25 μl. Amplification parameters were as follows: initial denaturation at 94°C for 5 min, followed by 35 cycles at 94°C for 30 s, 50°C for 30 s, 72°C for 45 s, with a final extension step of 10 min at 72°C, using an ABS2720 thermocycler.
Amplification of the 16S rDNA gene was carried out with universal primers SD-Bact-0008-a-S-20 and S-D-Bact-1495-a-S-20 [32]. Each reaction tube contained 1x PCR buffer (Invitrogen), 0.5 μM of each primer, 2.5 mM MgCl2, 200 ng of purified DNA, 0.2 mM dNTPs and 0.3 units of Taq polymerase (Invitrogen) and the total volume was adjusted to 25 μl. Samples were amplified according to the following cycle: an initial denaturation step at 94°C for 10 min, followed by 35 cycles at 94°C for 1 min, 55°C for 1 min, 72°C for 1 min and a final extension step of 10 min at 72°C, using an ABS2720 thermocycler. PCR amplicons were then purified using the QIAquick PCR Purification Kit (Qiagen) and sequenced.
Amplification of the V3-V5 region of the 16S rDNA: PCR amplification targeting the 16S rDNA genes was performed using the universal primers specific to the bacterial domain: 907r (5’-CCGTCAATTCCTTTGATGTTT-3’) and 357f (5’-TACGGGAGGCAGCAG-3’) [33]. A 40-bp GC-clamp was added to primer 357f to avoid complete denaturation of the DNA and allow the separation of DNA strands during migration in denaturing conditions [34–36]. Each reaction tube contained 1x PCR buffer (Invitrogen), 2.5 mM MgCl2, 0.12 mM dNTPs, 0.3 mM of each primer, 1 U of Taq DNA polymerase (Invitrogen) and 50 ng of DNA in a final volume of 50 μl. Amplification parameters were as follows: an initial denaturation step at 94°C for 4 min, 10 cycles at 94°C for 30 s, 61°C for 1 min and 72°C for 1 min, followed by 20 cycles at 94°C for 30 s, 56°C for 1 min and 72°C for 1 min. At the end of these cycles, a final extension step was performed at 72°C for 10 min.
DGGE analysis: PCR products were run on a 7% polyacrylamide gel in a 40%–60% denaturing gradient of urea and formamide for 16S rDNA analysis. DGGE was performed using a BioRad DCode Universal Mutation Detection System at 100 V at 59°C for 17 hr, in 1.0 × TAE buffer (20 mmol/L Tris, 10 mmol/L acetate, 1 mmol/L EDTA pH 7.4). After electrophoresis, gels were stained for 30 min with ethidium bromide.
Identification of the DGGE Bands: Excised bands of DGGE gels were washed twice with 1 mL sterilized distilled water in a 1.5-mL tube. A portion of the gel piece (< 1 mm3) was used as the direct template for PCR to recover DNA fragments. Amplification conditions for the V3-V5 region were as follows: an initial denaturation step at 94°C for 4 min followed by 35 cycles at 94°C for 30 s, 56°C for 1 min and 72°C for 1 min and a final extension step at 72°C for 10 min. Primers were identical to those described above except that the forward primer had no GC-clamp attached. The amplified products were purified with the QIAquick PCR Purification Kit (Qiagen) and then sequenced.
The 16S rDNA sequencing was carried out using the BigDye Terminator v3.1 Cycle sequencing Kit and the ABI 3130 sequence analyzer. The partial 16S rRNA gene sequences were compared with sequences available in the ribosomal database, release 11.4. Isolates were assigned at the species level on the basis of the 16S rRNA gene sequence similarity of the available sequences in the ribosomal database, measured by using the Seqmatch tool of RDP [37] (https://rdp.cme.msu.edu/). In addition, the partial 16S rDNA sequences were submitted to the BLASTn server of NCBI, using the 16S ribosomal RNA database (Bacteria and Archea) (http://blast.ncbi.nlm.nih.gov/Blast.cgi). The nucleotide similarity thresholds of the 16S rDNA sequences with the nearest neighbor were: ≥ 95% and 97.5% [38] applied at the genus and species levels, respectively.
All the analyses were conducted with the R-vegan package, v. 2.0–10 [39]. α-diversity was calculated using Shannon’s and Simpson’s diversity indices. Correspondance analysis (CA analysis) on the monthly data was carried out with the FactomineR package (https://cran.r-project.org/web/packages/FactoMineR/) using the R language (http://www.R-project.org).
All the published data concerning bacterial species identification associated with Phlebotomus and Lutzomyia species (the only two genera for which we have data) were compiled and analyzed. Studies describing the identification of the midgut bacteria at the family, class or phylum level were not considered. To assess bacterial richness associated with the adult sand fly, data were collected without taking into account the method of bacterial isolation (culture-dependent vs culture-independent) and identification (DNA sequencing of 16S rDNA, bacteriology). The overall dataset used in our analyses included ten Phlebotominae (L. cruzi [40], L. longipalpis [41,42], L. evansi [43], P. argentipes [22], P. duboscqi [44], P. halepensis [45], P. papatasi [21, 45–48], P. sergenti [45], P. perfiliewi [45], P. chinensis [49] and P. perniciosus) and their associated microbiota for the present study. Bacterial richness is visualized through network analysis using Cytoscape (http://www.cytoscape.org/) [50]. To achieve this goal, data were extracted from our own database (focused on Phlebotominae) as CSV files, containing vertices or nodes (representing hosts and bacteria) and edges (representing links). These files were loaded into Cytoscape v 3.4.0, a tool specializing in graphical representation. This graph was modified to keep only one edge between host and bacteria. Bacterial nodes were colored to show their degrees of interaction with hosts.
First, we determined the number of Colony Forming Units (CFU) of each individual midgut using PCA medium; they ranged from 5 to 121 per individual sand fly midgut. A total of 441 and 115 independent colonies were obtained from field-collected and colonized sand flies, respectively (Fig 1). Examples of the different types of bacterial colonies are shown in Fig 2A.
ITS-PCR analysis was used to dereplicate strain diversity among the 556 colonies. Each ITS profile is composed of one to five reproducible bands that display apparent molecular weights ranging from 50 to approximately 1,500 bp; these bands represent a phylotype (Fig 2B). Among the 441 colonies isolated from field-collected P. perniciosus, 25 distinct ITS-PCR profiles were identified (Fig 2B see *). Of a total of 115 independent colonies from colonized P. perniciosus, only 6 independent phylotypes were identified (Fig 2B). When possible, the 16S rDNA locus of a sample representative of each ITS-PCR profile was further amplified to attempt identification at the genus and species levels. All the bacterial colonies isolated from sand fly midguts belong to three phyla: Firmicutes, Actinobacteria, and Proteobacteria. For the field-collected P. perniciosus, the calculated midgut bacterial composition was: Firmicutes (53.5%), Actinobacteria (15.2%) and Proteobacteria (31.3%). For laboratory-reared sand flies, the midgut bacterial composition was Firmicutes (66.7%) and Proteobacteria (33.3%). We did not isolate bacteria belonging to the Actinobacteria phylum from the midgut of laboratory-reared sand flies. Nevertheless, only 35 females were processed and bacterial colonies were isolated solely using the PCA medium, which might have influenced the output of our analysis.
The results of isolating bacterial species from the midguts of field-collected and lab-reared P. perniciosus, performed in a culture dependent manner, are shown in Table 1. Of the six bacterial species identified in laboratory-reared sand flies (Table 1), three are also found in the midgut of field-collected sand flies (Stenotrophomonas maltophilia, Bacillus sp., Lysinibacillus sp.) (Table 1). We isolated Veillonella sp. and Burkholderia fungorum only from the laboratory-reared sand flies (Table 1). Overall, the bacterial richness recorded in field-collected sand flies, at the species level, seems to be more important than in laboratory-reared flies, even if the total number of lab-reared flies studied is small.
To further characterize the bacterial richness in field-collected sand flies, a culture-independent method (DGGE) was performed on the 45 dissected midguts (Fig 1). Despite variation in the number and intensity of the bands detected, the observed DGGE profile is composed of at least 12 distinguishable bands. Among these bands, six were successfully sequenced. In addition to bacteria already identified using culture-dependent methods, like Enterococcus sp. (Accession N° KY303721 and KY303722), we also identified Wolbachia sp. and Ehrlichia sp. (Fig 3). BLASTing the sequence from the DG5 band (459 bp Accession N° KY303723) indicated an overall similarity of 99% with the Pel strain of Wolbachia, isolated from Culex quinquefasciatus (NR-074127.1). The same query on the RDP database disclosed 98% similarity with Wolbachia inokumae DQ402518, which was already found in field collected P. perniciosus from Marseille, France [51]. A search in the RDP database with the sequence obtained from the DG1 band (718 bp, Accession N° KY322518) produced hits with various species of Ehrlichia, including 96% similarity with Ehrlichia canis-M73226. A similarity of 96% with Ehrlichia ewingii (NR-044747) was found when BLAST analysis was performed on the 718-bp DNA fragment (Fig 3). To our knowledge, this is the first report of the presence of Ehrlichia sp. DNA in sand fly midguts.
A meta-analysis was conducted to assess the bacterial species diversity of Phlebotomus and Lutzomyia microbiota. This analysis included previously published studies concerning adults of seven phlebotomine sand fly species (P. argentipes, P. chinensis, P. duboscqi, P. halepensis, P. sergenti, P. papatasi, P. perfiliewi) our study reported on P. perniciosus and previously published data reported on three Lutzomyia species (L. cruzi, L. evansi, L. longipalpis) [22,40–49]. Owing to the small number of studies conducted on the microbiota of Phlebotominae and the lack of information about sex in several cases, we chose to not take into account the genera of the specimen in order to highlight trends. This analysis shows that most bacteria identified from Old World sand fly species belong to the Firmicutes phylum, 39,8% (Fig 4A left panel) (41–42% for our study on P. perniciosus) and the Proteobacteria phylum, 46,8% (Fig 4A right panel) (37% for our study on P. perniciosus). Bacteria of the Bacteroides genus are not recorded in the present study and represent only 0.5% calculated from the pooled published data (i.e., Meta-set). Bacteria of the Actinobacteria phylum account for 11.9% of the Meta-set (20%, in our study on P. perniciosus). In Lutzomyia sp., more than 57% of bacteria currently characterized, belong to the Proteobacteria phylum (Gram-negative bacteria), Firmicutes representing 23.9% and Actinobacteria 5.6%. Bacteria of the Bacteroidetes phylum account for approximately 6% of the species in Lutzomyia but only 0.5% in Old World sand fly species (Fig 4A). Nevertheless, we did not notice significant differences in Bacterial phylum composition between Old World and New World sand flies (chi-squared = 5.8226, df = 2, p-value = 0.0544) (Fig 4A).
Within the Proteobacteria phylum, compared with the alpha-, beta- and deltaproteobacteria identified, gammaproteobacteria are by far the most frequently found bacterial class in Lutzomyia and Phlebotomus species (Fig 4B right panel). Within the Firmicutes phylum, a higher number of classes is observed in Lutzomyia, with bacteria belonging to Negativicutes, Bacilli, and Clostridia. The Bacilli class is almost the sole representative of Firmicutes class in the Old World sand fly species (Fig 4B left panel).
In the Gammaproteobacteria class, bacterial species of the Enterobacteriaceae family are the most represented (more than 60% so far isolated) in both the Old and New World sand fly species, followed by bacteria belonging to the Pseudomonadaceae and Moraxellaceae families (less than 20%) and Xanthomonadaceae, with less than 10% (Fig 4C right and left panel). Bacteria of the Coxiellaceae family have only been isolated from Old World sand fly species.
Our meta-analysis shows that bacteria of the Serratia genus has been identified in almost all Old World and New World sand fly species so far studied, but Serratia marcescens was characterized only in P. duboscqi. Bacteria of the Enterobacter genus are found in five of the eleven sand fly species studied. Enterobacter cloacae and Enterobacter aerogenes were recorded in three sand fly species, while Enterobacter gergoviae and Enterobacter ludwigii were found in two sand fly species. The most frequently isolated bacteria in sand flies are Stenotrophomonas maltophilia (Pseudomonadaceae), followed by Escherichia coli (Enterobacteriaceae), Klebsiella ozaenae (Enterobacteriaceae), and Staphylococcus epidermidis (Staphylococcaceae). Bacillus subtilis (Bacillaceae) and Acinetobacter baumannii (Moraxellaceae) were identified in three of the eleven sand fly species currently studied (Fig 5).
Despite that neither statistical nor bioinformatics analysis were performed to test the existence of biological patterns between sand fly species and their corresponding microbiota, the network representation displayed in Fig 6 suggests some relationships between the eleven studied New World and Old World sand fly species and the bacteria inhabiting their guts. As an example, the Bacillus genus is found in almost all Old World sand fly species. Bacillus subtilis was isolated from P. halepensis, P. papatasi and P. perniciosus. Bacillus megaterium was isolated from P. papatasi and P. argentipes. Bacillus oleronius, Bacillus brevis, Bacillus endophyticus, Bacillus pumilus, Bacillus circulans, Bacillus mojavensis, Bacillus firmus, Bacillus licheniformis, Bacillus vallismortis, Bacillus cereus, Bacillus amyloliquefasciens, Bacillus altitudinus and Bacillus flexus were isolated only from P. papatasi. Bacillus closei and Bacillus mycoïdes were isolated only from P. argentipes. Bacillus oleronius, Bacillus galactosidilyticus, and Bacillus casamensis were isolated only from P. perniciosus (Fig 6). Bacillus thuringiensis is the only species of the Bacillus genus that was isolated from L. evansi and P. chinensis, two sand fly species belonging to the New World and Old World, respectively (Fig 6).
Nevertheless, in the Meta-Set no significant differences in the microbiota composition at the genus level was observed as demonstrated in the Fig 7 that depicts the Shannon (left) and Simpson (right) indices of diversity for Old World (i.e Phlebotomus) and New World sand flies (Lutzomyia).
Fig 8A depicts the monthly proportions of each previously characterized bacterial species. To perform this analysis, bacterial species identification was linked to each individual phylotype recorded by PCR-ITS analysis. Then, the number of colonies harboring the same ITS-PCR profile was determined and their monthly proportion calculated.
In July, among the nine bacterial species identified in midguts of wild P. perniciosus, only bacteria belonging to two phyla, Proteobacteria and Firmicutes, were found. In September, the bacteria belonged to the Actinobacteria, Proteobacteria and Firmicutes phyla. In October, only bacteria belonging to Firmicutes were isolated (Fig 8A). Shannon and Simpson indices of diversity confirmed a lower diversity in October (Fig 8B). Correspondence analysis performed on data from monthly dynamics highlights the contrasting situation between October and September/July, depicted on the first axis. October is characterized by B. oleronus and unidentified Bacillus and Lysinibacillus species. The differences between July and September/October is depicted by the second axis of the correspondance analysis (Fig 8C). This analysis further reinforces our observation of a monthly evolution of the microbiota within the sand fly gut.
Currently, there are considerable efforts to study arthropod gut microbiota, especially those of medically important vectors. The microbiota is considered in the context of possible extended phenotypes conferred on the insect hosts that allow niche diversification and rapid evolution [52]. As early as 1929, Adler and Theodor [53] suggested that the presence of microorganisms in the guts of sand flies might impact the development of the parasite Leishmania. In the mid-80s, the team of Shlein and collaborators [12] observed a large number of “germ (bacteria) contaminations” in guts of wild-caught female P. papatasi. However, the composition of the sand fly’s gut microbiota was studied much later by Dillon et al. [54]. Ochrobactrum sp. was the first bacterium to be isolated from the midguts of P. duboscqi, a proven vector of L. major in Sub-Saharan Africa [44], and from other sand fly species [40], including laboratory-reared Lutzomyia longipalpis [55] and New World L. intermedia [56]. This bacterium, probably ingested by larva, passes to nymphs and up to the adults through transstadial transmission [44]. Recently, several publications were dedicated to the study of the microbial composition associated with the digestive tract of sand flies. Only a few studies concerning biotic and abiotic factors influencing the composition of the bacterial community of the midgut of sand flies were performed.
This study brings additional evidence on the microbiota composition in the midgut of P. perniciosus. Our results suggest that lab-reared P. perniciosus display a lower bacterial richness in their midgut than in field-collected sand flies. This difference is likely due in part to the type of food diet ingested by larvae and adults during rearing. In the laboratory, P. perniciosus larvae are fed sterile chaw (50% rabbit food plus 50% rabbit feces). After emergence, glucose is the main source of carbohydrates for adults [25]. Under natural conditions, larvae, as well as adult P. perniciosus, have a wide variety of diet including various sources of blood meals [18,57]. Therefore, the nature of the feeding regimen leads to a striking contrast between field-collected and laboratory-reared sand flies, which might explain the lower bacterial richness observed in colonized sand flies.
Among the bacterial genera found associated with P. perniciosus midgut, we identified isolates belonging to the Burkholderia genus and Stenotrophomonas maltophilia, an aerobic non-fermentative and a Gram-negative bacterium. We also identified bacterial species commonly found in the digestive tract of humans or other mammals, but which have not yet been described in the midguts of sand flies, like Veillonella sp. In addition Sporosarcina koreensis, Rhizobium pusense and Nocardia (a rare endophyte bacterium) have never been found in association with the sand fly gut. The richness of sand fly-associated bacteria, illustrated by the meta-analysis, point to some interesting outcomes. In Lutzomyia sp., more than 57% of identified bacteria belong to the Proteobacteria phylum (Gram-negative bacteria), whereas for Old World sand fly species, including P. perniciosus, Proteobacteria (47%) and Firmicutes (40%) are preponderant. Such a difference in the gut microbiota composition might be due to a number of factors, including the long divergence of evolution between the two subgenera [2]; some new studies are required to assess this observation. Another surprising finding is the high richness of Bacillus species found in Old World sand flies, in which the majority of these bacteria are host specific (Fig 6). Stenotrophomonas maltophilia, that has emerged as an important opportunistic pathogen [58] was found to inhabit the gut of most of the sand fly species so far studied. This bacterial species is a common microorganism found in aqueous habitats, plant rhizosphere, animal food and water sources. Thus, delineating the origin of the colonization of midguts by S. maltophilia and evaluating its role, if any, in the sand fly biology and physiology are of major importance.
Our results have, for the first time, disclosed monthly variation in the diversity of the sand fly’s gut microbiota, during the period of transmission of L. infantum. In fact, it appears that the richness of the gut microbiota is related to sand fly seasonal activity. This diversity could reflect the environmental conditions, such as temperature and humidity, but it may also be linked to variations in plant cover, such as flower blooming. At the beginning of the sand fly season (July), Ochrobactrum sp. and Serratia sp., both affiliated with the Proteobacterium phylum, were the principal bacterial genera isolated. The peak of activity of P. perniciosus occurs in September and October, a period that also corresponds to the L. infantum transmission season [59]. The analysis of the gut bacterial flora of sand flies collected in September reveals a higher diversity (Fig 8). In particular, we recorded the presence of Microbacterium, Micrococcus, Kocuria, Stenotrophomonas, and Bacillus sp. (Actinobacteria, Proteobacteria and Firmicutes). In July, O. intermedium and Serratia sp. are the dominant bacteria genera in the midgut of P. perniciosus and these bacteria became undetectable towards the main peak of sand fly activity identified in Tunisia, i.e., during the months of September and October [59]. The prevalence of L. infantum infection in the P. perniciosus population increases over the summer months and reaches a peak of 9% during September-October [60,61]. Ochrobactrum intermedium has been found previously to negatively affect Leishmania mexicana infection in L. longipalpis [55]. Certain strains of S. marcescens are capable of producing a pigment called prodigiosin, which ranges in color from dark red to pale pink depending on the age of the colonies. Derivatives of prodigiosin have recently been found to have anti-T. cruzi and anti-Leishmania (Leishmania mexicana) activity by promoting mitochondrial dysfunction leading to parasite programmed cell death [62,63]. To what extent such interplay between the bacterial colonies that exert toxic effects might interfere with the dynamic of L. infantum transmission awaits further investigation.
Sand flies are vectors of medical and veterinary importance. Understanding the establishment of the sand fly microbiota is critical towards clarifying underlying details of sand fly Leishmania-microbiota interactions [64]. Bacteria such as O. intermedium, which has been previously characterized in the guts of larvae, pupae, and adults of P. duboscqi [44], is an opportunistic pathogen to humans [65]. Serratia sp., an entomopathogenic bacteria found in this study, has been previously isolated from L. longipalpis [40] and L. intermedia [56]. Bordetella avium, isolated only once from a specimen caught during July, has never been previously isolated from sand fly midgut microflora. Bordetella avium is a highly pathogenic bacterium, causing the avian bordetellosis [66]. Klebsiella ozaenae, known also as a human pathogenic bacterium, has been found in four out of the ten studied sand fly species (not isolated in this study). K. ozaenae was isolated from the midgut of gravid and freshly fed females of P. papatasi and P. halepensis [45] and from some Lutzomyia species (Fig 5). Klebsiella species are ubiquitous in nature [67,68] and are recorded in all habitats where sand flies proliferate. Moreover, the presence of K. ozaenae in the midgut of gravid females [45] will highlight their capacity to survive in the gut of this insect. Nevertheless, as for all bacterial species known to be etiological agents of human diseases, the sole observation of their presence in sand fly gut is not sufficient to incriminate sand flies as a potential vector but gives information on the bacterial dissemination via blood-feeding insects. The data collected are not sufficient to incriminate sand flies as a biological vector of K. ozaenae but are enough to raise suspicion regarding their role in the dissemination of K. ozaenae. Furthermore, whether certain clinical outcomes from leishmaniasis may be linked to bacteria potentially deposited during the Leishmania-infected sand fly bite still remains to be fully investigated [68]. These studies will not only shed light on the effect of the gut bacterial community on the sand fly fitness but also on the establishment and the transmission of Leishmania parasites in endemic areas.
This meta-analysis aimed to identify the best bacterial candidate for a paratransgenic approach. Our study is based on data aggregated from various publications that use culture-dependent and culture-independent methodologies and various set of technical approaches used to study the sand fly microbiota. For these reasons, conclusions raised with this study should be taken with caution and analyzed in the light of the limitations and pitfalls inherently associated with the compilation of heterogeneous data. Among limitations, some are linked to the physiological state of the sample. The gut microbiome is highly dynamic [69] and therefore influences the outcome of the analysis. When using a culture-dependent approach, we have to keep in mind that only 20% of environmental bacteria can be grown on a growth medium [70]. Therefore, the composition of the microbiota is not a direct reflection of the bacterial community structure (abundance and richness) inside the insect, but an altered version of the ecosystem from where they came. Nucleic acid-based analysis, involving historically used methods (such as construction and Sanger sequencing of metagenomic clone libraries, automated ribosomal internal transcribed spacer analysis (ARISA), terminal restriction fragment length polymorphism (T-RFLP), denaturing gradient gel electrophoresis (DGGE)) and next generation sequencing technology require a critical step that must combine an efficient cell disruption without DNA degradation and uniform nucleic acid extraction. Unfortunately, no consensus protocol for microbial DNA extraction of insect-associated microbiota is currently available [70]. Although 16S rRNA gene sequencing is highly useful with regards to bacterial classification, it has a low phylogenetic power at the species level for some genera [71,72]. Depending on the 16S rRNA variable region targeted and the database used to perform the taxonomic profiling, misassignation of bacterial OTU at the species level could be frequent [73]. Nevertheless, taking into account all the above mentioned limits and pitfalls, we think that an exhaustive approach aimed at collecting a maximum of data on the microbiota of sand flies will give key information on the most commonly identified bacteria in sand fly species and those that are more specific.
Our groups are interested in the development of a paratransgenic platform to control the transmission of leishmaniasis. To that end, a strain of the non-pathogenic Bacillus species (Bacillus subtilis), isolated from P. papatasi, is proposed as a possible candidate for paratransgenic approach. In this study, we isolated B. subtilis from P. perniciosus midgut, in addition to other Bacillus species (Bacillus oleronius, Bacillus casamensis, Bacillus galactosidilyticus and Bacillus sp.). Bacteria belonging to the Bacillus genus seem to display a host-specific distribution, with only B. subtilis being isolated in more than one sand fly species (P. halepensis, P. papatasi, and P. perniciosus). In addition, we observed that no bacteria belonging to the Bacillus genus have been characterized to date in adult New World sand fly species. Therefore, even if this bacterium possesses the main advantages of being non-pathogenic, easy to cultivate and to perform genetic manipulation, its use for paratransgenic control of Leishmania can be challenged by its capacity to establish long-term colonies in the gut of various sand fly species. In particular, if a paratransgenic approach is developed using B. subtilis as a host, it will be essential to probe its capacity to efficiently colonize the gut of Lutzomyia species and of other Old World sand fly species in which this bacterium has yet not been found in the gut. Thus, it will be of major epidemiological importance to develop a regional strategy for each endemic area with different bacterial isolates.
The knowledge of interactions between sand flies, Leishmania and nonpathogenic microorganisms that inhabit the gut will help to delineate an appropriate bacterial host recipient that can be used for paratransgenesis designed to prevent Leishmania transmission. The identification at the species level of the midgut’s cultured flora of P. perniciosus, linked to its seasonal variation, is likely to provide new perspectives towards a better understanding of the role of the gut bacterial community on sand fly-pathogen interactions. This knowledge is crucial in order to implement control strategies for sand fly zoonotic visceral leishmaniasis.
|
10.1371/journal.pgen.1007901 | Functional divergence of a global regulatory complex governing fungal filamentation | Morphogenetic transitions are prevalent in the fungal kingdom. For a leading human fungal pathogen, Candida albicans, the capacity to transition between yeast and filaments is key for virulence. For the model yeast Saccharomyces cerevisiae, filamentation enables nutrient acquisition. A recent functional genomic screen in S. cerevisiae identified Mfg1 as a regulator of morphogenesis that acts in complex with Flo8 and Mss11 to mediate transcriptional responses crucial for filamentation. In C. albicans, Mfg1 also interacts physically with Flo8 and Mss11 and is critical for filamentation in response to diverse cues, but the mechanisms through which it regulates morphogenesis remained elusive. Here, we explored the consequences of perturbation of Mfg1, Flo8, and Mss11 on C. albicans morphogenesis, and identified functional divergence of complex members. We observed that C. albicans Mss11 was dispensable for filamentation, and that overexpression of FLO8 caused constitutive filamentation even in the absence of Mfg1. Harnessing transcriptional profiling and chromatin immunoprecipitation coupled to microarray analysis, we identified divergence between transcriptional targets of Flo8 and Mfg1 in C. albicans. We also established that Flo8 and Mfg1 cooperatively bind to promoters of key regulators of filamentation, including TEC1, for which overexpression was sufficient to restore filamentation in the absence of Flo8 or Mfg1. To further explore the circuitry through which Mfg1 regulates morphogenesis, we employed a novel strategy to select for mutations that restore filamentation in the absence of Mfg1. Whole genome sequencing of filamentation-competent mutants revealed chromosome 6 amplification as a conserved adaptive mechanism. A key determinant of the chromosome 6 amplification is FLO8, as deletion of one allele blocked morphogenesis, and chromosome 6 was not amplified in evolved lineages for which FLO8 was re-located to a different chromosome. Thus, this work highlights rewiring of key morphogenetic regulators over evolutionary time and aneuploidy as an adaptive mechanism driving fungal morphogenesis.
| Fungal infections pose a severe burden to human health worldwide. Candida albicans is a leading cause of systemic fungal infections, with mortality rates approaching 40%. One of the key virulence traits of this fungus is its ability to transition between yeast and filamentous forms in response to diverse host-relevant cues. The model yeast Saccharomyces cerevisiae is also capable of filamentous growth in certain conditions, and previous work has identified a key transcriptional complex required for filamentation in both species. However, here we discover that the circuitry governed by this complex in C. albicans is largely distinct from that in the non-pathogenic S. cerevisiae. We also employ a novel selection strategy to perform experimental evolution, identifying chromosome triplication as a mechanism to restore filamentation in a non-filamentous mutant. This work reveals unique circuitry governing a key virulence trait in a leading fungal pathogen, identifying potential therapeutic targets to combat these life-threatening infections.
| The fungal kingdom is recognized for its vast morphological plasticity, with many species capable of undergoing morphogenetic transformations in response to diverse environmental cues. For instance, filamentous fungi such as Aspergillus fumigatus undergo spore germination and branching throughout their life cycle, dermatophyte fungal pathogens produce arthroconidia, and thermally dimorphic fungi such as Histoplasma capsulatum exist as filamentous mycelia at ambient temperatures and transition to a yeast form upon exposure to mammalian physiological temperatures [1–3]. The purposes of such transformations are equally diverse, with morphogenesis playing critical roles in sexual reproduction, nutrient acquisition and virulence. For example, the basidiomycete Cryptococcus neoformans forms elongated filaments for the purpose of mating or monokaryotic fruiting [4], and the model yeast Saccharomyces cerevisiae undergoes invasive or pseudohyphal growth for nutrient acquisition under starvation conditions [5]. Finally, the fungal pathogen Candida albicans transitions from a yeast to hyphal state in response to a variety of host-relevant cues [2,6], which aids in tissue invasion, immune cell evasion, and biofilm formation, such that morphogenesis is critical for virulence of this pathogen [2,7,8]. Thus, fungal morphogenesis comprises a diversity of processes required for cellular survival, proliferation, and pathogenesis.
The past several decades have witnessed a surge in the frequency of life-threatening human fungal infections, making mycotic disease a serious public health problem. Candida species comprise one of the leading genera of human fungal pathogens, with C. albicans being the most prevalent [9]. Those most susceptible to these infections are the increasing population of immunocompromised patients, including those undergoing chemotherapy or transplantation, as well as those infected with HIV. C. albicans can cause fatal bloodstream infections, with mortality rates approaching 40% despite therapeutic intervention [9,10]. This is in part due to a limited number of antifungals available to treat systemic infections coupled with the frequent emergence of antifungal drug resistance in the clinic [11]. Further, the capacity of C. albicans to cause life-threatening disease in its human host is enabled by a complex repertoire of virulence factors, including the expression of surface structures that mediate adherence to epithelial cells, the secretion of hydrolytic enzymes that induce host cell damage, the capacity to produce biofilms that are intrinsically resistant to antifungal drugs, and the ability to transition between yeast and filamentous growth states [2,12–14]. Given the dearth of new antifungal classes uncovered by traditional approaches of targeting essential proteins required for viability, a complementary approach of targeting proteins required for pathogen virulence may prove to be a useful strategy to combat fungal infections [15].
The ability of C. albicans to transition between yeast and filamentous forms occurs upon exposure to a variety of different cues, including host febrile temperature, serum, and nutrient depletion [2,6]. Complex genetic circuitry underpins this transition, and distinct mechanisms are involved in initiating and maintaining filamentous growth [8,16–19]. One of the core pathways required for filamentation in C. albicans is the cAMP-protein kinase A (PKA) signaling cascade, for which multiple components of the pathway are required for filamentous growth in response to diverse conditions [2,6]. In C. albicans, the PKA complex is composed of the regulatory subunit Bcy1 and two catalytic subunits, Tpk1 and Tpk2 [20]. Increased cAMP levels produced by the adenylyl cyclase Cyr1 activate the PKA complex, resulting in phosphorylation and activation of the transcription factor Efg1, which governs the expression of many hyphal-specific genes [6,21,22]. Other signaling pathways including the Cek1-mitogen activated protein kinase (MAPK) pathway [23], Rim101 pH sensing pathway [24], and protein kinase C (Pkc1) cascade [25] also play important roles in enabling filamentation in response to diverse stimuli. Despite the continued identification and characterization of genes involved in C. albicans morphogenesis, an understanding of the regulatory networks governing morphogenesis remains largely elusive.
Systematic analyses of genes enabling filamentous growth have been performed on a genomic level in both S. cerevisiae and C. albicans [18,26]. Such analyses have revealed a striking divergence in the sets of genes required for filamentation between these species [18,26]. A recent study with the S. cerevisiae Σ1278b strain defined genes required for: a) diploid pseudohyphal formation in response to low nitrogen; b) haploid invasive growth in response to glucose depletion; and c) biofilm formation on semi-solid agar [26]. Despite largely distinct gene sets being important for each of these filamentation programs, there was a core gene set required for morphogenesis, including the previously uncharacterized protein Mfg1 [26]. In S. cerevisiae, Mfg1 forms a complex with the transcriptional regulators Flo8 and Mss11 to control the expression of hundreds of genes, including FLO11, which encodes a cell surface glycoprotein essential for filamentation [26,27]. Members of the S. cerevisiae Flo8-Mfg1-Mss11 complex primarily function in concert to promote morphogenesis. Of the 152 promoters bound by S. cerevisiae Mfg1, 89% are also bound by Flo8 and 78% are bound by Mss11 [26]. Further, deletion of S. cerevisiae MFG1 reduces Flo8 binding to 90.8% of its target promoters [26]. Mfg1 is also a key regulator of morphogenesis in C. albicans, and the physical interaction with Flo8 and Mss11 is conserved under basal conditions [26]. However, the mechanisms through which this complex regulates filamentous growth in C. albicans remains largely enigmatic given that the transcriptional targets of this complex remain unknown and that no FLO11 homolog has been identified in this fungal pathogen.
In this study, we characterized the role of the Flo8-Mfg1-Mss11 complex in regulating C. albicans filamentation. Although Flo8 and Mfg1 were required for filamentous growth in response to numerous conditions, they were not required for morphogenesis induced by compromised function of the molecular chaperone Hsp90, demonstrating that these mutants are capable of filamentation in response to specific cues. Further, overexpression of FLO8 resulted in constitutive hyphal growth yet overexpression of MFG1 did not, suggesting distinct functions of these regulators in the yeast-to-hyphal transition. We observed functional divergence of the complex members, as Mss11 was dispensable for filamentous growth and had a reduced physical interaction with Mfg1 and Flo8 in filament-inducing cues. Surprisingly, MSS11 overexpression was sufficient to induce morphogenesis in the absence of an inducing cue, highlighting the complex functional relationships. Chromatin immunoprecipitation (ChIP) of TAP-tagged Flo8 or Mfg1 followed by microarray analysis (ChIP-chip) revealed significant divergence between the transcriptional targets of Flo8 and Mfg1 in C. albicans, and highlighted dynamic temporal changes in promoter occupancy and gene expression in response to serum as a filament-inducing cue. However, Mfg1 and Flo8 cooperatively bound to a subset of key targets involved in morphogenesis, including TEC1, for which overexpression is sufficient to drive filamentation in the absence of either Mfg1 or Flo8. Finally, to gain further mechanistic insight into the circuitry through which Mfg1 regulates morphogenesis, we employed a novel experimental evolution strategy based on introduction of a drug resistance marker under the control of a filament-specific promoter to select for mfg1Δ/mfg1Δ mutants with a restored ability to filament in response to serum. Whole genome sequencing of mutants capable of filamentation in the absence of Mfg1 uncovered an adaptive mechanism involving amplification of chromosome 6, on which FLO8 is located. Thus, we highlight Mfg1 as a critical regulator of C. albicans morphogenesis, uncover distinct roles of Flo8, Mfg1, and Mss11 in promoting filamentation, and provide a striking example of aneuploidy formation as a mechanism of adaptive evolution to enable filamentous growth, offering broad insights into the biology, pathogenicity, and evolutionary strategies of a leading fungal pathogen.
Flo8 and Mfg1 are critical regulators of filamentation in C. albicans, and they physically interact with each other and with the transcriptional regulator Mss11 under basal conditions [26]. However, the mechanisms by which this complex regulates morphogenesis in C. albicans remain largely enigmatic. To characterize the role of the Flo8-Mfg1-Mss11 complex in the yeast-to-hyphal transition, we examined the morphology of C. albicans flo8Δ/flo8Δ, mfg1Δ/mfg1Δ, and mss11Δ/mss11Δ mutants in response to a variety of filament-inducing cues. Similar to previous reports [28], we observed that flo8Δ/flo8Δ mutants were completely blocked in filamentous growth in response to 10% serum, RPMI, Spider medium, or high temperature (Fig 1A). However, this strain was capable of forming robust filaments in response to Hsp90 inhibition with geldanamycin (Fig 1A). Similar to the flo8Δ/flo8Δ mutant, the mfg1Δ/mfg1Δ mutant was largely blocked in filamentation in the majority of cues examined, but was capable of filamenting in response to Hsp90 inhibition (Fig 1A). Notably, restoration of one allele of MFG1 [26] or FLO8 (S1 Fig) restored the capacity of the respective nulls to filament. In contrast to previous reports [29], no observable defect in filamentous growth was observed in two independently generated MSS11 mutants (Fig 1A and S2 Fig). As Flo8 and Mfg1 play central roles in filamentation induced by diverse cues, we also assessed their impact on filamentation induced by deletion of the repressors of filamentation, NRG1 [30] and LRG1 [25]. Strikingly, FLO8 was required for filamentation induced by loss of either NRG1 or LRG1, while MFG1 was largely dispensable (Fig 1B). Collectively, these results highlight that although Flo8 and Mfg1 are important for filamentation, mutants lacking these regulators are capable of polarized growth under specific environmental conditions and complex members have distinct roles in C. albicans filamentation.
Given that complex members are important to enable C. albicans filamentation in response to many cues, we next assessed if overexpression of any single component was sufficient to drive the filamentous growth program. Notably, overexpression of S. cerevisiae FLO8 or MSS11 has previously been shown to result in hyperfilamentous growth [31,32]. We overexpressed each of the three transcriptional regulators individually in C. albicans by replacing the native promoter of one allele with a tetracycline-repressible promoter, tetO, which drives strong and constitutive expression of the target gene in the absence of tetracycline (Fig 2A and 2B). Overexpression of FLO8 was sufficient to induce robust filamentous growth in the absence of an inducing cue, and resulted in a ~1,200-fold induction of the filament-specific transcript HWP1 relative to the wild-type strain (Fig 2A and 2C). This is reminiscent of other reports in which overexpression of C. albicans FLO8 resulted in wrinkly colony formation in 5% CO2 [33]. Consistent with previous reports [29], overexpression of C. albicans MSS11 was sufficient to induce some filamentation in the absence of an inducing cue and resulted in a 350-fold increase in expression of HWP1 relative to the wild-type strain (Fig 2A and 2C), highlighting that although this regulator is not necessary for filamentation, it does play a role in regulating this morphogenetic trait. Notably, the filaments induced by overexpression of FLO8 appeared distinct from those resulting from MSS11 overexpression, where the former resembled true hyphae, and the latter included many yeast-form cells, suggesting distinct mechanisms. Intriguingly, overexpression of C. albicans MFG1 did not result in filamentous growth in the absence of a filament-inducing cue, nor did it result in an induction of HWP1 (Fig 2A and 2C). This further supports our model that Flo8, Mfg1, and Mss11 have distinct roles in governing the yeast-to-hyphal transition.
Next, we examined if filamentation induced by overexpression of FLO8 or MSS11 was contingent upon Mfg1, given that MFG1 is required for filamentation in response to several inducing cues (Fig 1A). We replaced the native promoter of one allele of FLO8 or MSS11 with the tetO promoter, as above, in an mfg1Δ/mfg1Δ background (S3 Fig). Overexpression of FLO8 resulted in robust filamentous growth even in the absence of MFG1, and HWP1 transcript levels were induced by ~450-fold compared to the wild type (Fig 2C and 2D); this is less than half of the increase in HWP1 that was observed in the presence of MFG1, suggesting that Mfg1 promotes filamentation induced by Flo8. In contrast, overexpression of MSS11 was unable to drive filamentation in the absence of Mfg1, and did not result in induction of HWP1 (Fig 2C and 2D). Taken together, although Flo8, Mfg1, and Mss11 can exist as a complex in C. albicans under basal conditions, they also have distinct functions that influence morphogenesis.
Given our findings that C. albicans Flo8 and Mfg1 are key for filamentation in response to most cues (Fig 1A), we assessed if they displayed a conserved function with those complex members in S. cerevisiae. C. albicans FLO8 has been previously demonstrated to complement the filamentous growth defect of S. cerevisiae haploid flo8Δ and diploid flo8Δ/flo8Δ mutants [28]. Here, we cloned C. albicans MFG1 or FLO8 as a control, into an expression vector and expressed the constructs in the filamentation-competent S. cerevisiae background Σ1278b. Diploid cells were plated on nitrogen-limiting SLAD medium to monitor pseudohyphal growth. When cells expressed the empty vector, the wild-type diploid formed pseudohyphae and the flo8Δ/flo8Δ and mfg1Δ/mfg1Δ mutants were blocked in pseudohyphal growth (Fig 3A). As previously described, pseudohyphal growth was restored in the flo8Δ/flo8Δ mutant when C. albicans FLO8 was expressed (Fig 3A), demonstrating that it functionally complements the S. cerevisiae ortholog. We also observed that C. albicans MFG1 was able to functionally complement a S. cerevisiae strain lacking MFG1, resulting in pseudohyphal growth (Fig 3A). Similarly, expression of C. albicans FLO8 in the corresponding S. cerevisiae haploid deletion mutant was sufficient to restore haploid invasive growth, while expression of C. albicans MFG1 was able to partially restore invasive growth in the mfg1 haploid mutant (Fig 3B). Thus, C. albicans Mfg1 is able to at least partially functionally complement its S. cerevisiae ortholog.
In both S. cerevisiae and C. albicans, Flo8 functions downstream of PKA to regulate morphogenesis [28,34]. To determine if Mfg1 also acts downstream of the PKA complex in C. albicans, we overexpressed one of the PKA catalytic subunits, TPK2, by replacing the native promoter of one allele of TPK2 with the tetracycline-repressible promoter, tetO, to drive strong constitutive expression in the absence of tetracycline in the wild-type, flo8Δ/flo8Δ and mfg1Δ/mfg1Δ strains (Fig 4A). Cells were grown in rich medium at 30°C or 34°C, conditions that did not induce filamentous growth in wild-type cells, but caused substantial filamentation when TPK2 was overexpressed (Fig 4B). As expected, cells remained in the yeast form in the absence of FLO8 despite overexpression of TPK2 at 30°C or 34°C (Fig 4A and 4B), confirming that Flo8 acts downstream of PKA. Similarly, overexpression of TPK2 was insufficient to restore filamentation in the absence of MFG1 at either temperature (Fig 4A and 4B), suggesting that Mfg1 also acts downstream of PKA to enable C. albicans morphogenesis. Interestingly, TPK2 overexpression in the mfg1Δ/mfg1Δ background resulted in an enlarged cellular morphology. This demonstrates another phenotypic difference between flo8Δ/flo8Δ and mfg1Δ/mfg1Δ null mutants.
Given that perturbation of C. albicans Flo8 and Mfg1 had distinct consequences on filamentous growth (Fig 1 and Fig 2), we next pursued a global analysis of genes bound and regulated by complex members to further probe their contributions to morphogenesis. Notably, both Flo8 and Mfg1 are localized to the nucleus in basal conditions and upon treatment with serum (S4A Fig). We performed chromatin immunoprecipitation coupled with microarray analysis (ChIP-chip) with strains harboring Flo8 or Mfg1 tagged with a tandem affinity purification (TAP) epitope. Functionality of the TAP-tagged proteins was verified (S4B Fig). DNA binding of Flo8 or Mfg1 was measured in either untreated conditions, or upon exposure to a filament-inducing cue, 10% serum, for one hour or three hours, to identify promoter regions that were bound by these transcriptional regulators (S1 Table). This demonstrated a substantial increase in promoter occupancy of both Flo8 and Mfg1 upon exposure to serum, consistent with their roles as transcriptional regulators of morphogenesis (Fig 5A, S5A Fig and S5B Fig). To identify the Flo8 and Mfg1 targets that are also transcriptionally modulated by these regulators, we performed microarray analysis comparing the gene expression profiles of flo8Δ/flo8Δ, mfg1Δ/mfg1Δ, and flo8Δ/flo8Δ mfg1Δ/mfg1Δ mutants with a wild-type strain under basal and filament-inducing conditions (S2 Table). We noted stark differences in transcriptional regulation by Flo8 and Mfg1 in response to serum, where large sets of genes were specifically altered in expression in either the flo8Δ /flo8Δ or mfg1Δ/mfg1Δ mutant, or in the double mutant (S5C Fig and S5D Fig). There was a considerable temporal component to the effects on gene expression, where we observed significant differences between the one-hour and three-hour serum-exposed conditions (S5C Fig and S5D Fig).
In order to identify key downstream targets of Flo8 and Mfg1 through which they regulate C. albicans filamentation, we compared those genes both bound and transcriptionally regulated by Flo8 and Mfg1 (Fig 5A, S3 Table). In basal conditions, we identified only 17 and 11 genes whose promoters were bound by either Flo8 or Mfg1 respectively, and whose expression was altered (Fig 5A). These sets were expanded in the presence of serum, where we identified 77 genes whose promoters were bound by Flo8 and 101 bound by Mfg1 in filament-inducing conditions, and for which expression was altered in the respective null mutants at the respective time point (Fig 5A). Surprisingly, unlike S. cerevisiae Flo8 and Mfg1 which bind to overlapping targets [26], C. albicans Flo8 and Mfg1 bind and regulate largely distinct sets of targets (Fig 5A, S5A Fig and S5B Fig, S1 Table), which is striking since Mfg1 does not have a characterized DNA-binding motif. This highlights the unique functions of Flo8 and Mfg1 in regulating gene expression and identifies dynamic transcriptional changes that occur in response to serum.
To further explore the divergence in Flo8 and Mfg1 targets and to identify any potential additional Mfg1-binding partners, we performed affinity purification coupled to mass spectrometry (AP-MS) with C-terminally GFP-tagged Mfg1. Functionality of the tagged protein was verified (S4C Fig). Our previous AP-MS experiments demonstrated that C. albicans Mfg1 physically interacts with both Flo8 and Mss11 under standard conditions [26]. Here, we expanded the analysis to identify Mfg1 interaction partners in filament-inducing conditions (1 hour or 3 hours of growth in 10% serum at 37°C), and included sonication and benzonase-treatment to aid in the identification of chromatin-bound proteins. In untreated conditions, Mfg1 interacts strongly with Flo8 and Mss11, as expected (Fig 5B, S4 Table). Under filament-inducing conditions, Flo8 continues to interact with Mfg1, suggesting a key role for their interaction. However, interaction with Mss11 is decreased in filament-inducing conditions, consistent with it being dispensable for filamentation (Fig 1A, Fig 5B, S4 Table). Interestingly, while the AP-MS identified other proteins as Mfg1 interactors, there were no additional DNA-binding proteins that could mediate Mfg1 binding to DNA, even in response to serum (Fig 5B, S4 Table).
Many of the genes bound and transcriptionally regulated by Flo8 or Mfg1 upon exposure to serum have been implicated in morphogenesis (Fig 5A), suggesting that they could be key downstream targets through which Flo8 and Mfg1 regulate C. albicans filamentation. We leveraged this data to probe circuitry downstream of the relatively uncharacterized regulator, Mfg1. We initially focused on a subset of genes that were direct targets of Mfg1 and for which expression was altered in the corresponding condition in the mfg1Δ/mfg1Δ mutant, including the positive regulators of filamentation TEC1, ROB1, HSP21, IHD1 and SUR7, as well as the negative regulator CUP9. We determined if the effectors bound and regulated by Mfg1 were sufficient to promote filamentous growth in the absence of MFG1 by overexpressing each positive regulator in an mfg1Δ/mfg1Δ background using the strong tetO promoter, or by deleting negative regulators of filamentation (Fig 6A and S6 Fig). Overexpression of ROB1, HSP21, IHD1, or SUR7, or deletion of CUP9, did not restore the ability of the mfg1Δ/mfg1Δ mutant to filament (S6A Fig and S6B Fig), suggesting that none of these effectors alone are sufficient to modulate Mfg1-mediated filamentation. As previously reported, overexpression of the transcription factor encoded by TEC1 in an otherwise wild-type background resulted in robust filamentation, even in the absence of an inducing cue [35] (Fig 6A and S6C Fig). Strikingly, overexpression of TEC1 was sufficient to induce robust filamentation in the absence of Mfg1 (Fig 6A and S6C Fig), suggesting that TEC1 may be a key regulator of filamentation downstream of Mfg1. Interestingly, TEC1 overexpression also resulted in robust filamentation in the absence of Flo8 (Fig 6A and S6C Fig), suggesting that it acts downstream of both regulators.
To determine if Flo8 and Mfg1 cooperatively regulate TEC1 expression, we then examined binding of Flo8 and Mfg1 to the promoter of TEC1 in the absence of the other regulator. By ChIP-qPCR, we observed increased binding compared to the untagged control of both Flo8 and Mfg1 to the TEC1 promoter, especially in the presence of serum at 37°C for 1 hour, which was significantly reduced in the absence of the other regulator (Fig 6B). This suggests a functional relationship between Flo8 and Mfg1 at the TEC1 promoter. To further explore this relationship, we measured transcript levels of FLO8 and MFG1, and protein levels of Flo8-TAP and Mfg1-TAP, in the absence of the other regulator. While deletion of Flo8 or Mfg1 had no effect on transcript level of the other regulator (S7A Fig), deletion of FLO8 resulted in an approximately 4-fold decrease in Mfg1-TAP levels (S7B Fig). This decrease in Mfg1-TAP protein levels may contribute to the decreased binding observed at the TEC1 promoter in the absence of flo8Δ/flo8Δ (Fig 6B). Nevertheless, Flo8-TAP binding was reduced in the absence of Mfg1 despite the protein levels remaining stable (Fig 6B and S7B Fig), demonstrating a dependency of Flo8 on Mfg1 for binding to target promoters. We further confirmed this observation with another key regulator of filamentation, BRG1, where we observed that deletion of either MFG1 or FLO8 significantly reduced binding of the other regulator at the BRG1 promoter (S8A Fig). In addition, we observed that TEC1 and BRG1 expression were decreased in the absence of either Flo8 or Mfg1 (Fig 6C and S8B Fig), illustrating that these regulators both bind to the promoters and regulate expression of a subset of key regulators of filamentation.
Considering the key role for Tec1 in regulating filamentation, we further explored its relationship to Flo8 and Mfg1 by examining TEC1 expression in the tetO-FLO8/FLO8 overexpression strain. Interestingly, overexpression of FLO8 drives an approximately 4-fold increase in TEC1 expression relative to the wild type, which is decreased 2-fold in the absence of Mfg1 (Fig 6D). This is consistent with the fact that overexpression of FLO8 drives filamentation in the absence of an inducing cue, and to a lesser extent in the absence of MFG1 (Fig 2C and 2D). We further confirmed this observation by examining BRG1 expression in the tetO-FLO8/FLO8 strain, and similarly observed an increase in BRG1 expression upon overexpression of FLO8, which is reduced in the absence of Mfg1 (S8C Fig). Together, this suggests that Flo8 and Mfg1 cooperatively bind to the promoters of key regulators of filamentation, and that overexpression of FLO8 is sufficient to drive expression of these regulators and induce morphogenesis even in the absence of Mfg1.
To identify additional circuitry downstream of Mfg1 important for C. albicans filamentation, we turned to an alternative, unbiased approach and employed a novel selection strategy to evolve mutants capable of filamenting in a strain lacking MFG1. Our approach employed a dominant nourseothricin (NAT) resistance marker under the control of a filament-specific promoter (HWP1p), such that the expression of NAT occurs only when cells are undergoing filamentous growth, thereby enabling selection for filamentation on plates containing NAT. We introduced this system into an mfg1Δ/mfg1Δ mutant, and plated cells on filament-inducing conditions of rich medium containing 10% serum with a high concentration of NAT, with incubation at 37°C for two days. We evolved three independent lineages in the mfg1Δ/mfg1Δ background with a restored capacity to undergo robust filamentation in liquid medium containing serum as observed after six hours of growth (Fig 7A). Notably, this filamentation phenotype was largely lost after growth in serum for 24 hours, whereas wild-type cells were still filamentous, suggesting that filamentation could not be maintained in the absence of MFG1 (S9A Fig). This demonstrates that our HWP1p-NAT selection strategy provides adequate selective pressure to evolve the capacity for morphogenesis in filamentation-defective mutants.
To determine the genetic basis for this restoration of filamentation in the absence of MFG1, we performed whole genome sequencing of the three independently evolved mutants. Although no single nucleotide variants or small insertions or deletions in coding regions were identified that were common to all of the evolved lineages, we did observe a common decrease in copy number of chromosome 4 and an increase in copy number of chromosome 6 relative to the parent (Fig 7B). Upon closer examination, the apparent loss of chromosome 4 was due to a chromosome 4 trisomy in the parental strain, resulting in all evolved lineages having two copies of this chromosome. We therefore focused our analysis on the increased copy number of chromosome 6. While many genes involved in filamentation are located on chromosome 6, a likely candidate responsible for restoring filamentous growth was FLO8, as we previously observed that overexpression of FLO8 was sufficient to promote morphogenesis in the absence of MFG1 (Fig 2D). Increased expression of FLO8 in the evolved strains was verified by qRT-PCR (S9B Fig). To determine if the increase in FLO8 copy number was responsible for restoring filamentation of the mfg1Δ/mfg1Δ mutant, we deleted one allele of FLO8 from one of the evolved filamentous mutants and observed a corresponding decrease in ability of this strain to filament in response to serum (Fig 7C). This suggests that the chromosome 6 amplification may have enabled filamentation in response to serum by increasing the copy number of FLO8. Furthermore, we examined TEC1 and BRG1 expression in this evolved strain, as we had observed that FLO8 overexpression results in increased expression of these important regulators of filamentation, even in the absence of Mfg1 (Fig 6D and S8C Fig). Indeed, when grown in filament-inducing conditions, we observe a small but significant increase in both TEC1 and BRG1 transcript levels in an mfg1Δ/mfg1Δ evolved strain with a chromosome 6 amplification (S9C Fig). Notably, we also attempted to evolve filament-capable mutants in the flo8Δ/flo8Δ mutant and the mfg1Δ/mfg1Δ flo8Δ/flo8Δ mss11Δ/mss11Δ triple deletion mutant, yet were unable to obtain NAT-resistant colonies when plating cells on medium containing serum and a high concentration of NAT (S9D Fig), confirming that Flo8 is critical for morphogenesis in this experimental context. Together this suggests that Flo8 plays a pivotal role in enabling filamentation even in the absence of MFG1, and may do so by driving expression of key regulators of filamentation, including TEC1 and BRG1.
To further explore if FLO8 amplification underpins this adaptive mechanism to restore filamentation by chromosome 6 amplification, we engineered a strain in which both alleles of FLO8 were deleted from the endogenous location on chromosome 6 and integrated at a distinct location in the genome, on the right arm of chromosome 5. To confirm that FLO8 in a new chromosomal context is functional, we verified expression of two genes whose expression increase in the flo8Δ/flo8Δ null and mfg1Δ/mfg1Δ flo8Δ/flo8Δ double mutant (S2 Table), but are restored to wild-type levels when FLO8 is integrated on chromosome 5 (S10A Fig). Using our established selection regime, we successfully evolved five lineages in the mfg1Δ/mfg1Δ background with FLO8 located on chromosome 5 that were capable of undergoing filamentation upon growth in liquid medium containing serum (Fig 8A), with many but not all reverting to yeast form growth after 24 hours (S10B Fig). Whole genome sequencing revealed that none of the newly evolved lineages possessed an aneuploidy of chromosome 6 (Fig 8B), providing evidence that the presence of FLO8 on chromosome 6 was an important determinant of aneuploidy formation in the original lineages (Fig 7). Consistent with the importance of FLO8 in aneuploidy formation, one of the newly evolved lineages acquired an amplification of the right arm of chromosome 5 (Evo #C1-1), on which FLO8 was located in this background (Fig 8B). A significant increase in FLO8 expression in this evolved lineage was verified by qRT-PCR (S10C Fig). Intriguingly, the other newly evolved lineages acquired distinct chromosomal alterations that were independent of FLO8 copy number, including amplification of the right portion of chromosome 1 (Evo #C1-2), amplification of the left portion of chromosome 7 (Evo #C1-1 and Evo #C1-2), and loss of the right portion of chromosome 7 (Evo #C1-2) (Fig 8B). Interestingly, NRG1 is located on the right arm of chromosome 7, and deletion of NRG1 allows filamentous growth in an mfg1Δ/mfg1Δ mutant (Fig 1B), suggesting that reduced NRG1 dosage might be an adaptive mechanism to promote filamentation in this evolved strain. In addition to chromosomal alterations, a few SNPs were identified in the evolved lineages, including in the predicted kinase gene YAK1 and upstream of the putative transcription factor gene HOT1 (S5 Table). Together, these results suggest that amplification of FLO8 is a major driving force of aneuploidy formation in order to restore filamentation in the absence of MFG1, and emphasizes that additional factors remain to be discovered.
Fungal species rely on morphogenetic transitions for diverse biological processes, including reproduction, stress adaptation and virulence [2,3]. The transcriptional regulators Flo8, Mfg1 and Mss11 work as a complex in the model yeast S. cerevisiae to control the expression of genes that enable filamentous growth [26]. Although these regulators are conserved in C. albicans, as is their physical interaction under basal conditions, we established here that Flo8, Mfg1 and Mss11 possess divergent roles in regulating C. albicans filamentation (Fig 1 and Fig 2). Genome scale analyses supported the model that although Flo8 and Mfg1 are capable of acting in concert to regulate gene expression, the majority of their transcriptional targets are distinct and environmentally contingent (Fig 5 and S5 Fig). Moreover, we described a novel approach for mapping genetic circuitry important for C. albicans filamentation, and discovered that increasing FLO8 copy number can drive aneuploidy formation to restore morphogenesis in the absence of MFG1 (Fig 7). Thus, we reveal complex genetic circuitry through which Flo8, Mfg1 and Mss11 regulate filamentation, and implicate a new facet of genomic plasticity in pathogen adaptation.
A global analysis exploring regulators of filamentation in S. cerevisiae and C. albicans identified the previously uncharacterized Mfg1 as a core regulator of morphogenesis in both species, and showed that it physically interacts with transcriptional regulators Flo8 and Mss11 [26]. In S. cerevisiae, these three proteins act in concert to enable pseudohyphal growth, biofilm formation and invasive growth through transcriptional regulation of target genes [26]. Genome-scale analyses demonstrated that deletion of any component of this complex largely abrogated binding of other complex members to target genes [26], highlighting the functional interdependence of these transcriptional regulators. In contrast, we identified considerable divergence in complex member function in C. albicans. Although both Flo8 and Mfg1 were required for filamentous growth, Mss11 was dispensable for hyphal formation in response to any cue that we tested, and its physical interaction with Mfg1 was reduced in filament-inducing conditions (Fig 1 and Fig 5B). Although Flo8 and Mfg1 target largely different genes, they do cooperatively bind to a subset of key regulators of filamentation, including TEC1 and BRG1. Mfg1 and Flo8 binding to these promoters is reduced in the absence of the other regulator (Fig 6B, S8A and S10D Figs), leading to reduced expression of targets such as TEC1 and BRG1 (Fig 6C and S8B Fig). However, overexpression of FLO8 was sufficient to result in increased expression of TEC1 and BRG1, resulting in filamentous growth even in the absence of Mfg1 (Fig 2, Fig 6D, S8C Fig, and S10D Fig), and suggesting that Mfg1 is dispensable for Flo8 binding to at least a subset of key targets or that overexpression of FLO8 can compensate for the loss of Mfg1. Although Flo8 was required for filamentation in response to loss of the filamentation repressors Nrg1 and Lrg1, Mfg1 was dispensable, highlighting another distinct facet of their consequences on morphogenesis. Interestingly, mfg1Δ/mfg1Δ mutants resembled mutants lacking Eed1, a regulator of hyphal extension, which form germ tubes after a short exposure to filament-inducing cues, but revert to yeast after prolonged exposure to the inducing cue [36]. This suggests that Mfg1 may be a key regulator of hyphal extension as opposed to initiation. Together, these findings support evolutionary rewiring of complex member function compared with that observed in S. cerevisiae.
Experimental evolution has been used extensively to investigate circuitry governing microbial drug resistance, virulence and filamentation. Such studies have revealed that aneuploidies are a prevalent adaptive mechanism in fungi, involving ploidy gain or reduction, upon exposure to stressful environmental conditions [37]. The acquisition of aneuploidies in C. albicans upon exposure to azoles, the most widely deployed class of antifungal, occurs readily in both in vitro experimental evolution experiments, as well as in clinical settings [38–40]. Most common is the formation of an isochromosome on the left arm of chromosome 5 (i(5L)), which increases the copy number of the azole drug target gene ERG11, as well as a gene encoding a transcriptional regulator of an efflux pump, TAC1 [41]. Experimental evolution has also identified adaptations that enable filamentation of non-filamentous Candida within mammalian macrophages, where filamentation is favoured as it enables host cell escape [42,43]. A single nucleotide polymorphism in the chitin synthase gene CHS2 was found to restore filamentation in Candida glabrata [43], and a mutation in SSN3, encoding a component of the Mediator complex, facilitated filamentation in a non-filamentous C. albicans strain [42]. We employed a novel selection method to restore morphogenesis in non-filamentous C. albicans mutants by placing a dominant NAT resistance cassette under the control of the filament-specific HWP1 promoter. A key advantage of this strategy is that it can be easily applied to diverse genetic backgrounds and environmental conditions. With this approach, we restored filamentous growth in an mfg1Δ/mfg1Δ mutant and identified an increase in chromosome 6 copy number as a mechanism to enable filamentation (Fig 7). Notably, recent findings identified a chromosome 6 trisomy in a parasexual isolate with increased virulence and capacity to filament [44]. Furthermore, preferential amplification of C. albicans chromosome 6 has been identified upon passage in a murine model of oropharyngeal candidiasis [45], suggesting that amplification of chromosome 6 is also important for adaptation in vivo. Analogous adaptive strategies have been reported in S. cerevisiae, where the gain of a single chromosome was sufficient to transition between smooth and fluffy colony morphologies [46]. Although genomic plasticity has been associated with altered morphologies in fungal pathogens [45,47–52], our work provides the first example of aneuploidy formation as a causative mechanism enabling filamentous growth in the context of C. albicans experimental evolution.
As the frequency and severity of fungal infections continue to climb, the necessity for novel therapeutic interventions is of utmost importance. Given the limited number of antifungal classes uncovered by traditional approaches of targeting essential proteins, targeting regulators of pathogen-specific virulence traits as a therapeutic strategy is attracting increasing interest [15]. For C. albicans, the transition between yeast and hyphae is of particular relevance, as both forms are required for virulence. Many small molecules with antifungal properties also have profound effects on fungal dimorphism, including the azole class of antifungals [53], compounds that target the molecular chaperone Hsp90 [54], the promiscuous protein kinase inhibitor staurosporine [55], the natural product beauvericin [56], and the broad-spectrum metal chelator DTPA [57]. Various other small molecules have also been described as able to modulate the transition between yeast and hyphae [58]. Although large-scale chemical biology screens typically monitor pathogen viability to reveal bioactive molecules, recently a screen to identify inhibitors of C. albicans adhesion uncovered filastatin, an inhibitor of hyphal formation [59]. Analogous approaches using expanded libraries of chemical matter have the potential to reveal additional small molecules capable of attenuating fungal virulence. Intriguingly, targeting specific aneuploidies with small molecules has also been described as a potential strategy, specifically to treat azole-resistant C. albicans harbouring (i(5L)) [60]. Thus, exploring circuitry governing C. albicans filamentation has the potential to reveal new strategies to cripple fungal pathogens and also to uncover fascinating biological insights into adaptive mechanisms governing key virulence traits.
Archives of C. albicans and S. cerevisiae strains were maintained at −80°C in rich medium (YPD) or in synthetic defined (SD) medium with 25% glycerol. YPD was prepared as follows: 1% yeast extract, 2% bactopeptone, 2% glucose, with 2% agar for solid medium. SD was prepared as follows: 6.7 g/L yeast nitrogen base, 2% glucose, supplemented with amino acids as necessary. Strains were propagated in YPD or SD medium as required. Spider medium was prepared as previously described [61]. RPMI was prepared as follows: 10.4 g/L RPMI-1640, 3.5% MOPS, 2% glucose, supplemented with an additional 5 mg/mL histidine as required, pH 7. Synthetic low ammonia dextrose (SLAD) solid medium plates supplemented with 15 μM leucine and 10 μM histidine were prepared as previously described [26,62]. Geldanamycin was obtained from LC laboratories, G-4500, and was dissolved in DMSO. Heat-inactivated newborn calf serum (Gibco) was used in YPD at 10%. All strains used in this study are listed in S5 Table and construction is described in S1 Text. To select for nourseothricin (NAT)-resistant mutants, NAT (Jena Bioscience) stock solution was prepared in water at a concentration of 250 mg/mL and YPD plates were supplemented with 150 μg/mL NAT. The SAP2 promoter was induced to drive expression of the FLP recombinase to excise the NAT marker cassette [63,64]. All plasmids used in this study are listed in S6 Table and all oligonucleotide sequences used in this study are included in S7 Table.
Each microarray experiment was conducted in triplicate. Cells were grown overnight in YPD at 30°C, diluted to OD600 of 0.2, and grown to mid-log phase, at which point 10% serum was added and cultures were moved to 37°C for 1 or 3 hours. Cultures were pelleted at 3000 rpm for 5 min and frozen overnight at –80°C. RNA was isolated using the QIAGEN RNeasy kit and RNasefree DNase (QIAGEN). Microarray experiments were performed essentially as described, using a high-density tiling array containing 240,798 unique 60-mer probes [65,66]. Briefly, 20 μg of RNA was reverse transcribed using Superscript III Reverse Transcriptase (Invitrogen) and oligo(dT)21 in the presence of Cy3- or Cy5-dCTP (Invitrogen). Template RNA was degraded using 2.5 units RNase H (USB) and 1 μg RNase A (Pharmacia), incubated at 37°C for 15 min. Labeled cDNA was purified with a QIAquick PCR Purification Kit (QIAGEN). The hybridization was carried out with DIG Easy Hyb Solution (Roche Diagnostics) containing 0.45% salmon sperm DNA and 0.45% yeast tRNA at 42°C for 24 hours in a SlideBooster Hyb chamber SB 800 (Advalytix, Brunnthal, Germany) with regular microagitation. The slides were washed once in 1.0% SSC (0.15 M NaCl and 0.015 M sodium citrate) with 0.2% SDS at 42°C for 5 min; twice in 0.1% SSC with 0.2% SDS at 42°C for 5 min; and once in 0.1% SSC at 24°C for 5 min, followed by four rinses in 0.1% SSC. The microarray slides were air dried before being scanned using a ScanArray Lite microarray scanner (Perkin Elmer). Microarray data were analyzed with GeneSpring GX v7.3 (Agilent Technologies), and genes with statistically significant (P<0.05) changes in transcript abundance of ≥1.5-fold were identified with a volcano plot and compared to other lists of significantly modulated genes. Data are accessioned at NCBI GEO GSE117477.
ChIP-chip was performed as previously described [65,66]. In brief, binding locations were determined in duplicate ChIP-chip experiments using a high density tiling array containing 240,798 unique 60-mer probes. Fluorescence intensities were quantified using ImaGene software 9.0 (BioDiscovery Inc.), background corrected, and normalized for signal intensity (Lowess normalization). The significance cut-off was determined using the distribution of log-ratios for each factor. It was set at two standard deviations from the mean of log transformed fold enrichments (cut off log ratio of 0.4). ORFs with a binding peak within the 1500 bp 5’ region were identified as targets. Values shown are an average of two biological replicates of tagged and mock constructs. Data are accessioned at NCBI GEO GSE117477.
To select for filamentous mutants, 4x107 cells from independent overnights were plated on YPD+10% serum +250 μg/mL NAT. NAT resistant colonies appeared after 2 days at 37°C. Genomic DNA was isolated with phenol chloroform, as described previously [67]. Libraries were prepared using the NexteraXT DNA Sample Preparation Kit following the manufacturer’s instructions (Illumina). Libraries were purified with AMPure XP beads (Agencourt) and library concentration was quantified using a Bioanalyzer High Sensitivity DNA Chip (Agilent Technologies) and a Qubit High Sensitivity dsDNA fluorometric quantification kit (Life Technologies). DNA Libraries were sequenced using paired end 2x250 flow cells on an Illumina MiSeq (Creighton University). Copy number variation was visualized using YMAP [68]. The sequence data is publicly available on the NCBI Sequence Read Archive with accession number SRP124459.
Flo8-TAP and Mfg1-TAP binding at the BRG1 and TEC1 promoters were analyzed by ChIP-qPCR, as described in reference [69], with the following modifications. Cells were grown in YPD to middle-late log phase (OD600 4–5), collected by centrifugation and reinoculated to OD600 of 1 in 50 mL pre-warmed 30°C YPD or 75 mL 37°C YPD with 10% new-born calf serum (Gibco) and grown for 1 hour before fixation with 1% formaldehyde for 20 minutes. Cells were lysed by bead beating for 5 x 1 minute, and chromatin was sheared by probe sonication (40% amplitude) for 4 x 16 seconds and in a Bio-disruptor water bath sonicator (high setting; 30 seconds on/ 30 seconds off) for 4 x 5 minute intervals. 300 μL of sonicated cell lysates were incubated with 20 μL rabbit IgG agarose beads (Sigma) at 4°C overnight. Beads were extensively washed with wash buffer, deoxycholate buffer and TE. Final ChIP products were eluted and reverse-crosslinked in TE-SDS buffer and purified by PCR purification column (Qiagen). An untagged parental strain (SN95) was tested in parallel as a control, where the tagged cells grown in untreated or hyphal-inducing conditions were normalized to the untagged parents grown in either untreated or hyphal-inducing conditions, respectively. Data analyses were performed similarly as described in reference [69]. Binding was assessed by qPCR using primers oLC7051/oLC7054 (BRG1) and oLC7371/oLC7372 (TEC1).
Immunoblotting to monitor Flo8-TAP and Mfg1-TAP protein levels was performed as described in reference [69]. Crude cell lysates were resolved on 6% SDS-PAGE and probed by a rabbit polyclonal TAP antibody (Invitrogen; CAB1001). A gel slice that does not contain immunoblotting signals was stained by Simple Blue Safe (Invitrogen) as a loading control.
Cells were grown overnight in YPD at 30°C, diluted to an OD600 of 0.1, and grown at 30°C to mid-log phase. To verify overexpression of various genes in the mfg1Δ/mfg1Δ background, cells were subcultured to an OD600 of 0.1 in YPD in the presence of 10% serum and grown at 37°C to mid-log phase. To examine expression of TPK2, cells were grown as above, but subcultured to OD600 of 0.1 and grown at 30°C or 34°C to mid-log phase. Cultures were pelleted and frozen at -80°C. RNA extraction, complementary DNA synthesis and PCR were performed as previously described [18]. Reactions were performed in triplicate, for two biological replicates and data were analyzed using the BioRad CFX Manager 3.1. Transcript levels were examined using the primers, oLC1988/oLC1989 (TEF1), oLC2285/oLC2286 (ACT1), oLC3796/oLC751 (HWP1), oLC5038/oLC5320 (MFG1), oLC5040/oLC5323 (MSS11), oLC5036/oLC5322 (FLO8), oLC4839/oLC4840 (SUR7), oLC4831/oLC4832 (HSP21), oLC2637/oLC2638 (ROB1), oLC5699/oLC829 (TPK2), oLC6472/oLC6473 (PMA1), oLC6476/oLC6477 (RIP1), oLC6738/oLC6739 (orf19.3897), oLC6736/oLC6737 (PGA26), oLC1456/oLC1457 (IHD1), oLC6718/oLC6719 (TEC1), and oLC2635/oLC2636 (BRG1). All oligonucleotide sequences are listed in S7 Table.
Pseudohyphal growth of diploid S. cerevisiae cells was assayed on SLAD solid medium supplemented with 15 μM leucine and 10 μM histidine as previously described [26,62]. Plates were incubated at 30°C for 11 days. Images of single colonies were taken on a Zeiss Axio Observer.Z1 (Carl Zeiss). Invasive growth of haploid S. cerevisiae was assayed by spotting equal cell dilutions on YPD plates with 2% agar. Plates were incubated at 30°C for 4 days before washing with water to remove non-invasive cells. Images were taken before and after washing with a Canon Power Shot A610. To image C. albicans, cells were grown overnight in YPD at 30°C. Cultures were diluted to an OD600 of 0.1 and grown in the indicated conditions. Cells were imaged using differential interference contrast (DIC) microscopy using a Zeiss Axio Imager.MI (Carl Zeiss). All images are representative of multiple fields of view from at least biological duplicate experiments.
For nuclear staining and microscopic analysis of cell morphology, cells were harvested, washed once with PBS and resuspended in 1 mL PBS containing 5 μg/mL Hoechst 33342. Following an incubation of 15 min in the dark at room temperature, cells were spun down and resuspended in 50 μL of PBS and imaged. Imaging was performed on a Zeiss Imager M1 upright microscope at 40X magnification on the green fluorescent protein (GFP) channel for GFP tagged proteins, the DAPI (4’,6-diamidino-2-phenylindole) channel for nuclei stained with Hoechst 33342 and the DIC channel. At least three fields were imaged for each strain, in at least two biological replicates.
|
10.1371/journal.pgen.1000145 | Antagonism between DNA and H3K27 Methylation at the Imprinted Rasgrf1 Locus | At the imprinted Rasgrf1 locus in mouse, a cis-acting sequence controls DNA methylation at a differentially methylated domain (DMD). While characterizing epigenetic marks over the DMD, we observed that DNA and H3K27 trimethylation are mutually exclusive, with DNA and H3K27 methylation limited to the paternal and maternal sequences, respectively. The mutual exclusion arises because one mark prevents placement of the other. We demonstrated this in five ways: using 5-azacytidine treatments and mutations at the endogenous locus that disrupt DNA methylation; using a transgenic model in which the maternal DMD inappropriately acquired DNA methylation; and by analyzing materials from cells and embryos lacking SUZ12 and YY1. SUZ12 is part of the PRC2 complex, which is needed for placing H3K27me3, and YY1 recruits PRC2 to sites of action. Results from each experimental system consistently demonstrated antagonism between H3K27me3 and DNA methylation. When DNA methylation was lost, H3K27me3 encroached into sites where it had not been before; inappropriate acquisition of DNA methylation excluded normal placement of H3K27me3, and loss of factors needed for H3K27 methylation enabled DNA methylation to appear where it had been excluded. These data reveal the previously unknown antagonism between H3K27 and DNA methylation and identify a means by which epigenetic states may change during disease and development.
| Methylation of DNA and histones exert profound and inherited effects on gene expression. These occur without changes to the underlying DNA sequence and are considered epigenetic effects. Disrupting epigenetic states can cause developmental abnormalities and cancer. Very little is known about how locations in the mammalian genome are chosen to receive these chemical modifications, or how their placement is regulated. We have identified a DNA sequence that acts as a methylation programmer at the Rasgrf1 locus in mice. It is required for methylation of nearby DNA sequences and can also influence the levels of local histone methylation. The methylation programmer has different effects on paternally and maternally derived chromosomes, directing DNA methylation on the paternal allele and histone H3 lysine 27 trimethylation on the maternal allele. These two methylation marks are not only mutually exclusive; they are also mutually antagonizing, whereby one blocks the placement of the other. Manipulations that cause aberrant changes in the levels of one of these marks had the opposite effect on the other mark. These observations identify novel mechanisms that specify epigenetic states in vivo and provide a framework for understanding how pathological epigenetic changes can arise, including those emerging at tumor suppressors during carcinogenesis.
| In mammals, imprinted loci are expressed from only one allele. Accompanying and controlling monoallelic expression are allele-specific epigenetic modifications influenced by an imprinting control region (ICR). Within this region, there is a differentially methylated domain (DMD) that is subject to acquisition of epigenetic modifications, typically DNA methylation and histone modifications. These modifications are placed in a parent-of-origin specific manner and impose an epigenetic state that dictates allele-specific gene expression at imprinted loci [1].
Previously, we characterized the mechanisms by which the ICR controls allele-specific methylation and expression at the imprinted Rasgrf1 locus. The ICR, located 30 kbp upstream of the transcriptional start site, is a binary switch consisting of a repeated element and the DMD. The repeated element functions as a methylation programmer, that is necessary for the establishment and maintenance of DNA methylation at the DMD on the paternal allele and sufficient for establishing gametic imprints in both germlines ([2],[3] and YJP, HH, AML, Ying Gao and PDS, in preparation). The DMD is a methylation sensitive enhancer blocker that binds CTCF on the unmethylated maternal allele and limits enhancer to promoter interactions, silencing the maternal allele [4]. DNA methylation that is directed to the paternal DMD by the repeats prevents CTCF binding, allowing expression of the paternal allele. The repeats constitute the first identified, and one of only a few known, naturally occurring DNA methylation programmers in mammals [5]–[8].
Epigenetic analysis of Rasgrf1 done by others examined DNA methylation across an expanded region centered on the ICR ([9] and Hisato Kobayashi and Hiroyuki Sasaki unpublished data) and histone modifications at the ICR [10]. The DNA methylation data suggested that a broader DMD exists in somatic tissue and in the male germline than was previously appreciated [9]. The histone methylation data indicated that several allele-specific histone modifications accompany the DNA methylation differences, including H3K27me3 and H4K20me3 on the maternal allele and H3K9me3 on the paternal allele [10].
The Rasgrf1 locus presents some unusual paradoxes: The paternal allele is active yet it carries DNA methylation and other repressive marks, whereas the maternal allele is silent and lacks DNA methylation but carries other repressive marks. It is unclear if and how the primary DNA sequence controls each of these parent-specific marks. We have identified the DNA sequences that are necessary and sufficient for programming the establishment and maintenance of DNA methylation on the paternal allele, however, nothing is known about the cis-acting DNA sequences that control placement of repressive histone modifications in this region, or whether there is any coordination between the histone and DNA methylation modifications. In many organisms, distinct epigenetic marks coordinately determine the transcriptional status of genes. For instance, recruitment of DNA methylation can depend upon pre-established histone H3 methylation at lysine 9 [11]–[13]; histone modifications can be lost when DNA methylation is impaired [14]; and some histone modifications become redistributed in histone methyltransferase mutants [15].
Here we report the analysis of a 12 kbp region at Rasgrf1 for locations bearing histone modifications and DNA methylation. Our data reveal the mutual exclusion of the repressive H3K27 methylation and DNA methylation modifications. Furthermore, by experimentally manipulating the levels of DNA and H3K27 methylation possible at the locus, we demonstrate that these two marks are mutually antagonistic, whereby the placement of one mark prevents the placement of the other, and removal of one mark allows the encroachment of the other. Additionally, we found that the tandem repeat sequences, which are necessary and sufficient for programming DNA methylation marks, are also important for directing H3K27 and H3K9 modifications to the proximal DMD and that H3K9 methylation is needed for optimum establishment of DNA methylation on the paternal allele.
There are two regions rich in C and G residues and CpG dinucleotides over a 200 kbp interval at the Rasgrf1 locus. One CpG cluster is in the ICR and the other in the promoter region of Rasgrf1 (Figure S1A, B, C). By analyzing the DNA methylation pattern of these two CpG clusters in somatic DNA using methylation sensitive restriction enzymes, we found that only the DMD CpG cluster is methylated while the one in the promoter is not (Figure S1D). When Kobayashi et al. performed a comprehensive analysis of allele-specific DNA methylation at the Rasgrf1 ICR in embryonic day 12.5 DNA and in the male germline, they observed that the somatic and germline DMD was larger than had been previously appreciated ([9] and Hisato Kobayashi and Hiroyuki Sasaki unpublished data). We expanded upon this by characterizing the distribution of both histone modifications and DNA methylation over a 12 kbp region centered on the DMD within the ICR, and also by evaluating the influence of the tandem repeats within the ICR on these epigenetic marks (Figure 1).
We performed bisulfite sequencing to characterize DNA methylation in 86 CpGs present in eight segments (labeled D1 through D8 in Figure 1) containing 4,118 bp from the 12,020 bp interval. Our analysis of methylation in the soma used DNA from neonatal brain and our analysis in the male germline used DNA isolated from sperm. Somatic DNAs were from F1 progeny of 129S4Jae and PWK strains. Polymorphisms between these strains allowed us to determine which bisulfite sequences were from the maternal and paternal alleles in the soma. The 129S4Jae-derived allele was either wild type or lacked the Rasgrf1 tandem repeats constituting the DNA methylation programmer (Figure 2A). Sperm DNAs were from mice homozygous for wild type or tandem repeat-deficient alleles of Rasgrf1. Our characterization of the somatic methylation states from animals carrying the wild type 129S4Jae allele was in strong agreement with the results of Kobayashi, even though sources of somatic DNAs differed: Kobayashi used midgestation embryos. In neonatal brain DNA, we detected paternal allele-specific DNA methylation, which covers at least the 7.6 kbp interval between segments D4 through D7 and includes the ICR. We also found a region of methylation on both alleles over a 1.4 kbp interval upstream of the ICR containing segments D1 and D2. None of the somatic DNA methylation patterns changed on either the paternal or maternal alleles in mice harboring a deletion of the tandem repeats on the maternal allele (Figure 2C, D). In contrast, all paternal allele-specific DNA methylation we detected in regions D4 to D8 was lost from somatic DNA when the tandem repeats were absent from the paternal allele (Figure 2D). This indicates that the range of action of the Rasgrf1 DNA methylation programmer within the tandem repeats is not confined to the narrowly defined 400 nt DMD previously studied, but its reach spans at least 7 kbp in somatic tissue.
Imprinted DNA methylation patterns that are established in the germlines are typically maintained and can even spread during somatic development. To determine the extent of the methylation in sperm DNA and the range of action of the DNA methylation programmer in the male germline, we performed bisulfite analysis on sperm DNA from mice carrying an intact repeat element and also from mice carrying a deletion of the repeats. In mice with the intact repeats, we found that Rasgrf1 methylation in sperm DNA was present not only the originally defined 400 bp DMD, but it extended an additional 1.6 kbp upstream, in agreement with results from Hisato Kobayashi and Hiroyuki Sasaki (unpublished). However, in mice bearing a deletion of the repeats that constitute the Rasgrf1 DNA methylation programmer, only the DNA methylation at the originally defined DMD was lost. The DNA methylation on the additional 1.6 kbp was unaffected, indicating that the range of action of the Rasgrf1 DNA methylation programmer in the tandem repeats is limited to the 400 bp proximal DMD in the male germline (Figure 2B). Because loss of DNA methylation on that narrowly defined sequence was sufficient to disrupt imprinted expression of Rasgrf1 [2], we infer that this differentially methylated portion of the locus is essential for its imprinting and we refer to it as the core DMD.
We next characterized histone methylation status across the same 12 kbp interval over which the DNA methylation was characterized. Specifically, we sought to determine where histone modifications were distributed, if any modifications were allele-specific, if their placement required the same DNA methylation programmer that imprinted DNA methylation requires, and if there is any coordination between modification states on histones and DNA. We limited our analysis to di-, and tri-methylation of histone H3 at lysine 9 and 27 because they are associated with gene silencing and DNA methylation, which are observed at the maternal and paternal alleles respectively. For this analysis, we performed chromatin immunoprecipitation (ChIP) using mouse embryonic fibroblasts (MEFs) and antibodies specific to H3K9me2, H3K9me3, H3K27me2, and H3K27me3. Our initial tests were controls to verify that the antibodies detected histone modifications with proper specificity (Figure S2). For these tests, we amplified immunopreciptates using primers from Charlie, Actin and Hoxa9. H3K9me2 and H3K27me2 are known to reside at Charlie [16], H3K9me3 at Actin, and H3K27me3 at Hoxa9 [17]. The expected PCR products were observed for each immunoprecipitation, indicating the antibodies were indeed specific. In addition, PCRs done using DMD primers detected only H3K9me3 and H3K27me3 at the DMD; therefore, subsequent ChIP studies primarily used antibodies recognizing these marks (Figure S2).
We then extended our H3K27me3 and H3K9me3 analysis to six segments (labeled C1 through C6 in Figure 1) that included 10,451 bp surrounding the core DMD and methylation programmer using two separate immunopreciptates from wild type MEFs, and MEFs carrying a deletion of the DNA methylation controlling repeats (RepΔ. Because we used two immunoprecipitates, these analyses report the general distribution of histone marks in the region rather than providing reliable quantification of their abundance. Our PCRs in regions C1, C2, C4, C5 and C6 did not distinguish the parental alleles and our PCR of the DMD at C3 used wild type allele-specific primers. The ChIP analysis of wild type MEFs indicated that both H3K9me3 and H3K27me3 were most abundant at the core DMD at region C3 with some H3K27me3 signal extending downstream of the tandem repeats (Figure 3). Analysis of MEFs carrying a deletion of the DNA methylation controlling repeats suggested that the repeats could influence the distribution of histone modifications at the DMD and elsewhere in the region.
To provide statistically significant measures of methylated H3K9 and H3K27 at the DMD and to assess if any modifications were allele-specific, we analyzed a total of six to twelve independent immunoprecipitations by quantitative PCR using primers spanning the DMD at region C3 (Figure 2A). Our data confirmed that the DMD is enriched for trimethylated lysines but lacks dimethylated ones (Figure 4A). To determine if these histone marks over the DMD were on the maternal or paternal alleles, we repeated the ChIP assays using mice carrying the engineered polymorphisms shown in Figure 2A that enabled us to amplify the wild type maternal and paternal DMD sequences separately. Results demonstrated that the maternal allele has H3K9me3 and H3K27me3, whereas the paternal DMD has only the H3K9me3 mark (Figure 4B and C). This is in partial agreement with other data describing H3K27me3 as being maternal allele specific and H3K9me3 as being paternal allele specific at Rasgrf1 [10]. H3K9me3 that we detect on the two alleles may be placed by different mechanisms. Our data correlate well with previous findings that DNA methylation can be coregulated with H3K9me3 [11],[13],[18],[19], but generally not with H3K27me3 [20],[21].
Because the tandem repeats act as a DNA methylation programmer, playing an essential role both in establishment and maintenance of DNA methylation at the DMD ([2],[3] and YJP, HH, AML, Ying Gao and PDS, in preparation), we wanted to determine if they also influence placement of methylated histone marks at the DMD. We did this by repeating the ChIP analysis using MEFs carrying a deletion of the repeats (RepΔ) and amplifying the wild type allele and the mutated allele separately. Our analysis showed that the repeat element indeed has a significant influence of histone modification status at the DMD, in addition to controlling its DNA methylation (Figure 4D): When the repeats were absent from the maternal allele, the levels of maternal allele-specific H3K27me3 and H3K9me3 were respectively 1/2 and 1/6th the levels seen when the repeats were present. Similarly, when the repeats were absent from the paternal allele, the level of paternal allele-specific H3K9me3 was 1/3rd that seen when the methylation programmer was absent. Interestingly, deletion of the repeats from the paternal allele led to a three-fold increase in the accumulation of H3K27me3 on the paternal allele. This is consistent with our locus wide ChIP analysis spanning intervals C1 to C6, which suggested H3K27me3 can encroach into areas where it is normally absent, both 5′ and 3′ of the DMD, when the paternal repeats are deleted (see sites C2, C5, C6 in Figure 3B). These observations provide evidence that DNA methylation and H3K27me3 are mutually exclusive epigenetic marks at Rasgrf1. When we superimposed the DNA and H3K27 methylation data for wild type animals and animals carrying a deletion of the paternal methylation programmer from Figures 2 and 3, the mutual exclusion of H3K27me3 and DNA methylation over the core DMD was apparent (shown separately in Supporting Figure S3A and B for clarity).
Mutual exclusion of H3K27me3 and DNA methylation can arise by different mechanisms. In one scenario, the two marks may be placed in different compartments of the nucleus and the DNA cannot reside in both places. Alternatively, distinct factors needed for H3K27me3 and DNA methylation may require the same DNA binding site, which cannot be simultaneously occupied by the two sets of factors. A third possibility is that DNA and H3K27 methylation are mutually antagonizing, whereby one inhibits placement of the other. This last possibility is mechanistically different from mere mutual exclusion. If antagonism between these two marks is occurring, then we can predict what happens to one mark if the other is experimentally manipulated.
In order to explore more directly the possible antagonism between DNA methylation and H3K27me3, we repeated our allele-specific ChIP studies using MEFs that had been treated with the DNA methyltransferase inhibitor 5-azacytidine. If DNA methylation can antagonize H3K27 methylation, then we expected that 5-azacytidine treatments should increase the levels of H3K27me3 at the DMD as assayed by ChIP. This is precisely what we observed. 5-azacytidine treatments increased the signals from our ChIP analysis by more than six fold when we assayed H3K27me3 on the two parental alleles (Figure 5A). Although the maternal allele lacks imprinted DNA methylation, there is DNA methylation at sites D1, D2 and D8. Reductions in methylation at those sites might augment accumulation of H3K27me3 across the entire region.
If DNA methylation antagonizes H3K27 methylation, then an additional expectation is that inappropriate placement of DNA methylation on the maternal DMD should exclude accumulation of H3K27me3 marks. To test this, we took advantage of a transgenic system we developed to test if the tandem repeats, which are necessary for programming DNA methylation at Rasgrf1, are sufficient to impart imprinted methylation to the DMD at an ectopic location in the genome. Independent transgenic founders harboring three to five ectopic copies of the Rasgrf1 ICR underwent proper establishment of DNA methylation at the transgenic DMD in the male germline and erasure of that methylation in the female germline, recapitulating the essential features of imprinted methylation establishment seen at the endogenous locus (YJP, HH, AML, Ying Gao and PDS in preparation). We were able to distinguish the transgenic ICR from the endogenous copy because the transgenic repeats were flanked with loxP sites and had the same structure as the WT-flox allele shown in Figure 2A. This allowed us to assay DNA methylation and perform ChIP analysis of the transgene. The transgene was useful for the studies we describe here because the unmethylated state that was established on the transgene after female transmission could not be maintained if there was any history in the pedigree of transmission through a male (Figure 5B). This system of transgenerational epigenetic memory allowed us to generate two different sets of MEFs, both of which were derived by maternal transmission of the transgene from a common founder. For one set of MEFs, the transgene was unmethylated at the transgenic DMD, whereas in the second set, the same transgene was methylated on the DMD (Figure 5B, upper two panels). If there is antagonism between H3K27me3 and DNA methylation at Rasgrf1, then we predicted our MEFs with a methylated transgene should exclude H3K27me3, whereas our MEFs with an unmethylated transgene should allow its placement. This is also precisely what we observed (Figure 5B, lower panel), providing additional independent evidence that DNA methylation antagonizes placement of H3K27me3.
We next wondered if the antagonism between DNA methylation and H3K27me3 might be reciprocal, meaning; H3K27me3 is able to exclude DNA methylation. To test this possibility we analyzed the DNA methylation state of the Rasgrf1 DMD in ES cells, embryoid bodies or trophoblast outgrowths that lack either of two factors needed for H3K27me3 by the PRC2 complex. PCR2 includes SUZ12, EED and EZH2, the H3K27 methyltransferase. YY1, the mammalian ortholog of the Drosophila PHO protein, is a DNA binding factor that binds to EED and recruits PRC2 to sites of action [22]–[24]. Mice and cells with deficiencies in either SUZ12 or YY1 fail to acquire normal levels of H3K27me3 genome wide [22],[25],[26], and the deficiency is lethal for mice, but SUZ12-deficient ES cells are viable [26],[27].
If conditions necessary for proper placement of H3K27me3 are in fact required to antagonize placement of DNA methylation on the maternal DMD of Rasgrf1, then DNA methylation at the DMD will increase in the absence of SUZ12 and YY1. Because DNA methylation at the Rasgrf1 DMD is normally restricted to the paternal allele, which is completely methylated, any increase in DNA methylation would arise on the maternal allele. To monitor the level of Rasgrf1 DMD methylation in SUZ12- and YY1-deficient materials, we used COBRA [28]. This involved treating DNAs with bisulfite and amplifying them using primers not overlapping with CpG dinucleotides, which will amplify templates without bias for either methylation state. We then digested the PCR products with BstUI. Methylated templates will retain the BstUI recognition site (CGCG) after amplification and will be digested, whereas unmethylated templates that underwent bisulfite conversion of either CG in the recognition site to TG will resist digestion. There should be an equal amount of digested and undigested PCR product when the maternal allele is completely unmethylated and the paternal allele is completely methylated. This is what we saw in embryoid bodies and blastocysts that were heterozygous respectively for the Suz12 and Yy1 mutations. This indicated that our COBRA assays accurately reported the presence of both methylated and unmethylated templates expected in these Suz12 and Yy1 expressing materials; however, it is not clear why Suz12 heterozygous ES cells did not show this pattern. When we performed COBRA analysis on SUZ12-deficient embryoid bodies (EB) that had differentiated for six (P6) or nine (P9) days in vitro (Figure 6A, B) or on trophoblast outgrowths from YY1-deficient blastocysts (Figure 6C), we found a dramatic increase in the levels of the digested PCR product in three out of four samples of Suz12 −/− material and in the Yy1 −/− material, indicating that loss of SUZ12 or YY1 resulted in increased Rasgrf1 DMD methylation. The near complete acquisition of DNA methylation in P9 EB lacking SUZ12 was confirmed by bisulfite sequencing (Figure 6B lower panel), whereas unmethylated DNA was present in EB with a single functional allele of Suz12, though it is possible there is a quantitative increase in Rasgrf1 DNA methylation when only one functional copy of Suz12 is present. We do not know why only three out of four of the Suz12 −/− DNAs show hypermethylation. This could be an artifact of cell cultures, which can exhibit frequent and cyclic changes in DNA methylation [29]. Also, mutation of Eed, another component of PRC2, is known to cause hypermethylation and hypomethylation simultaneously, depending upon which CpGs are queried [30]. Given these precedents, it is possible that the eight CpGs we assayed in the BstUI sites are predominantly hypermethylated in cultured cells lacking SUZ12. Nonetheless, our data provide evidence that the antagonism between DNA and H3K27 methylation is reciprocal and that H3K27me3 antagonizes placement of DNA methylation. Furthermore, this mutual antagonism exists in at least three DNA sources: MEFs, embryoid bodies and trophoblast outgrowths.
We also explored the relationship between H3K9 and DNA methylation at Rasgrf1. H3K9 methylation has been strongly correlated with DNA methylation (reviewed in [31]): Loss of the SUV39H1 and SUV39H2 H3K9 methyltransferases in mice simultaneously impairs accumulation of H3K9me3 across the genome [32] and accumulation of DNA methylation at pericentric major satellite repeats, but not at minor satellite or C-type retroviral repeats [13]. DNA methylation deficiencies were also noted in plants lacking H3K9 methyltransferases [12],[19] with one study reporting that maintenance of DNA methylation was affected [18]. To investigate the relationship between H3K9me3 and DNA methylation at Rasgrf1, we asked if H3K9me3 controlled by SUV39H1 and SUV39H2 affected imprinted DNA methylation at Rasgrf1. To address this, we performed methylation analysis on adult testes DNA using COBRA, bisulfite sequencing and a PCR assay that detected methylation status at a series of five HhaI sites in the DMD. Testes primarily contain cells of the germline, which will carry paternal epigenotypes, but some somatic cells are also present, which will carry both paternal and maternal epigenotypes. The COBRA analysis suggested that the DNA was hypomethylated in the SUV39H1- and SUV39H2-doubly deficient testes (Figure 7A). When we measured the extent of DNA methylation using HhaI site-spanning Q-PCR assays, it was clear that the loss of Rasgrf1 DNA methylation was significant (Figure 7B). Bisulfite sequencing provided additional confirmation with higher resolution – there was a significant decrease in the number of DNA templates that were more than 80–100% methylated and an increase in the number that were 40–80% methylated in SUV39H1- and SUV39H2-doubly deficient testes (Figure 7C) but there was no change in the abundance of DNAs that were completely unmethylated. The reduction in DNAs with the 80–100% methylated paternal epigenotype, and the increase in DNAs with the 40–80% methylated epigenotype suggests that SUV39H1 and SUV39H2 control the efficiency with which imprinted DNA methylation is established in mice. In Arabidopsis, the SUVH4 H3K9 methyltransferase is known to control maintenance of DNA methylation [18].
We report here the epigenetic states that exist within a 12 kbp interval centered on the Rasgrf1 ICR. Both parental alleles were marked by DNA methylation in somatic tissue on a 1.4 kbp segment at the very 5′ end of this 12,020 nt interval (D1–D2, Figure 2C, D). Downstream of this were segments that spanned the ICR that were paternally methylated in somatic DNA (D3–D8), and sperm DNA as well (D3–D5, Figure 2B,D). Not every CpG was assayed in this 12,020 interval, including those within the tandem repeats that constitute the DNA methylation programmer. H3K9me3 was present on both parental alleles at the core DMD immediately 5′ of the tandem repeats and within the ICR. H3K27me3 was present at this same location, but exclusively on the maternal allele. There was no appreciable dimethylation of these H3 residues at the core DMD (Figure 4A, B, C).
The tandem repeats, consisting of approximately 40 copies of a 41 nt unit, influenced the placement of histone and DNA methylation (Figures, 2B, D, 3 and 4D) and can be considered a cis-acting methylation programming sequence, one of only a few naturally occurring ones known in mammals. Paternal allele DNA methylation was particularly sensitive to these tandem repeats, which control establishment of DNA methylation in the male germline at a 400 nt core DMD lying just 5′ of the repeats (Figure 2B and [2]). The repeats also control spreading and maintenance of paternal allele DNA methylation in somatic tissue over a broader domain (Figure 2B, D and [3]).
Marking the core DMD with DNA methylation on the paternal allele and H3K27me3 on the maternal allele are coordinated and mutually exclusive events in wild type cells with DNA methylation largely confined to the core DMD on the paternal allele and H3K27me3 on the maternal allele (Figures 2D, 3B and 4). The mutual exclusion arises because one epigenetic mark antagonizes the placement of the other. Five independent lines of evidence led to this conclusion. First, MEFs taken from mice lacking DNA methylation on the paternal DMD inappropriately accumulated H3K27me3 on the paternal allele (Figure 4D). Second, MEFs treated with the DNA methyltransferase inhibitor, 5-azacytidine, accumulate elevated levels of H3K27me3 marks (Figure 5A). Third, MEFs taken from mice with a maternally transmitted Rasgrf1 ICR transgene that lacked DNA methylation had H3K27me3 on the transgenic DMD, whereas H3K27me3 was excluded by manipulations that inappropriately placed DNA methylation on the transgene (Figure 5B, lower panel). Fourth, mutation of the Suz12 component of PRC2, which is needed for activity of the EZH2 H3K27 methyltransferase in PRC2, ablated normal placement of H3K27me3 and enabled the maternal allele to inappropriately acquire DNA methylation (Figure 6A, B). Fifth, mutation of the Yy1 gene, which is needed to recruit PRC2 to DNA and, like Suz12, is needed for effective placement of H3K27me3 also enabled the maternal allele to inappropriately acquire DNA methylation (Figure 6C). Other studies have documented the cross-dependency of some histone modifications and DNA methylation [11], [13], [14], [19], [33]–[37], and it has also been observed that H3K27me3 and DNA methylation can be mutually exclusive [21]. The studies described here provide evidence that H3K27me3 and DNA methylation are in fact mutually antagonizing epigenetic marks and that H3K27me3 facilitates allele-specific DNA methylation that exists at imprinted loci.
H3K9me3 was detected on both parental alleles indicating this mark is controlled differently from H3K27me3. However, it too participates in imprinted DNA methylation because the H3K9 methyltransferases, SUV39H1 and SUV39H2, are needed for optimal establishment of DNA methylation at the DMD in the male germline (Figure 7).
We do not know how DNA and H3K27 methylation antagonize each other's placement; however, the literature highlights several molecular and developmental events, as well as protein factors that may be involved. Among these is the transcriptional state that is known to influence which of two mutually exclusive histone modifications is placed by the competing activities of polycomb (PcG) and trithorax (Trx) group proteins [38]. Additionally, differentiation state is known to influence genome wide epigenetic patterns in ES, MEF and neuronal progenitor cells [39]. At Rasgrf1, developmental stage also influences epigenetic states [40]; the methylation programmer controls establishment of DNA methylation in the germline and maintenance in peri-implantation embryos [2],[3], but not later in development. Interestingly, this same period is a critical interval for control of H3K27 methylation [41]. Finally, there may be a role for CTCF in the mutual exclusion of H3K27 and DNA methylation at Rasgrf1. CTCF and its binding sites have been shown to influence H19 DNA methylation [42]–[47] and CTCF binds at Rasgrf1 as well [4]. Genome-wide ChIP analysis identified locations enriched for CTCF [48] and H3K27me3 [49] in MEFs and Chi squared analysis reveals a significant co-localization of these marks at imprinted versus non-imprinted loci (Table S3). This raises the possibility that, in addition to its role in preventing DNA methylation at other imprinted loci, CTCF helps to place H3K27me3 at Rasgrf1. CTCF functions in coordination with its binding partner, YY1, in activating the X chromosome [50] and YY1 also inhibits DNA methylation at Rasgrf1 (Figure 6C), most likely through its ability to recruit PRC2 [22]–[24]. Depending upon the consensus sequence considered, between one and twelve YY1 sites are predicted to lie within the DMD and repeat region (data not shown). Like CTCF, YY1 sites are enriched at other imprinted loci as well [51]. CTCF has additional binding partners including CHD8, which is associated with DNA methylation [52]. Using ChIP analysis, we could not detect CHD8 on either Rasgrf1 allele (data not shown), suggesting that at Rasgrf1, other CTCF binding partners and functions might be more important, such as YY1.
Normal placement of DNA methylation on the paternal allele and H3K27me3 on the maternal allele both require the same tandemly repeated DNA sequence element, which we previously showed has DNA methylation programming activity ([2],[3] and YJP, HH and PDS unpublished). However, DNA methylation is more rigidly dependent on the repeated sequence than are the histone modifications. Whereas DNA methylation on the paternal core DMD was typically completely lost when the repeats were deleted, H3K27me3 and H3K9me3 on the maternal DMD were respectively reduced to levels only 1/2 and 1/6 of those seen when the repeats were present. Repeated sequences have been shown to have methylation programming activity in other systems [53],[54]. Notably, at the DM1 locus in humans, a repetitive element is associated with heterochromatin accumulation [55]. Interestingly, like the maternal Rasgrf1 DMD and repeat sequences [4], the DM1 repeat also is a CTCF-binding insulator. CTCF appears to restrict the boundary of heterochromatinization at DM1, but it is not known if CTCF has a similar effect at Rasgrf1. Sequences with appreciable similarity to the Rasgrf1 tandem repeats are not abundant in the mouse genome. However, the Rasgrf1 repeat unit has striking similarity to the B repeat sequences on Xist (Figure S4). Because Xist RNA regulates placement of H3K27me3 on the inactive X chromosome and at an autosomal transgenic site in cis [56],[57], it is possible there is mechanistic overlap between epigenetic regulation by Xist and the Rasgrf1 repeats.
We do not know what functional motifs enable the methylation programmer at Rasgrf1 to control either DNA or H3 methylation. Its repeated nature may be sufficient [54], possibly involving an RNA-dependent mechanism [58]. Other potentially important features include the CpG present in 36 of the 40 repeat units; GGGG tetramers that may facilitate the formation of G-quadruplex structures [59], which in turn may alter the sensitivity of DNA to methyltransferase action [60]; or CTCF sites known to lie in the Rasgrf1 methylation programmer [4]. BORIS, the male germline paralog of CTCF [61], may also be important for function of the Rasgrf1 methylation programmer.
Figure 8 describes a model for the placement of DNA methylation and H3K27me3 in response to the Rasgrf1 methylation programmer, their antagonism, and the developmental timing of these events. However, it is unlikely that a universal rule dictates the regulation of DNA and H3K27 methylation at all loci within a species or among species. In human cell lines, some loci have been found at which DNA and H3K27 methylation occur simultaneously with one mark requiring the placement of the other [62], whereas in Arabidopsis, DNA methylation does not seem to be closely associated with H3K27me3 [20] and in fact can be mutually exclusive [21]. Nonetheless, identifying the various rules that influence epigenetic programming of normal developmental states will provide insights for manipulating them for therapeutic benefit.
Mice used for DNA methylation analysis across the 12 kbp interval were F1 progeny of PWK and 129S4SvJae parents. Polymorphisms in these strains facilitated the assignment of a given clone from bisulfite PCR to one of the two parental alleles. Mice used to prepare MEFs for ChIP analysis across the 12 kbp interval were from strain 129S4SvJae backcrossed to C57BL/6 and included wild type animals, animals carrying a repeat deletion [2] and animals containing an engineered polymorphism at the DMD that did not disrupt imprinting [3]. All allele specific ChIP analyses were done using MEFs from mice carrying one of these mutations. Mice carrying the Rasgrf1 ICR transgene will be described in a separate report (YJP, HH, PDS, in preparation). Previous reports describe the Suz12 mutation and preparation of homozygous ES cells and embryoid bodies [26] and the Yy1 mutation and preparation of trophoblast outgrowths [27].
MEFs from 13.5 day old F1 embryos of C57BL/6 and 129S4Jae parents were used for ChIP analysis as described in Text S1. Modified histone-specific antibodies were from Millipore/Upstate (H3K9me2 item 07-441 lot 29698, H3K9me3 item 07-442 lot 24416, H3K27me2 item 07-452 lot 24461, H3K27me3 item 07-449 lot 24440) and Thomas Jenuwein, IMP, Austria (H3K9me3) [15]. Specificity of antibody from Thomas Jenuwein's lab has been reported [15]. Validations of commercial antibody specificities are publicly available from the manufacturer (see http://www.millipore.com for certificates of analysis for each catalog item and lot number). The DNA recovered after ChIP was used for Q-PCR with input chromatin and mock immunoprecipitations without antibody serving as controls. Q-PCR was performed in triplicate with SYBR green detection using primers listed in Table S1. Ratios of bound to input signals are reported.
Treatment of genomic DNA with bisulfite was performed as previously described [2], with the added difference that we used 1.5 M betaine and 5% DMSO to enhance the yield in PCR of AT-rich, converted DNA. ExTaq HS DNA polymerase (Takara, Japan) was used for hotstart PCR. Primers used are provided in Table S1 and additional experimental details are in Text S1. The bisulfite converted and amplified DNA was either cloned and sequenced or subjected to COBRA [28] using BstUI digestions. In this assay, cytosine methylation enables digestion, whereas absence of methylation prevents it. Assays for DNA methylation using HhaI digested DNAs were described [2].
|
10.1371/journal.pbio.0060105 | Biogenesis of the Trypanosome Endo-Exocytotic Organelle Is Cytoskeleton Mediated | Trypanosoma brucei is a protozoan parasite that is used as a model organism to study such biological phenomena as gene expression, protein trafficking, and cytoskeletal biogenesis. In T. brucei, endocytosis and exocytosis occur exclusively through a sequestered organelle called the flagellar pocket (FP), an invagination of the pellicular membrane. The pocket is the sole site for specific receptors thus maintaining them inaccessible to components of the innate immune system of the mammalian host. The FP is also responsible for the sorting of protective parasite glycoproteins targeted to, or recycling from, the pellicular membrane, and for the removal of host antibodies from the cell surface. Here, we describe the first characterisation of a flagellar pocket cytoskeletal protein, BILBO1. BILBO1 functions to form a cytoskeleton framework upon which the FP is made and which is also required and essential for FP biogenesis and cell survival. Remarkably, RNA interference (RNAi)-mediated ablation of BILBO1 in insect procyclic-form parasites prevents FP biogenesis and induces vesicle accumulation, Golgi swelling, the aberrant repositioning of the new flagellum, and cell death. Cultured bloodstream-form parasites are also nonviable when subjected to BILBO1 RNAi. These results provide the first molecular evidence for cytoskeletally mediated FP biogenesis.
| Trypanosomes are ubiquitous unicellular parasites that infect humans, animals, insects, and plants. African, Asian, and some South American trypanosomes have evolved the amazing ability to change their surface coat proteins, an essential strategy for their survival. The surface coat proteins are recycled and targeted to the surface of the parasite via an endocytic and exocytotic organelle called the flagellar pocket, which is sequestered in the trypanosome cell's cytoplasm. The flagellar pocket is also used to remove host-derived antibodies that are bound to the surface of the parasite, making this organelle critical for the parasite's evasion of the host immune system. We describe a novel protein, “BILBO1,” which was identified from the insect-form parasite of the African trypanosome Trypanosoma brucei. We show that BILBO1 is part of a ring or horseshoe-like cytoskeletal structure that is located in a region of the flagellar pocket called the collar. When BILBO1 transcripts were knocked down with inducible RNA interference, trypanosome cells became arrested in a post-mitotic cell-cycle stage. Induced cells lost the normal flagellum-to-cell-body attachment, were unable to regulate endocytosis and exocytosis, and most importantly, were unable to construct a new flagellar pocket. These results provide molecular evidence for the idea that flagellar pocket biogenesis is cytoskeletally mediated.
| Endocytosis and exocytosis in trypanosomes is performed by the flagellar pocket (FP), an important organelle that is sequestered within the cytoplasm of the posterior region of the cell. On the basis of its protein composition, the FP membrane is biochemically distinct from the flagellar or pellicular membranes [1–3] and is also required for the molecular trafficking and recycling of glycosylphosphatidylinositol (GPI)-anchored proteins such as procyclin and variable surface glycoproteins (VSG). Both procyclin and VSG are surface coat proteins that are trafficked and recycled from the cytoplasm via the FP to the cell surface, where they function in the survival strategies of the cell. The molecular processes involved in these trafficking events are complex and require clathrin, actin, and a number of important GTPase Rab proteins [4,5].
An additional important feature of the FP is that it is physically linked to the cytoskeleton. This linkage can be observed in two areas of the pocket: (1) the axoneme of the flagellum traverses the FP prior to exiting the cytoplasm, and (2) the neck of the FP originates from a cytoskeletal structure that appears to be attached to the flagellum [6]. Similar to the sequence of events observed in kinetoplast segregation, the segregation of the new FP is precisely temporally and spatially coordinated, which implies that this process maybe mediated by the flagellum [7]. The site of FP biogenesis may also be mediated by the flagellar axoneme, as is observed for the Golgi apparatus of T. brucei [8]. The existence of a FP–cytoskeleton linkage would therefore explain the exquisite precision in positioning and segregation of the FP; namely that, the FP is always located on the proximal, cytoplasmic portion of the flagellum axoneme, and FP segregation is tightly coordinated with the flagellum biogenesis and segregation cycle.
The formation of a new flagellum is tightly associated with the maturation and elongation of the probasal body that is associated with the old flagellum [6]. The new axoneme then traverses the luminal core of the FP and exits the cell at a constricted site called the flagellar pocket collar (FPC) [9,10]. Although a number of proteins have been characterised to be specific to the FP, most are not essential, and none function in FP biogenesis [11–17]. For example clathrin and the proteins needed for receptor-mediated endocytosis are sequestered to the FP but are never exposed to the cell surface [2]. To date, data on FP organisation have been extremely limited and mainly based on ultrastructural studies. No proteins of the FPC have been identified, which is surprising in that the FP has the important role of controlling the targeting of molecules to and from the cell surface to avoid the host immune system [18–22]. Furthermore, endocytosis via the FP is not only used for the trafficking of parasite-derived molecules, but is used in the clearance of host antibodies bound to the cell surface [22–26].
In this study, we describe the identification and characterisation of the first cytoskeletal flagellar pocket protein: BILBO1. BILBO1 is a component of a cytoskeletal framework that is essential for biogenesis of the FPC. RNA interference (RNAi) ablation of BILBO1 in cultured procyclic insect forms (PF) of T. brucei prevents FP biogenesis, thus disturbing endocytotic activity and inducing vesicle accumulation, Golgi swelling, gross repositioning of the new flagellum, and cell death. Furthermore, cultured bloodstream forms (BSF) are not viable when subjected to BILBO1 RNAi in vitro. BILBO1, therefore, provides an interface between the cytoskeleton and the endocytotic and exocytotic systems, and represents the first molecular component of the FPC to be identified.
T. brucei has proven to be an excellent model for the study of cytoskeletal biogenesis [10,27]. T. brucei has several single-copy organelles, such as the mitochondrion, the kinetoplast (mitochondrial genome), a single Golgi apparatus, and a single FP. The endocytotic and exocytotic activity is limited to the posterior region of the cell via the FP. Figure 1A illustrates the morphology of a PF cell and shows the overall position of the FP within the cell. In PF cells, the FP is always closely associated with the flagellum, and both structures are always located in the posterior of the cell.
To attempt to characterise minor but essential proteins in the flagellum of T. brucei, salt-extracted flagellar proteins were separated by polyacrylamide gel electrophoresis (PAGE), and slices of the gel were used to immunise mice. Polyclonal serum obtained from these mice was then used to probe for novel proteins in immunofluorescence and western blotting studies. Proteins that appeared novel by immunofluorescence analysis were further investigated and eventually identified by mass spectrometry. From these studies, we identified a novel 67.3-kDa flagellar protein that we named BILBO1.
A single-copy gene, located on chromosome 11 of the T. brucei genome, encodes the BILBO1 protein. BLAST analysis, using parasite GeneDB databases, identified eight orthologs of BILBO1: one in T. brucei gambiense, one in T. congolense, two in T. cruzi (the South American trypanosome), one in T. vivax, one in Leishmania major, one in L. infantum, and one in L. braziliensis. With the exception of T. congolense, these genes have very similar locations with regard to their respective flanking genes, indicating that BILBO1 gene synteny is preserved amongst these species. BLAST analysis of the genes of non-kinetoplastid organisms lacking a FP did not identify any other homologs to BILBO1. The primary and secondary structures of BILBO1 do not predict any localisation or cytoskeletal functions; however, this protein does possess two putative EF-hand calcium-binding motifs (amino acids [aa] 185–213 and 221–249), suggesting the existence of calcium binding sites and possible roles in regulation. The large C-terminus coiled-coil domain (aa 263–566) signifies a role in oligomerisation or protein–protein interactions.
In vivo overexpression of enhanced green fluorescent protein (eGFP)-tagged BILBO1 (Figure S1A) in PF cells localised the protein to the FPC (Figure 1B). In addition to the eGFP labelling experiments, we also raised antiserum to recombinant BILBO1 protein or peptides. Immunoelectron microscopy and immunofluorescence studies on PF cytoskeletons confirmed the eGFP-tagged BILBO1 localisation data (Figure 1C–1G). Identical immunofluorescence FPC–FP localisation was observed on BSF cytoskeletons (Figure S1B).
The anti-BILBO1 immunogold labelling observed in Figure 1C forms a horseshoe structure that is oriented around the emerging axoneme. No label is observed directly on the axoneme, suggesting that BILBO1 is present on the cell body side of the FPC as opposed to the flagellum side. However, when cytoskeletons are treated with 1 M NaCl, little or no BILBO1 protein is extracted, and the BILBO1 signal remains associated with the flagella preparation (Figure S1C). This indicates that, although BILBO1 is located on the cell side of the flagellum, the FPC and the BILBO1 protein both remain tightly linked to the flagellum. This observation is further supported by the presence of BILBO1 protein in the T. brucei flagella proteome [28].
In order to understand the biogenesis of the FPC, we examined cells in different cell-cycle stages labelled with the anti-BILBO1 antiserum. We observed that early in the kinetoplast S phase (as observed by kinetoplast DAPI staining), the old maternal FPC elongates and grows along its long principal axis, followed by a complete constriction of the short principal axis, thus forming two FPC structures (Figure 1E and 1F). During kinetoplast S phase, one of the FPC structures is moved towards the cell posterior along with new flagellum migration (Figure 1F and 1G). Figure 1F illustrates that division and segregation of the FPC occurs before kinetoplast S phase is completed. The new FPC appears to be segregated simultaneously with the new flagellum.
In wild-type (WT) cells, new flagellum segregation is accomplished by a subpellicular microtubule-mediated mechanism, which moves only the new flagellum towards the posterior end of the cell [7]. Since the FP is always physically linked to a flagellum, the simultaneous separation of the new flagellum and FP suggests that they may both be segregated by the same microtubule-mediated mechanism. Taken together, these data indicate that the mother FPC participates in daughter FPC biogenesis.
We used the tetracycline-inducible RNAi system to assess the function of BILBO1 in PF cells and BSF cells [29,30]. Cell growth was arrested in the PF cells after 24 h of induction, followed by cell death (as judged by a reduction in cell numbers over time) after 48–72 h of induction. In induced BSF cells, cell death, (as judged by a reduction in cell numbers over time), began after approximately 24 h of induction (Figure S1G and S1H). Note that western blot studies show that BILBO1 protein was not completely depleted in PF cells at 72 h after induction (Figure 2A). Densitometry data of PF cells indicate that at 24 h of induction, BILBO1 protein levels had dropped to 44.2% of parental levels and to 27.5% and 19.6% at 48 h and 72 h, respectively. When we observed PF cells by immunofluorescence after 36 h of BILBO1 RNAi induction, the BILBO1 signal was weak and only detectable on the mother FPC (unpublished data).
During BILBO1 RNAi induction, we observed that PF cells were elongated and supported new motile flagella but displayed an aberrant flagellum–cell body attachment. New flagella were attached to the cell body only through the basal body and were relocated to the distal portion of the aberrantly elongated posterior end of the cell. Antibody labelling of the basal body or paraflagellar rod (PFR) (a flagellar structure required for flagellar motility) indicated that the new flagellum was positive for the basal body and PFR proteins of these structures. The new flagellum was also closely associated with a new kinetoplast, as observed by immunofluorescence and DAPI staining (Figure 2B). Thin-section transmission electron microscopy observation of induced cells illustrated the astonishing finding that the FPs of these cells were not duplicated. Thus, no new FP were formed at the site of new flagellum growth (Figure 2D and 2E). However, the kinetoplast had duplicated and remained attached to, and was segregated by, the basal bodies of the new flagellum (Figure 2B, 2D, and 2E).
Taken together, these data indicate that (1) the formation of the FP requires BILBO1 protein and (2) that a reduction of BILBO1 protein levels by approximately 50% prevents FP formation and leads to cell death.
The image shown in Figure 2C illustrates a WT PF cell longitudinally sectioned at the FP level. This image clearly illustrates that the transition zone of the mature basal body (as shown by the arrowhead in the figure) is positioned within the FP lumen [31,32] and the PFR originates at the point where the axoneme exits the pocket [6,29]. In BILBO1 RNAi-induced cells, however, flagellum-to-cell body attachment has been disrupted, and the new basal body and transition zone are external to the cell body (Figure 2B and 2E). As with control cells, the origin of the new PFR in induced cells is also distal to the transition zone. Because the new flagellum of induced cells is attached to the cell only through the basal body region, this observation suggests that the axoneme itself contains the information necessary for determining where the PFR originates, as opposed to a signal or marker derived from attachment to the cell body. A higher magnification image of the basal body region of an induced cell (Figure 2E) illustrates the absence of a FP but also that the kinetoplast remains associated with, and segregated by, the basal body; it also shows the abnormal presence of microtubules in the cytoplasm at the proximal end of the basal body. The electron-dense material corresponding to the FPC at the exit site of the flagellum is clearly visible in noninduced cells but is not visible at the exit site of the new flagellum in induced cells, supporting the perception that BILBO1 RNAi cells do not form a new FP or a FPC (Figure 2D and 2E).
Overexpression of nontagged BILBO1 in PF cells did not produce any obvious aberrant phenotypes other than a slight delay in growth rate compared to WT cells (unpublished data). Immunofluorescence studies on cells overexpressing BILBO1 showed that the protein localised to both the mother and daughter FPCs (unpublished data). Intriguingly, overexpression of amino or carboxyl terminal eGFP-tagged BILBO1 for 24 h induced a large accumulation of the tagged protein at the mother FPC. Longer induction of eGFP-tagged BILBO1 (48 h) induced growth arrest (unpublished data). As with the RNAi cells, induction of eGFP-tagged BILBO1 produced cells with new motile flagella that were relocated to an aberrantly elongated posterior portion of the cell. These new flagella were not associated with any BILBO1 immunofluorescence signal or BILBO1-eGFP fluorescence signal, suggesting that no new FPC was formed. Induction of eGFP-tagged BILBO1 also initiated a disruption of new flagella-to-cell body attachment (unpublished data). Thin-section electron microscope images of these cells showed that they had accumulated abnormally large numbers of cytoplasmic vesicles, indicating that they had considerable endo- and exocytotic defects (unpublished data). These cells displayed similar phenotypes to the RNAi-induced cells described previously thus we propose that lethality is not due to overexpression of BILBO1-eGFP per se, but rather due to the inhibition of BILBO1 function via a dominant-negative effect.
Cell counts using DAPI-stained PF cells (Figure 3A) indicated that the ratio of cells with two kinetoplasts and two nuclei (2K2N) increased from 11.73% (standard error [SE] ± 0.63%, n = 1,542) in the nontransformed parental cell line to 24.41% (SE ± 5.73%, n = 811) in induced cells after 36 h of induction. We also detected a decrease in the population of cells with one kinetoplast and one nucleus (1K1N) from 70.24% (SE ± 2.22%, n = 1,542) in the parental cell line to 42.62% (SE ± 4.33%, n = 811) in induced cells after 36-h induction. Interestingly, only 3.57 ± 1.43% of the population were multinucleated, as compared to 1.46% ± 0.7 in the noninduced cells, indicating that induced cells do not continue through mitosis but instead undergo a cell-cycle block at the 2K2N stage (Figure 3A). Within the induced 2K2N population (Figure 3B), 60.06% (SE ± 2.76%, n = 535) of cells possessed an elongated posterior end. Furthermore, 91.04% of induced 2K2N cells had the mispositioned flagellar phenotype, 8.96% of noninduced cells had mispositioned flagella, whereas 3.36% of WT cells had this phenotype. In all BILBO1 RNAi-induced cells, mispositioned new flagella always maintained a disrupted flagellum-to-cell body attachment.
Five distinctive 2K2N phenotypes were observed in induced PF cells (Figure 3B): (1) 2K2N cells that appeared normal in kinetoplast and nuclear positioning (KNKN [8.96% SE ± 0.82%]), (2) KNKN cells with a disrupted loss of new flagellum–cell body attachment phenotype (20.56% SE ± 1.76%), (3) cells with two kinetoplasts and two nuclei positioned kinetoplast–kinetoplast, nucleus–nucleus (KKNN) with a disrupted loss of new flagellum–cell body attachment phenotype (9.63% SE ± 0.63%), (4) elongated KNKN cells (18.33% SE ± 3.01%), and (5) elongated KKNN cells (41.73% SE ± 2.3%) (Figure 3B). The reason for production of KKNN cells is not clear, but it may be related to where the cell is positioned within its cell cycle (e.g., early or late in mitosis) when new FPC biogenesis is inhibited by RNAi knockdown. Intriguingly, a KKNN organisation is observed in normal WT BSF trypanosomes, thus this organelle arrangement may reflect a modified mechanism of organelle segregation in BSF cells compared to PF cells. Noticeably, in all of the induced cells, the new flagella were shorter than the mother flagella (Figure 2B), suggesting that these cells were also experiencing difficulties in delivery of cargo for construction of the new flagellum.
Electron microscopy reveals that induced PF cells possess what appear to be stacks of membranes that resemble a Golgi apparatus. These cells also amass large numbers of vesicles (Figure S1D and S1F). Since Golgi duplication in procyclic T. brucei cells involves Centrin-2 [33], and Golgi separation in T. brucei is basal body mediated [8], we wanted to test whether the observed Golgi swelling influenced Golgi duplication or segregation in induced BILBO1 RNAi-elongated cells. BILBO1-induced cells (36-h induction) were therefore probed with anti-GRASP antibody (Golgi marker) [33] and viewed by immunofluorescence to observe Golgi duplication and segregation. Similar to previous studies on WT cells, we observed two or more major separate Golgi-positive signals in all induced 2K2N phenotypes (Figure 4A–4F) [8,33]. The extended posterior portion of induced 2K2N cells varied in length; therefore, we scored Golgi signals of induced cells that were present in the extreme posterior distal half of the extension as “basal body segregation positive” and Golgi signals in the proximal anterior half as “basal body segregation negative.” In 2K2N WT cells, 98.56% (SE ± 0.26%, n = 764) were segregation positive, whereas 19.98% (SE ± 3.31%, n = 456) of induced cells (36-h induction) were segregation positive. These data indicate that in BILBO1 RNAi cells, Golgi duplication is not inhibited, but the basal body-dependent Golgi segregation machinery is disrupted. This latter observation is due, most likely, to malformations observed in the duplicated Golgi that may block the formation of essential components of the segregation machinery, but also probably related to the loss of cytoskeleton organisation and function in the absence of a FPC at the relocated posterior flagellum.
In trypanosomes, a cytoskeletal structure, called the flagellum attachment zone (FAZ) is thought to be involved in the organisation of the flagellum and cytokinesis. This structure is located in the subpellicular cytoskeleton, where it subtends the flagellum [6,34,35]. FAZ proteins are required for flagellum attachment, and loss of the FAZ induces both a flagellum-to-cell body detachment and an inhibition of cytokinesis [36,37]. The L3B2 monoclonal antibody recognises the cytoplasmic filament of the FAZ in immunofluorescence and in immunoelectron microscopy [34]. We have used the anti-FAZ antibody L3B2 to study the organisation of the FAZ in the context of flagellum positioning in induced cells. Immunofluorescence studies demonstrate that in BILBO1 RNAi-induced PF cells, no new L3B2-positive FAZ filaments are formed, and the FAZ signal observed remains associated only with the old maternal flagellum (Figure 4G–4I). These data illustrate that flagellum and basal body formation are not sufficient for FAZ formation and could imply that the FPC or FP is required for FAZ formation. If this is the case, the absence of the FAZ could induce the absence of normal flagellum-to-cell body attachment. The lack of a FAZ has previously been observed to produce loss of flagellum-to-cell body attachment [36–39]. Alternatively, the absence of FAZ formation could be explained by the fact that new flagella exhibit a rapid flagellum-to-cell body detachment.
As the new flagellum of induced cells consistently exhibits a flagellum-to-cell body detachment, we therefore wanted to determine when in the cell cycle does flagellum detachment occur, and does the new flagellum remain associated to the old flagellum while within the FP? We probed PF cytoskeletons with AB1, which is a monoclonal antibody that localizes to a protein component of the flagellar connector (FC) [40]. The FC is a flagellum–flagellum linkage that is formed during cell division in PF cells [39]. It is present on the distal tip of the new flagellum and is normally tethered at the tip to the lateral aspect of the old flagellum. It is involved in the replication of the helical cell pattern and polarity of trypanosomes. Studies using trypanosome intraflagellar transport (IFT) knockdown cells showed that in the absence of a new flagellum, the FC can still migrate along the old flagellum. Therefore, new flagellum–FC attachment is not essential for FC movement and suggests that the FC has a novel motor for movement along the old flagellum [41].
AB1 labelling of noninduced cells showed the normal attachment of the new flagellum to the old flagellum within the FP, a similar observation to those published previously [40]. The immunofluorescence data presented in Figure 5A show a noninduced PF cell that has been probed with AB1 and a monoclonal antibody (L8C4) that targets the PFR. The merged immunofluorescence and phase contrast images of this cell indicate that a short new flagellum has formed, and the AB1 anti-FC staining shows the presence of the FC new-to-old flagellar attachment site. However, no L8C4 signal is observed on this new flagellum, indicating that it is only a few microns long and is located within the FP. As cells progress through the cell cycle (Figure 5B, 5C, 5E, and 5F), the new flagellum emerges from the FP. The distal tip of the new flagellum remains attached to, and moves along, the old flagellum (Figure 5A–5C).
BILBO1 RNAi-induced and AB1-probed cells showed that there was a FC-positive signal present on the old flagellum, but this signal remained in the FP (proximal to the origin of the old PFR signal) (Figure 5G–5L). Induced non-elongated cells also had formed the FC-positive signal, which, similar to elongated cells, remained in the FP. Based on its short length, and in comparison to WT cells, the new flagellum of an induced cell early in the cell cycle should normally be located within the FP. The example shown in Figure 5M and 5P illustrates that the new flagellum is PFR negative, indicating that it would normally be located within the FP and should be attached to the old flagellum. However, in this case, attachment of the new flagellum did not occur or was transient. The location of this short new flagellum indicates that the new basal body of BILBO1 RNAi-induced cells can “dock” in the proximity of the old pocket in a similar manner to that found in WT cells. Figure 5N and 5Q or Figure 5O and 5R illustrate induced cells with short flagella, but in both cases, they are PFR positive. In all induced cells, however, the new flagellum is never attached to the old flagellum even though the FC has formed. This suggests that attachment did not occur or was not stable enough to maintain the new-to-old flagellar linkage. If FC attachment did occur, it was lost at an early point in the growth stage of the new flagellum.
To test whether induced cells were capable of orthodox endocytotic activity, we carried out live PF cell endocytosis analysis using the fixable fluorescent lipophylic dye FM4-64X. WT 2K2N cells showed strong endocytotic activity at the base of both flagella, suggesting endocytosis activity via the old and the new FP (Figure 6A and 6B). In induced cells, we observed no endocytotic activity at the site of the new flagellum but considerable activity at the old FP (Figure 6C and 6D). Additionally, induced cells at the site of the new flagellum were negative for markers of early endocytosis such as clathrin or Rab5A (Figure S2), illustrating that in the absence of the FPC, and despite the fact that the new flagellum is still formed, no endocytotic activity is associated with this new flagellum.
In order to identify perturbations in trafficking and pocket targeting systems, we probed induced cells with (1) antiserum to the cysteine-rich acidic transmembrane protein (CRAM) (a FP protein of unknown function but postulated to be a lipoprotein receptor) [42]; (2) antiserum to procyclin, a major surface coat protein (GPEET) expressed in PF cells [43]; and (3) antiserum to p67, a lysosomal protein [44]. In all cases, noninduced cells gave localisation signals similar to control cells in work published previously. However, induced cells gave strong vesicle and/or vacuolar labelling patterns (Figures S3 and S4), and in the case of CRAM, the whole cytoplasm was positive for this protein (Figure S3E–S3H). CRAM localization was also checked by immunoelectron microscopy and confirmed that induced cells rapidly accumulate CRAM-positive vesicular structures (Figure S1E). Western blotting of BILBO1, clathrin, Rab5A, CRAM, and procyclin (GPEET) proteins after BILBO1 RNAi indicated that clathrin and Rab5A levels appear relatively constant, but that CRAM protein levels increased considerably (unpublished data).
Preliminary studies on cultured BSF cells show that there is a significant difference between the phenotype seen in PF cells versus that seen in BSF cells following BILBO1 ablation. In BSF cells, the immediate morphological effect observed was the rapid formation of spherical cells. Cells began rounding up as early as 12 h after RNAi induction. No aspects of the BSF cells were elongated after BILBO1 knockdown. This rounding up of induced cells prevented a clear analysis of kinetoplast and nucleus number or organisation (Figure S5). Furthermore, immunofluorescence labelling with the anti-PFR monoclonal antibody L8C4 showed that induced cells did not have the flagellum-to-cell body detachment phenotype observed in PF cells, and immunofluorescence labelling with the anti-FAZ monoclonal antibody L3B2 showed that in contrast to PF cells, induced cells often possessed two FAZ signals (unpublished data). Together, this indicates that BILBO1 RNAi in BSF cells has very different effects on cytoskeleton function and organisation in comparison to PF cells.
We have identified a novel protein (BILBO1) that is located around the axoneme of T. brucei as it exits the FP. Using a variety of techniques, we have demonstrated that BILBO1 is part of a “horseshoe” or “ring” of a detergent-insoluble cytoskeletal structure known as the flagellar pocket collar (FPC). BILBO1 is the first component of the FPC to be identified and characterised. The FPC is important for the cell because it forms an “adhesion zone” of electron-dense material located between the pellicular, flagellar, and sequestered FP membranes [6,18]. New flagellum growth is supported in the absence of new FP construction in the case of RNAi knockdown of BILBO1, but the deficiency of a new FPC and FP directly or indirectly results in cell death in both insect and bloodstream forms of T. brucei.
In T. brucei, it has been demonstrated that FP selectivity exists to retain certain proteins since, for example, the CRAM protein and transferrin receptors are restricted to the FP, whereas procyclin and VSGs are found on the flagellum, FP, and cell surface [14,45]. The nature of this selectivity of distribution is unknown, but it may be developmentally regulated or associated with interactions between the FP membrane, the pellicular membrane, and the FPC.
Numerous structures that physically link axonemes to the cell body or the cytoskeleton have been identified in lower and higher eukaryotes [46–49]. One interesting example is the “ciliary necklace.” This structure has been identified in all 9+2 and 9+0 mammalian and invertebrate cilia, but it is not universally found in sperm. The necklace is located at the basal plate of cilia where the axonemal membrane “pinches in” [50]. Additionally, numerous proteins, including centrin, have been identified to be associated with basal bodies/centrioles and the cytoskeleton [51]. However, the molecular nature or function(s) of many of these structures remain to be identified [52]. The ciliary necklace-like structure of T. brucei is visible at the transition zone region, between the axoneme and the flagellar membrane [53], but it does not appear to be associated with the FPC [54]. Furthermore, extensive searches to define necklace proteins in any organism and to characterise their function have been fruitless. A more comprehensive search for these proteins should now be possible since centriole, cilia, and flagella proteomes have been published [28,55–58].
Recent evidence illustrates that certain primary cilia can function as sensory organelles that detect changes in fluid flow and initiate gene expression accordingly [52,59,60]. Notably, some of these primary cilia have structures similar to FPs with some electron-dense material at the exit point of the cilium, similar to the organisation of the FPC. The pocket-like structure is called the axonemal “vesicle” or sheath; it is thought to be Golgi derived and extends along with the growing ciliary axoneme within the cytoplasm [61]. The function of this vesicle is in all probability to provide a distinct, isolated compartment separated from the cytoplasm to allow intraflagellar transport for axonemal elongation. However, the molecular functions of the vesicle of primary cilia are not known.
Certain primary cilia can retract if subjected to physiological stress [62]. Membrane that is bound to the proximal region of the primary cilia axoneme is observed when they retract [63]. Presumably, the membrane in these structures is derived from the pellicular membrane, but exactly how this membrane is maintained as a uniform and organised “sack” or “pocket” around the primary cilia remains unresolved. Electron microscopy data suggest that the primary cilia vesicle is also able to carry out endocytotic activity via coated pits [64,65]. The presence of pits implies an organisation of membrane and proteins that separate the plasma membrane from the primary cilia vesicle. In this regard, we propose that a structure additional to the necklace and analogous to the FPC of trypanosomes may exist in primary cilia and may be important for positioning of cilia and, possibly, in trafficking processes.
In mammals and yeasts, actin and actin-binding proteins are the major cytoskeleton components associated with endo- and exocytosis [66,67]. These proteins are essential for the reshaping of the plasma membrane to facilitate endocytosis. They are often found associated with coated pits in the form of transient patches tightly associated with primary endocytotic vesicles. The actin poisons Latrunculin A and Jasplakinolide partially inhibit endocytosis in mammalian cells but initiate a complete endocytotic block in Saccharomyces cerevisiae [66–69]. In trypanosomes, actin has a differential role whereby it is essential and required for the formation and trafficking of endocytotic vesicles in BSF cells of T. brucei. Loss of actin by RNAi in BSF cells prevents endocytosis and results in enlargement of the FP, followed by cell death. In contrast, actin is neither essential nor associated with the FP in procyclic cells [4]. Furthermore, in trypanosomes, actin has not been observed as polymers or bundles, rather it localises to the endocytotic pathway but does not associate with the subpellicular cytoskeleton or the FP, illustrating that it is not a component of the FPC [4].
The FAZ is thought to attach the trypanosome flagellum along the cell body and to coordinate correct cytokinesis [10]; thus, the FAZ plays an important role in the regulation of cell division. In BILBO1 RNAi-induced PF cells, the flagellum-to-cell body attachment of the new flagellum was disrupted and the expected new FAZ was absent, implicating an important relationship between the FAZ and the FPC/FP. The lack of FAZ formation is striking, but consistent with the orientation of the new flagellum being detached from the cell body. Alternatively, the FAZ is absent because the new flagellum (1) rapidly loses a flagellum–cell body attachment or (2) never initiates an attachment to the cell body. In either case, the absence of a new FAZ raises interesting questions regarding the control of FAZ formation and its relationship with other structures of the cytoskeleton.
A unique feature of the trypanosome cell cycle is that defects in cytokinesis do not necessarily trigger mitosis checkpoints, so that cells become multinucleated when cytokinesis is blocked [70]. However, in BILBO1 RNAi-induced cells, only 3.57% were multinucleated, suggesting the stimulation of a true cell-cycle block. This apparent S phase and mitotic block is unlikely to be due completely to the loss of FAZ. Previous studies have shown that interfering with correct FAZ formation, by RNAi knockdown of a FAZ protein called FLA1, induces flagellar detachment and cytokinesis block, but not mitosis, because induced FLA1 RNAi cells develop a multinucleated phenotype [36].
In BILBO1 RNAi-induced PF cells, the presence of a single FAZ may pose difficult cytokinesis-related problems for the cells. Our studies suggest that the FP or FPC plays a more substantial role in the cell cycle than does the FAZ; however, we are unable to define whether it is the FPC or the FP that initiates this cell cycle block. Even though BILBO1 is expressed in both PF and BSF cells, reduction of expression in PF cells induces the formation of many 2K2N cells that arrest in a phenotype in which basal bodies are located on the posterior side of the two nuclei (KKNN) instead of an alternated KNKN conformation. These data raise questions as to whether BILBO1 or the FPC, in interaction with the cytoskeleton, function in controlling cell-shape differentiation.
It appears that the new basal body of BILBO1 RNAi-induced cells can “dock” in the proximity of the old pocket in a fashion similar to that found in WT cells, but it then moves away to the extreme posterior end of the elongated cell, possibly because of failure to assemble the FP and FPC. It does not appear to depart from its position next to the old basal body, or migrate into the cytoplasm to an incorrect position, before extending a measurable length of axoneme. The mispositioning of the new flagellum towards the cell posterior and the absence or early loss of flagellum-to-flagellum attachment supports the supposition that the new flagellum grows into the pellicular membrane and/or remnants of the old FP membrane during or early after new axoneme growth is initiated.
In these induced cells, the PFR grows in parallel with the new flagellum and is independent of attachment to the cell body, thus indicating that the PFR is dependent on the axoneme for initiation and formation rather than signals from the cell body. Our data also show that new flagellar growth of procyclic cells is autonomous of the FC, or an attachment to the old flagellum. Indeed, new flagella of induced cells rapidly lose or may not establish flagellum-to-flagellum or flagellum-to-cell body attachment early in the cell cycle. The work of Davidge et al. (2006) [41] showed that new flagellum formation is not essential for FC movement. They also showed that a new FAZ was formed after intraflagellar transport (IFT) knockdown inhibited new flagellum growth. Why the FC is limited to the FP after BILBO1 RNAi remains to be determined, but it could be argued that absence of FC movement could be related to the absence of FAZ formation. To date, no proteins of the FC have been identified, thus the dependency relationships between the old and new flagella via the FC cannot as yet be studied in more detail.
The extreme posterior localisation of the new flagellum in BILBO1 RNAi-induced cells is intriguing and is observed only in 2K2N cells. This location in the cell cycle is not coincidental; otherwise, one would expect to observe the site of the new flagellum to be distributed randomly on the cell surface and in any cell-cycle stage. One possibility is that the new flagellum may be pushed to the posterior end of the cell by the growth of new or preexisting subpellicular microtubules. This would suggest a loss of control over the polymerisation of the microtubules involved in the posterior extension of the cell during normal division. Why subpellicular microtubules of induced cells elongate to such an extent is also interesting and requires further investigation.
Other workers have observed a posterior-end extension in trypanosomes after expression or RNAi knockdown of proteins related to differentiation or control of the cell cycle. Overexpression in PF cells of TbZFP2, a zinc finger protein implicated in differentiation from BSF to PF cells, causes a “nozzle” phenotype (a posterior extension of the cytoskeleton) as well as the occurrence of multinucleated and multiflagellated cells [71]. The authors showed that the nozzle is a result, at the posterior end of the cell, of polarized extension of microtubules rather than interdigitating short microtubules. RNAi depletion in PF cells of the cyclin CYC2, an essential PHO80-like cyclin, and the cyclin-related kinases CRK1+CRK2 also induced a polarized extension of posterior-end microtubules [72]. In all these studies, the extended or nozzle phenotypes were only observed in 1K1N/2K1N cells (cells arrested in G1), as opposed to the postmitotic 2K2N cell-cycle stage observed in BILBO1 knockdown PF cells. The absence of an elongated posterior end in BSF BILBO1 RNAi cells is similar to the observations of Hammarton et al. (2004) [73] in that RNAi of CYC2 induces a nozzle phenotype in PF cells, but not in BSF cells. The reasons for the production of nozzle or extended posterior-end phenotypes are unclear; however, these results clearly indicate that cytoskeleton elongation is heavily influenced by cell-cycle and/or cell-differentiation checkpoints.
The FPC remains intact and attached to flagella after detergent and salt extraction. The FPC thus most likely consists of a complex of proteins in addition to BILBO1, because BILBO1 itself does not appear to have any obvious membrane-targeting domains, but it does have a large coiled-coil domain consistent with protein–protein interactions. One function of the FPC complex is to physically link the flagellum to the neck of the FP and the cell body. More precisely, this link would produce an intimate bridge between the FP membrane, pellicular membrane, and the flagellum membrane. This bridge complex forms a barrier or an adherens junction-like plaque between the flagellum and the subpellicular cytoskeleton. It is well documented that the trypanosome axoneme exits the FP via the FPC, but little data have been published on the organisation of this structure. It follows that structural homologs are likely to be present in many organisms in order to define and localise the exit site of cilia or axonemes. A schematic diagram of the positioning of the FPC and its role in noninduced or induced cells is shown in Figure 7. This figure also illustrates the distribution and organisation of organelles before and after BILBO1 RNAi knockdown.
BILBO1 RNAi-induced cells arrest and die before all BILBO1 protein is lost from the old FPC. The new flagellum of induced cells does not remain attached to the old flagellum via the FC, even within the old mature FP. Surprisingly, the membrane at the base of the new flagellum of induced cells is not capable of carrying out endocytotic function (as demonstrated using FM4-64FX uptake); however, we should consider the possibility that activity could be below the level of detection using this fluorescence-based assay. Nevertheless, with only one functional FP, BILBO1 RNAi-induced cells appear stressed in the sense that this single FP must function for two FPs; this probably induces an endocytotic imbalance. Loss of the FPC appears to disrupt all components of the endocytotic pathway as observed by electron microscopy or via endocytotic and lysosomal markers. Exocytosis is possibly disrupted also, which raises questions regarding the ability and consequences of these cell types to carry out procyclin, VSG, or invariant surface glycoprotein (ISG) trafficking in insect form or BSF cells.
In summary, FP biogenesis, endocytotic activity, flagellar positioning, and cell division all have strict dependency relationships with the FPC in PF trypanosomes. The discovery of BILBO1 and identification of its partners will facilitate studies on the trafficking of surface proteins involved in parasite survival strategies and on FP biogenesis. The identification of nonparasite-specific BILBO1 partner proteins may help to identify generic axoneme and cilia positioning structures.
Genomic DNA of Trypanosoma brucei TREU927/4 GUTat10.1 [74] was used to amplify by PCR the BILBO1 ORF. T. brucei procyclic cell line EATRO1125-T7T and BSF line 427 90–13 single marker [75,76] were grown and transformed as described in [77]. Transformants were screened by immunofluorescence and cell morphology after tetracycline induction (1 μg · ml−1), and cloned.
Flagellar proteins were prepared as follows: 1 × 1010 T. brucei EATRO 1125 cells were harvested by centrifugation (1,000 × g, 20 °C, 10 min) washed in PBS (pH 7.2), 10 mM EDTA. Cells were lysed in PBS, 2 mM MgCl2, 0.25% NP40, and protease-inhibitors (539134; Calbiochem). Genomic DNA and total RNA were digested with 200 U Benzonase. Cytoskeletons were extracted in 1 M NaCl (final concentration) and incubated for 10 min on ice. Flagella were harvested (30 min, 4 °C, 8,422 × g), washed in PBS, 2 mM MgCl2, 36 U Benzonase, washed in PBS and stored in PBS at −80 °C.
Flagellar proteins (10 mg total) were preseparated in denaturing conditions (5% Ampholines [pH 3–10], 2% CHAPS, 7 M urea, 2 M thiourea, 50 mM DTT) on a Bio-Rad Rotofor. Individual fractions were run on SDS-PAGE gels, and protein bands (60 kDa to 80 kDa) were excised and subjected to liquid chromatography–tandem mass spectrometry (LC/MS/MS) analysis.
The protein sequence of BILBO1 was identified using a signature of ten individual peptide sequences. The corresponding ORF was identified by WU-BLAST on GeneDB database has been deposited at GenBank.
The plasmid p3960SL contains a “sense/antisense” cassette targeting a 600-bp fragment of the T. brucei BILBO1 ORF (nucleotide position 1 to 600) in the pLew100 vector [78] and was constructed as follows. A sense fragment of 600 bp was amplified by PCR (with Taq polymerase) using the primers HindIII-3960 (5′GGTCGCaagcttATGGCGTTTCTCGTACAAGTAGCA3′) and 3960–600-XbaI (5′CTCAACtctagaCACACGGTTACCCTTTACATCGA3′) and cloned into pCR2.1-TOPO (p39–600). A 650-bp antisense fragment was amplified with the primers BamHI-3960 and 3960–650-XbaI (5′GTAGCTtctagaAAGTTGAGATTAAACACAGTGAA3′) and cloned in the pCR2.1-TOPO (p39–650). After digestion of p3960–600 by HindIII-XbaI, and p3960–650 by BamHI-XbaI, the excised fragments were simultaneously cloned into pLew100 between the HindIII and BamHI sites (p3960SL). For RNAi in BSF, a fragment corresponding to the last 514 bp of BILBO1 ORF was amplified by PCR (using the primers BamHI-514-3960 5′tcaggatccCAGAGACGCTGATATCGTGAAA3′ and 3960-HindIII 5′GGTCGCaagcttATGGCGTTTCTCGTACAAGTAGCA3′) cloned between the HindIII and BamHI sites of the p2T7–177 plasmid [79]. To overexpress in vivo the recombinant BILBO1-eGFP protein, the BILBO1 ORF was amplified by PCR with the primers HindIII-3960 and 3960-NoStop-XbaI (5′ATATtctagaATCTCGCGGATAGGACCTC3′) and cloned into the pLew79-GFP1 vector [80] to make p3960-GFP.
We thank the following researchers for antibodies: K. Gull (anti-PFR2, L8C4, anti-FAZ, L3B2, anti-FC, AB1, and anti-basal body, BBA4), M. Field (anti-Rab5A and anti-Clathrin), G. Warren (anti-GRASP), M. Lee (anti-CRAM), J. Bangs (anti-p67), and I. Roditi (anti-GPEET procyclin). To produce anti-BILBO1 antiserum, two peptides (H2N-SFPSRPSISELTRSAE-CONH2 and H2N-GSRSPVSHRSESQQAR-CONH2) were synthesized, conjugated to carrier protein (Eurogentec), and then injected into rabbits to produce polyclonal antiserum. Additionally, recombinant 6 histidine-BILBO1 protein was overexpressed in bacteria, purified in urea on Ni-NTA resin, and then injected into a mouse which was then used to produce a monoclonal antibody. All antisera, including the monoclonal, were tested by immunofluorescence and western blotting.
Whole cells (1 × 107 cells) or cytoskeleton proteins were prepared as described in [77]. Membranes were blocked 1 h in TBS, 0.2% Tween-20, 3% milk (or TBS, 5% milk for K1), incubated overnight at 4 °C with mouse polyclonal anti-BILBO1 antibody diluted in blocking buffer 1:200. After washing, anti-CRAM, anti-Clathrin, anti-RAB5A, or anti-GPEET procyclin (K1), diluted 1:2,000 was used as in [43]. Filters were processed as in [77] or [43]. Western blots were scanned at 300 dpi, and densitometry analysis was done using NIH Image 1.62.
Whole cells were washed in PBS, then spread on poly- l-lysine–coated slides. For cytoskeleton preparations, cells were extracted with 0.25% NP40 in PIPES buffer (100 mM PIPES [pH 6.9], 1 mM MgCl2) for 5 min, and then washed twice in PIPES buffer. Cells or cytoskeletons were fixed in −20 °C methanol or 3.7% paraformaldehyde or 3.7% paraformaldehyde with 0.025% glutaraldehyde. In the latter case, cells were neutralised for 15 min in 200 mM glycine washed in PBS, blocked in 1%–10% bovine serum albumin (BSA), and probed with anti-FC (AB1) 1:5, anti-PFR2 (L8C4) neat, anti-basal bodies (BBA4) 1:20, anti-CRAM 1:250, anti-GRASP 1:300, anti-Clathrin 1:250, anti-RAB5A 1:250, anti-GPEET procyclin (K1) 1:200, anti-p67 1: 400, or anti-BILBO1 rabbit polyclonal 1:500 and anti-BILBO1 monoclonal 1:10. Slides were processed as in [77] using Jackson, Sigma, or Molecular Probes secondary anti-IgG– or anti-IgM–specific secondary antibodies conjugated to FITC, Oregon Green, or Texas Red. Prior to K1 labelling, cells were permeabilised in 0.1% NP40 in PBS for 1 min, and then washed 3 × 10 min in PBS. Prior to Clathrin, RAB5A, and CRAM labelling, cells were permeabilised in 0.1% Triton X-100 for 1 or 10 min and blocked with BSA (1% or 10%, respectively for 10 min to 1 h). Slides were DAPI stained and mounted with Slowfade Lite (Molecular Probes). Images were acquired with Metavue 4.4 software, on a Zeiss Axioplan 2 microscope, using a Roper CCD 1300-Y/S digital camera, and processed with Adobe Photoshop 8. Brightness was reduced on DAPI images when used in merged presentations.
Microscopic analysis of FM4-64FX uptake was carried out by a modification of the assay described by Hall et al., 2005 [81]. A total of 1 × 107 induced (36 h) and noninduced BILBO1 RNAi procyclic cultures were harvested by centrifugation, washed in PBS, and then resuspended in 1 ml of PBS, 0.1 mM adenosine, and 10 mM glucose. FM4-64FX was added (to a final concentration of 2.5 μg/ml), and the cells were incubated for 15 min in the dark with mild agitation (25 rpm) on a rotary shaker. After incubation, the cells were kept on ice to block endocytotic activity. All following solutions and protocols were done at 4 °C. The cells were washed in PBS, deposited on poly-l-lysine–coated slides for 5 min in the dark, and then fixed for 15 min in the dark with 4% paraformaldehyde in PBS, followed by two 5-min washes in PBS. Finally, the slide was DAPI stained, 10 μg/ml, in PBS for 4 min, washed 2 × 5 min in PBS, mounted, and then viewed at room temperature as for immunofluorescence.
A total of 50 ml of mid-log phase WT or 48-h RNAi-induced cells were harvested by centrifugation at 1,000 × g for 15 min. Block preparation and protocol was performed exactly as in [77].
A total of 50 ml of RNAi-induced cells (36 h) at 1 × 107/ml were harvested and resuspended in 25 ml of 4% paraformaldehyde, 0.025% glutaraldehyde in PBS for 2 h. Fixed cells were washed, dehydrated, and embedded in Lowicryl HM20 mono-step (EMS). Sections were cut, neutralised in 100 mM glycine for 10 min, blocked in PBS 1% BSA for 10 min, and probed with anti-CRAM 1:250 in blocking buffer 4 °C overnight. Sections were washed 4 × 10 min in PBS 1% BSA, then probed with 10-nm gold-conjugated protein A or G (Aurion) 1:30 in blocking buffer for 2 h. Sections were washed 4 × 10 min in PBS 1% BSA, then 4 × 10 min PBS, fixed in 1% glutaraldehyde in water for 1 min, stained in 2% uranyl acetate for 15 min, washed 4 × 5 min in water, and then viewed as described in [77]. Cytoskeletons were prepared as above with the following exceptions: extraction was in 25 ml of 1% NP40 in Pipes buffer, washed in Pipes buffer, and resuspended in 25 ml of 4% paraformaldehyde, 0.25% glutaraldehyde in PBS for 30 min. Blocks were prepared as above and probed with mouse anti-BILBO1 polyclonal antiserum 1:50 at 4 °C overnight. Sections were washed, probed with 10-nm gold-conjugated anti-mouse 1:30 and processed as above.
Standard error was used for cell counts and n = 3 (3 different experiments in all cases). Sample size (n) is indicated as total cell number counted for the three experiments.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for BILBO1 is DQ054527, and the GeneDB (http://www.genedb.org/) accession number is Tb11.01.3960.
|
10.1371/journal.ppat.1005983 | Potent and Broad Inhibition of HIV-1 by a Peptide from the gp41 Heptad Repeat-2 Domain Conjugated to the CXCR4 Amino Terminus | HIV-1 entry can be inhibited by soluble peptides from the gp41 heptad repeat-2 (HR2) domain that interfere with formation of the 6-helix bundle during fusion. Inhibition has also been seen when these peptides are conjugated to anchoring molecules and over-expressed on the cell surface. We hypothesized that potent anti-HIV activity could be achieved if a 34 amino acid peptide from HR2 (C34) were brought to the site of virus-cell interactions by conjugation to the amino termini of HIV-1 coreceptors CCR5 or CXCR4. C34-conjugated coreceptors were expressed on the surface of T cell lines and primary CD4 T cells, retained the ability to mediate chemotaxis in response to cognate chemokines, and were highly resistant to HIV-1 utilization for entry. Notably, C34-conjugated CCR5 and CXCR4 each exhibited potent and broad inhibition of HIV-1 isolates from diverse clades irrespective of tropism (i.e., each could inhibit R5, X4 and dual-tropic isolates). This inhibition was highly specific and dependent on positioning of the peptide, as HIV-1 infection was poorly inhibited when C34 was conjugated to the amino terminus of CD4. C34-conjugated coreceptors could also inhibit HIV-1 isolates that were resistant to the soluble HR2 peptide inhibitor, enfuvirtide. When introduced into primary cells, CD4 T cells expressing C34-conjugated coreceptors exhibited physiologic responses to T cell activation while inhibiting diverse HIV-1 isolates, and cells containing C34-conjugated CXCR4 expanded during HIV-1 infection in vitro and in a humanized mouse model. Notably, the C34-conjugated peptide exerted greater HIV-1 inhibition when conjugated to CXCR4 than to CCR5. Thus, antiviral effects of HR2 peptides can be specifically directed to the site of viral entry where they provide potent and broad inhibition of HIV-1. This approach to engineer HIV-1 resistance in functional CD4 T cells may provide a novel cell-based therapeutic for controlling HIV infection in humans.
| HIV-1 infection persists and requires life-long therapy. Approaches to prevent viral replication in the absence of treatment will likely require effective antiviral immune responses, but this goal has been confounded by HIV-1’s ability to target CD4 T cells that coordinate adaptive immunity. We describe a novel approach to confer HIV-resistance to CD4 T cells using peptides from the HIV-1 gp41 heptad repeat-2 (HR2) domain to inhibit infection. By linking a 34 amino acid HR2 peptide to the amino terminus of CCR5 or CXCR4 we were able to use the physiologic trafficking of these coreceptors to deliver the inhibitory peptide to sites of viral fusion where they exerted potent, specific and broad resistance irrespective of viral clade or tropism. This effect was highly dependent on the positioning of the peptide and most effective when conjugated to CXCR4. In vitro and in humanized mice, primary CD4 T cells were protected and expanded following HIV-1 infection. This work represents a proof of concept that T cells can be genetically engineered to resist infection in vivo and provides a rationale to explore this approach as a novel cell-based therapeutic in strategies to augment antiviral immune responses that target viral reservoirs and for long-term control of HIV-1.
| HIV-1 infection persists in the face of suppressive anti-retroviral therapy, and following cessation of treatment, typically rebounds rapidly, generating new rounds of infection [1–4]. Viral persistence results from long-lived reservoirs that include memory CD4 T cells [5–7] and perhaps other cell types [8] that are established early after infection in humans [9] and, in pathogenic models of SIV infection in nonhuman primates [10]. While there is a single example of an individual cured of HIV infection following a stem cell transplant from a donor lacking CCR5 [11], in vitro [12, 13] and in vivo models [14] have strongly suggested that host immune responses will be required to eliminate or control virus in these sites. However, confounding immunologic approaches to control HIV-1 is the tropism of this virus, which targets CD4 T cells that are required to generate cellular and humoral anti-viral immune responses [15, 16].
To protect and/or enhance host immune responses to HIV-1, many approaches have been developed based on engineering primary CD4 T cells to become resistant to HIV-1 infection. Findings from our group and others, have shown that gene therapy for HIV-1 is feasible and capable of generating modified CD4 and CD8 T cells that persist in HIV-infected subjects [17–20], traffic to mucosal compartments where HIV-1 infection is frequently initiated and sustained [21], and are capable of exerting selection pressure on the virus [22]. Tebas and coworkers have recently shown that autologous peripheral CD4 T cells, rendered CCR5-negative through zinc-finger nuclease treatment and expanded ex vivo, could be re-infused safely into HIV-infected subjects where they persisted for months to years and expanded in the context of an interruption in anti-retroviral therapy [21]. In this study there was a striking correlation in frequency of disrupted CCR5 alleles and HIV-1 control after ART was removed, indicating that approaches to increase the number of protected CD4 T cells may lead to more durable control. However, while editing peripheral CD4 T cells to be CCR5-negative is feasible and can confer resistance to R5-tropic viruses, there are logistical concerns in that for maximal effect both alleles must be targeted and because this approach would be ineffective for X4- or dual-tropic HIV-1s [23]. Although complementary methods to ablate both CCR5 and CXCR4 are feasible, this dual mutagenesis occurs with relatively lower efficiency and carries the risk of causing possible functional defects [23, 24].
Many strategies to confer resistance to primary CD4 T cells involve stable expression of a transgene to produce an inhibitory protein or nucleic acid [25]. Building on the theme that soluble peptides from the HIV-1 envelope heptad repeat-2 domain (HR2) can inhibit viral entry by blocking formation of the gp41 6-helix bundle that is required for membrane fusion, inhibitory proteins containing the C-terminal 46 (C46) or 36 (C36) amino acids from HR2 conjugated to a membrane-associated scaffold protein have been shown to broadly inhibit R5- and X4-tropic HIV-1s [26–29]. These membrane anchored constructs exhibited antiviral effects when introduced into primary CD4 T cells [26, 27, 30, 31] and were well tolerated and non-immunogenic in a human trial [32]. However, their antiviral activity is dependent on high levels of expression on the cell surface, which can vary considerably in different cell types, and is further influenced by the design of the anchoring protein and cis-acting regulatory elements in the vector [33, 34]. Work in this area has gone forward in nonhuman primate simian-human immunodeficiency virus (SHIV) models using hematopoetic stem cells transduced with a C46-containing protein where a survival advantage of transduced cells was shown along with a reduction in plasma viremia [35]. However, no therapeutic benefit or clear antiviral effect was observed in a human trial when peripheral CD4 T cells were transduced ex vivo with a similar vector and re-infused into patients, most likely due to insufficient levels of gene-protected T cells [32].
Considering these findings, we reasoned that the potency and effectiveness of cell-based HR2-peptide inhibition could be increased if this peptide were brought to the precise site of viral entry by conjugation to molecules directly involved with HIV-1 entry, rather than to artificial scaffold proteins expressed nonspecifically on the cell surface. We introduced a 34 amino acid peptide from HR2 (C34) onto the amino termini of either CD4 or coreceptors, CCR5 and CXCR4. Strikingly, C34-conjugated coreceptors exhibited potent HIV-1 inhibition, with the greatest effect observed for C34-conjugated CXCR4. Considerably less inhibition was observed when C34 was fused to CD4. C34-coreceptor inhibition was dependent on peptide sequence, occurred irrespective of viral tropism for CCR5 or CXCR4 and on multiple viral clades, and occurred for HIV-1 isolates that were resistant to the soluble HR2 peptide inhibitor, enfuvirtide. Primary CD4 T cells expressing C34-conjugated coreceptors, particularly C34-CXCR4, were resistant to HIV-1 in vitro and in vivo in NOD/SCID IL-2Rγnull (NSG) mice, as seen by expansion of these cells during HIV-1 infection. Collectively, these findings demonstrate that stable expression of C34-containing coreceptors on peripheral CD4 T cells can confer potent and broad resistance to HIV-1 and may provide a novel strategy to augment anti-viral immune responses that complement approaches to target or control HIV-1 reservoirs in infected individuals.
A 34 amino acid peptide from the gp41 HR2 domain, corresponding to amino acids 628–661 in HxB2, was fused directly to the amino terminus of either CCR5 or CXCR4 flanked by an N-terminal alanine and C-terminal leucine, lysine linkers (S1 Fig). When these C34-CCR5 or C34-CXCR4 coreceptors were transiently expressed with human CD4 on Cf2-Luc reporter cells and viral entry assessed using R5-tropic BaL or X4-tropic HxB2 HIV-1, respectively, only background levels of entry were detectable relative to unconjugated coreceptors (Fig 1A). When stably introduced via a lentiviral vector into human CD4+ SupT1 cells, from which endogenous CXCR4 had been ablated, (a line termed, A66 [36]) C34-CCR5 and C34-CXCR4 were expressed on the cell surface, as detected by a monoclonal antibody (mAb) specific for the C34 peptide and with mAbs reactive with the CCR5 or CXCR4 extracellular loops (Fig 1C and 1D). A66 cells bearing C34-conjugagted CCR5 or CXCR4 migrated in a chemotaxis assay in response to CCL4 and CXCL12, respectively, indicating that coreceptors containing this amino terminal peptide continued to exhibit physiologic responses to their cognate chemokines (S2 Fig). Consistent with the results of viral entry assays on Cf2-Luc cells, both C34-CCR5 and C34-CXCR4 were highly resistant to BaL and HxB2 infection (Fig 1B) indicating that the amino terminal HR2 peptide was likely able to access and inhibit HR1 domains of the pre-hairpin intermediate to block 6 helix-bundle formation and viral entry. Similar results were seen with additional R5- (YU2, JRFL) and dual-tropic (R3A) HIV-1 isolates (S3 Fig).
The ability of C34-conjugated coreceptors to interact with HR1 domains was also assessed quantitatively using 5-Helix, a partial mimetic of the gp41 6-helix bundle. 5-Helix is composed of three gp41 HR1 segments and two gp41 HR2 segments, and contains a single HR2 binding site of high specificity and affinity [37–39]. Using an interaction assay employing a rhodamine-conjugated 5-Helix construct, A66 cells stably expressing C34-CCR5 or C34-CXCR4 were shown to have approximately 52,000 and 55,000 molecules per cell, respectively, with KD values less than 15 pM. Negligible binding was seen on A66 cells expressing unconjugated receptors (S4 Fig). Collectively, these findings indicate that C34-conjugated coreceptors could be processed and presented on the cell surface, retained the ability to interact with high affinity to HR1 domains that contribute to 6-helix bundle formation during fusion, and when positioned on the coreceptor amino terminus were able to prevent HIV-1 entry and infection.
We next determined if C34-conjugated coreceptors could inhibit HIV-1 when co-expressed with unconjugated coreceptors, and if so, whether inhibition would occur in a homologous (i.e., C34-CXCR4 inhibiting X4-tropic HIV-1 from using unconjugated CXCR4; C34-CCR5 inhibiting R5-tropic HIV-1 from using unconjugated CCR5) or a heterologous manner (i.e., C34-CCR5 inhibiting X4-tropic HIV-1; C34-CXCR4 inhibiting R5-tropic HIV-1). Using the Cf2-Luc transfection assay, we assessed HIV-1 entry when increasing amounts of C34-conjugated coreceptors were cotransfected with a fixed amount of unconjugated coreceptors. Cf2-Luc cells were transfected with CD4 and unconjugated CCR5 or CXCR4 alone or in combination with varying ratios of C34-conjugated CCR5 or CXCR4, inoculated with R5-tropic (BaL) or X4-tropic (HxB2) HIV-1s, and RLUs quantified as an indicator of entry. As shown (Fig 2A, Top Panels), homologous inhibition by C34-coreceptors was evident for both R5- and X4-tropic HIV-1. Relative to fusion with unconjugated CCR5 or CXCR4, for both BaL and HxB2, respectively, levels of entry were comparable to cells transfected with only a GFP control, and inhibition was seen up to a 1:10 ratio of plasmids encoding C34-conjugated to unconjugated coreceptors. At a ratio of 1:10, subtracting background, BaL inhibition by C34-CCR5 was approximately 85% and HxB2 inhibition by C34-CXCR4 was >95%. Similarly, heterologous inhibition was also seen (Fig 2A, Bottom Panels). At input ratios up to 1:10 of C34-conjugated to unconjugated coreceptors, C34-CXCR4 could inhibit BaL from using CCR5 (85%); while C34-CCR5 could inhibit HxB2 from using CXCR4 (>95%). Both homologous and heterologous inhibition were progressively lost at higher dilutions of C34-conjugated receptors.
To evaluate homologous and heterologous inhibition of a spreading HIV-1 infection by C34-conjugated coreceptors, A66 cells were stably transduced to express CCR5 or CXCR4 alone or with C34-CXCR4 or C34-CCR5. C34-conjugagted coreceptors were clearly detectable, as determined by staining with an anti-C34 antibody (S5 Fig). As expected, BaL could infect A66 cells expressing CCR5 but not CXCR4; HxB2 could infect A66 cells expressing CXCR4 but not CCR5. However, co-expression of either C34-CCR5 or C34-CXCR4 with unconjugated coreceptors potently inhibited infection by either virus (Fig 2B).
Thus, inhibition of HIV-1 by C34-conjugated coreceptors could be mediated in a trans-dominant manner, irrespective of viral tropism, and was highly potent, with inhibition occurring at input ratios of expression plasmids of ≥1 to 10 C34-conjugated to unconjugated receptors. Of note, no inhibition occurred for SIVmac239 in SupT1 cells stably expressing C34-CCR5 (S6 Fig), consistent with the findings that peptides from HR2, including enfuvirtide, are poorly inhibitory for SIVmac [40, 41].
The specificity of C34-conjugated coreceptor inhibition of HIV-1 entry and infection was assessed by creating C34-CXCR4 constructs in which the sequence of the C34 peptide was altered at 4 (C34-S4) or 8 (C34-S8) positions shown previously to be critical for inhibiting 6-helix bundle formation and fusion [42, 43] (Fig 3A, Top Panel). These constructs were then transfected into Cf2-Luc cells with CD4 and evaluated for the ability to support infection by X4- (HxB2) or dual-tropic (R3A) HIV-1s. While C34-CXCR4 inhibited fusion for both viruses, fusion for CXCR4 conjugated to C34-S4 occurred at approximately 50% of unconjugated CXCR4, while CXCR4 conjugated to C34-S8 supported fusion at near wildtype levels (Fig 3A, Bottom Panel). Thus, residues within the C34 peptide required for blocking 6-helix bundle formation are also required for the inhibitory effect of the C34 peptide when conjugated to CXCR4. These results clearly indicate that C34-coreceptor inhibition is highly specific.
To evaluate the role of C34 peptide positioning in HIV-1 inhibition, we next determined whether this peptide would block HIV-1 entry when conjugated to the amino terminus of CD4. C34, followed by a Glu-Phe C-terminal linker, was inserted between amino acids 5 and 6 of the mature CD4 amino terminus (i.e., including the signal peptide and after the CD4 sequence, KKVVL), and this construct (designated C34-CD4) or wildtype CD4 were transiently expressed with CXCR4 or CCR5 in Cf2-Luc cells. Infection was assessed for R5- (BaL), X4- (HxB2), or dual-tropic (R3A) HIV-1s. C34-CD4 expression was verified with an anti-CD4 mAb reactive with the CD4 D1 domain and with an anti-C34 mAb (not shown). Relative to wildtype CD4, C34-conjugated CD4 when co-expressed with CXCR4 permitted fusion of HxB2 and R3A, and when co-expressed with CCR5, permitted fusion of BaL and R3A (Fig 3B). In each case, fusion was 70–100% of the levels observed with unconjugated CD4. Thus, the ability of the C34 peptide to inhibit fusion was highly dependent on its positioning. These findings suggest that on the CD4 amino-terminus the C34 peptide was less able to access the HR1 binding site during 6-helix bundle formation and was therefore more limited in its ability to inhibit fusion, in contrast to when positioned on the more membrane-proximal coreceptor amino-terminus.
Viral resistance to the soluble HR2-derived peptide enfuvirtide has been well documented in vitro and in vivo and typically involves mutations involving gp41 amino acids 26–45 in HR1 [29, 44]. To determine the extent to which mutations that confer resistance to a soluble HR2 peptide could also confer resistance to C34 peptide when conjugated to a coreceptor amino terminus, we introduced 3 mutations (I37K, V38A, and N43D) individually into the HIV-1 R3A HR1 domain. Mutations at each of these positions have been implicated in enfuvirtide resistance in patients undergoing anti-retroviral therapy [45–47]. As shown (Fig 4A), when assayed in a pseudovirus entry assay on transfected Cf2-Luc cells, each mutation conferred concentration-dependent resistance on CCR5 fusion by enfuvirtide (i.e., IC50 for parental R3A was 71 nM, but was ≥1000 nM for I37K, V38A and N43D Envs). However, when entry was assayed on C34-conjugated CCR5 or C34-conjugated CXCR4, all Envs were highly susceptible to inhibition (Fig 4B). As a control, no inhibition by enfuvirtide or C34-coreceptors was seen for a VSV-G Env. Thus, these findings strongly suggest that when anchored to a coreceptor amino terminus and in the context of a membrane-associated intermolecular interaction, an HR2-derived peptide exhibited enhanced potency that extended to viruses resistant to soluble HR2 peptides.
Given the ability of C34-coreceptors to inhibit HIV-1 entry when expressed in Cf2-Luc and T cell lines, we evaluated their effects on primary CD4 T cells. Purified CD4 T cells from healthy donors, stimulated with anti-CD3/CD28 coated beads and maintained in IL2-containing media, were transduced with lentiviral vectors encoding C34-conjugated CCR5 or CXCR4 or, as controls, GFP or C34-conjugated CD4, given its poor ability to inhibit HIV-1 (Fig 3). Uninfected CD4 T cells transduced with C34-CCR5 or C34-CXCR4 maintained expression at levels 85–95% of total T cells for up to 14 days. In contrast, expression of C34-CD4 decreased over time; whereas 92.6% of cells were positive at day 0, only 44.1% of T cells expressed this construct at day 14 (Fig 5A). Additionally, in response to CD3/CD28 stimulation, CD4 T cells transduced with C34-coreceptor proliferated and showed no differences in expression of intracellular cytokines (IL2, TNFα, IFNγ, and CCL4) upon T cell restimulation (S7 Fig), suggesting that this construct does not interfere with T cell activation or function.
Cells were inoculated with R5-, X4- or dual tropic HIV-1s from different clades and infection monitored for 14–17 days by flow cytometry for intracellular p24-Gag. Results for R5-tropic JRFL 5 days after viral inoculation are shown in Fig 5B (Top Panels). For non-transduced, GFP-, or C34-CD4-transduced cells, 57.0%, 45.4% or 20.8% p24-Gag+ cells, respectively, were seen. In contrast, in C34-CCR5 or C34-CXCR4 transduced cells, p24-Gag expression was markedly reduced to 0.7% and 1.2% in C34-CCR5- or C34-CXCR4-transduced cells, respectively. Over time, this inhibition persisted with <1% p24-Gag+ cells at Days 8 and 14 (Fig 5B, Middle and Lower Panels).
The kinetics of JRFL p24-Gag+ expression and inhibition are shown in S8 Fig. While we observed modest protection of CD4 T cells expressing C34-CD4, much more robust resistance to HIV infection was seen in CD4 T cells expressing either C34-CCR5 or C34-CXCR4. The finding that JRFL infection was inhibited by both C34-CCR5 and C34-CXCR4 demonstrated that the trans-dominant homologous and heterologous inhibition of HIV-1 seen on T cell lines also occurred on primary cells. Similar results were seen for other X4-, R5- and dual-tropic HIV-1 isolates (Table 1).
To determine whether C34-coreceptor expressing primary CD4 T cells were selectively enriched during HIV-1 infection, transduced and non-transduced cells were mixed prior to infection at a ratio of 1:3 and the proportion of C34-expressing cells was assessed over time in multiple donors. As shown (Fig 5C), following JRFL infection, C34-coreceptor transduced cells increased over time and by Day 14, C34-CCR5- and C34-CXCR4-expressing cells had increased to 51.5% and 66.1%, respectively. In contrast, C34-CD4-transduced cells decreased to 5.5%, with a similar decrease seen when cells were transduced with GFP (Fig 5C, Lower Panels). An expansion of C34-CCR5- or C34-CXCR4-expressing cells was also seen in cultures inoculated with additional primary R5- (US1), X4-tropic (CMU-02), and dual R5/X4-tropic (SF2) HIV-1 isolates (Fig 6). We observed highly restricted replication of dual tropic HIV-1 R3A in cultures containing C34-coreceptor-transduced CD4 T cells, and the expansion of these cells when diluted with untransduced CD4 T cells are shown in S9A and S9B Fig, respectively. Of note, in assays performed using multiple donors and with different HIV-1 isolates, a greater expansion of cells expressing C34-CXCR4 was seen than for cells expressing C34-CCR5 (S1 Table).
Thus, expression of C34-CCR5 or C34-CXCR4 in primary cells does not impair their ability to proliferate or to produce cytokines following T cell activation, and cells expressing these constructs exhibited resistance to multiple primary HIV-1 isolates irrespective of tropism, and showed selective expansion in the setting of HIV-1 infection with greater expansion seen for C34-CXCR4-transduced cells.
To evaluate the ability of C34-conjugated coreceptors to protect CD4 T cells in vivo, we employed the NSG mouse model that has been used extensively to evaluate the ability of a wide range of antiviral agents to protect T cells from HIV-1 infection in vivo [23, 48–51]. We infused 10 million untransduced, GFP-transduced, or C34-CXCR4 transduced T cells into cohorts of mice. Once T cell engraftment was confirmed and quantified (Fig 7A) mice were challenged with mixture of R5 (US1) and X4-tropic (CMU-02) HIV-1 isolates. After eight days viral load was measured, at which time CD4 T cell levels were comparable (Fig 7B). Mice engrafted with C34-CXCR4 expressing T cells had significantly lower viral loads than mice engrafted with either untransduced or GFP-transduced T cells (Fig 7C). After an additional 20 days, mice were sacrificed and the numbers of human CD4 T cells in spleens was measured (Fig 7D), given that in this model, spleens have been shown to contain the majority of engrafted T cells. Mice engrafted with C34-CXCR4 expressing T cells showed a marked increase in CD4 T cells compared to mice engrafted with untransduced or GFP-transduced T cells. Given that our in vitro data in peripheral blood CD4 T cells indicated that survival of cells expressing C34-CXCR4 was superior to cells expressing C34-CCR5 (S1 Table), we compared survival of C34-CXCR4 and C34-CCR5 expressing cells in a second experiment. After CD4 T cell engraftment and HIV-1 infection, mice were bled at 10-day intervals to assess T cell survival and expansion. Remarkably, CD4 T cells expressing C34-CCR5 survived poorly with levels at days 16 and 21 that were comparable to GFP-transduced cells (Fig 7E). However, in marked contrast, C34-CXCR4-transduced cells persisted throughout the period of HIV-1 infection in both peripheral blood and spleens (Fig 7E, 7F and 7G). Together, these data demonstrate that T cells expressing C34-CXCR4 are highly resistant to HIV-1 infection in vivo and exhibit a survival advantage over C34-CCR5 expressing cells.
The persistence of HIV-1 in reservoirs remains the principal obstacle to curing infected individuals [52–55]. Although a substantial proportion of viruses archived in these compartments are defective and/or unable to replicate in CD4 T cells [56], a genetically diverse array of replication competent viruses persists for the lifetime of the infected host and is poised to replicate upon discontinuation of antiretroviral therapy [57, 58]. While pharmacologic approaches are being explored to reverse HIV-1 latency and drive reservoirs to a more active state that could be vulnerable to antiviral interventions [55, 59, 60], there has been no proof of concept to date that these agents alone can impact the size of the reservoir or its capacity to generate new infectious viruses. Indeed, in vitro studies have strongly suggested that whether the goal is elimination of HIV-1 reservoirs or their long-term control in the absence of antiretroviral therapy, an immunologic response will likely be required [12, 13] and will need to persist and be broad enough to recognize the genetic diversity within the reservoir. This response will also need to be resistant to the immunopathogenic effects of HIV infection on CD4 T cells that provide help to initiate and sustain adaptive immunity.
In this report we show that conjugating a fusion-inhibitory peptide from the gp41 HR2 domain to the amino terminus of HIV-1 coreceptors CCR5 or CXCR4 exerts potent, broad, and specific inhibition of genetically diverse HIV-1 isolates. The conjugated C34 peptide exhibited picomolar binding affinity to soluble 5-Helix, which presents HR1 domains in a trimeric context that likely reflects a structure formed as the gp41 pre-hairpin fusion intermediate transitions to the 6-helix bundle during viral entry [37–39]. C34-conjugated CCR5 and CXCR4 were unable to be used for infection by R5- and X4-tropic viruses, respectively, and exhibited potent inhibition over unconjugated receptors with inhibition at input ratios of expression plasmids as low as 1 C34-conjugated to 10 unconjugated receptors. Strikingly, this trans-dominant inhibition occurred irrespective of viral tropism and across genetically diverse clades. The potency of inhibition was further reflected by the ability of C34-conjugated CCR5 to inhibit viruses that were resistant to the soluble HR2 peptide, enfuvirtide, consistent with more effective binding of the conjugated peptide to gp41 in the setting of a highly localized intermolecular interaction. In vitro and in NOG/SCID mice, CD4 T cells stably expressing C34-conjugated coreceptors appeared to be resistant to HIV-1 infection by their selective outgrowth during HIV-1 infection. Collectively, these findings demonstrate a novel approach to enhance the fusion-inhibiting properties of HR2 peptides and to confer broad and durable protection from HIV-1 infection to CD4 T cells by directly targeting the peptides to the precise site of fusion and viral entry.
How do C34-conjugated coreceptors inhibit HIV-1 irrespective of viral tropism and with such high stoichiometric potency? Several reports have shown that CD4, CCR5 and CXCR4 reside in cholesterol-rich microdomains, termed lipid rafts, on the plasma membrane [61–63]. Although some reports have indicated that in contrast to CCR5, CXCR4 may be only partially present in these domains [64–66], there is general agreement that in the context of HIV-1 gp120 and virion binding to CD4, lipid rafts can serve as sites for the recruitment, concentration and colocalization of CCR5 and CXCR4 to facilitate cooperative interactions with the envelope trimer that are required for entry [64, 66]. Indeed, disrupting coreceptor localization in lipid rafts by cholesterol depletion potently inhibits infection and entry of both R5- and X4-tropic HIV-1s [63, 64]. It is likely that C34-conjugated CCR5 and CXCR4 retain their physiologic trafficking and, as a result, are able to colocalize in lipid rafts with unconjugated receptors. In addition, HIV-1 entry requires highly cooperative interactions with multiple coreceptor molecules [67, 68] and there is evidence for direct interactions between CCR5 and CXCR4 [64] in lipid rafts including their ability to form heterodimers [69, 70]. Collectively these findings suggest that HR2 peptides conjugated to CCR5 or CXCR4 are well positioned to disrupt cooperative interactions between the viral envelope and coreceptors regardless of viral tropism. Lastly, it has been well documented that HR2 peptides target a transient intermediate state (i.e. the pre-hairpin fusion intermediate), and as a consequence, the potency of these inhibitors depends not only on how tightly they bind, but on how rapidly they associate [37–39] [71–73]. Thus, our findings are consistent with the model diagrammed in Fig 8 that C34-conjugated coreceptors exploit physiologic trafficking of chemokine receptors to present HR2 peptides at the site of HIV-1 entry, and that the proximity of the C34-conjugated amino termini on CXCR4 or CCR5 is well positioned to interrupt formation of the 6-helix bundle and to take advantage of a kinetic window required for this structure to be generated.
To inhibit HIV-1 fusion, an HR2 peptide conjugated to a cell surface molecule must be able to be presented to trimeric heptad repeat-1 domains (HR1) on gp41 that are exposed during formation of the pre-hairpin fusion intermediate and prior to its conformational transition to the 6-helix bundle [74, 75]. Membrane-associated structures of CXCR4 [76, 77] and CCR5 [78] have been resolved at the atomic level for CCR5 in association with the small molecule inhibitor maraviroc [78] and for CXCR4 bound to a small molecule or a cyclic peptide antagonist [77] and the viral chemokine vMIP-II [76]. Although the amino termini of chemokine receptors play important roles in chemokine binding, a large number of N-terminal residues were missing in CXCR4 structures bound to vMIP-II and to US28 or CX3CL1 (i.e., 22 and 14 aminio acids, respectively), presumably due to its flexibility. Similarly, nearly the entire N-termini were also not resolved in the small molecule inhibitor-bound structures of CCR5 and CXCR4 (18 and 26 residues, respectively). Flexibility of the CXCR4 and CCR5 amino termini, which is a structural theme for many G-protein coupled receptors, likely plays a role in binding to gp120 by facilitating an interaction with the gp120 bridging sheet that forms following CD4 activation [79–81]. As noted above and diagrammed in Fig 8, this flexibility, as well as the membrane-proximal positioning of the CXCR4 and CCR5 amino termini, could serve to present the C34 peptide to HR1 domains following the insertion of the proximal amino terminal gp41 fusion peptide into the cell membrane. Although structural information on this pre-hairpin fusion intermediate is lacking, modeling strongly suggests that a spatial relationship between the C34 peptide on the coreceptor amino terminus and the trimeric HR1 domain anchored in the cell membrane is favorable for such an interaction to occur (personal communication, Irina Kufareva, UCSD, San Diego, CA). Interestingly, when the C34 peptide was conjugated to the amino terminus of CD4, it exhibited considerably less potency in preventing fusion, even though this construct was competent for binding to CD4-specific mAb and initiating fusion (Fig 3). It is likely that in this context the C34 peptide was poorly positioned on the more extended CD4 molecule and/or that gp120 binding itself prevented the peptide from accessing gp41 HR1 domains. Overall, our findings are consistent with the view that in the context of an intermolecular interaction between the envelope glycoprotein trimer, CD4 and coreceptors, that the positioning of the C34 peptide and its tethering to a flexible domain on the anchoring chemokine receptor was critical for inhibiting fusion.
HIV-1 peptides containing 36 or 46 amino acids from HR2 have been shown to inhibit HIV-1 infection in primary T cells when conjugated to membrane-associated scaffold proteins derived from the nerve growth factor receptor or CD34 [26, 27, 30, 33]. The effect of these constructs has been proposed to result from inhibition of 6-helix bundle formation and viral entry following the insertion of gp41 fusion peptides into the target cell membrane [25, 26, 29]. However, their antiviral activity has been shown to be dependent on high and stable levels of surface expression, which can vary considerably in different cell types [33, 34]. Although in human trials these constructs were well tolerated and non-immunogenic when transduced ex vivo in peripheral CD4 T cells and re-infused into patients, no selective advantage of transduced cells or antiviral effects were observed [32]. In nonhuman primates infected with a simian-human immunodeficiency virus (SHIV) bearing an HIV-1 envelope, stem cells transduced with a C46-containing construct showed a modest survival advantage, although no antiviral effects were seen [35]. While a head to head comparison of these constructs with the C34-conjugated coreceptors described in this report has not been conducted, the stoichiometric relationship of one C34-conjugated coreceptor to several unconjugated coreceptors suggests a highly efficient mechanism of fusion inhibition. Interestingly, in primary CD4 T cells in vitro and, particularly in humanized mice, the survival of CD4 T cells expressing C34-CXCR4 was greater than for C34-CCR5. Although there could be steric and/or structural attributes of the N-termini of these receptors that account for these differences, it is also possible that the level of C34-CXCR4 surface expression was greater over time in primary cells.
To be efficacious in HIV-infected humans, T cells expressing C34-conjugated coreceptors that have been rendered resistant to HIV-1 infection will also need to exhibit physiologic functions that permit their expansion and persistence and the ability to promote effective adaptive immune responses. Peripheral blood CD4 T cells expressing C34-conjugated CCR5 or CXCR4 proliferated and exhibited levels of cytokines when stimulated by CD3/CD28-mediated T cell receptor activation that were identical to untransduced T cells (S7 Fig). C34-conjugated coreceptors expressed on cell lines were also shown to mediate chemotaxis in response to their cognate chemokines, CCL4 for C34-CCR5 and CXCL12 for C34-CXCR4, although C34-CXCR4 appeared to be less efficient than wildtype CXCR4 (S2 Fig). It remains to be determined whether the ability to mediate chemotaxis would be advantageous to primary T cells in vivo or to what extent this function would affect the ability of these cells to provide help for adaptive immune responses. However, the potency of protection from HIV-1 infection conferred by these C34-conjugated coreceptors provides a rich opportunity for future studies to explore the full anti-viral potential of these cells. This question is of particular interest given possible antiviral effects observed during an interruption of anti-retroviral therapy in patients who received autologous CD4 T cells that had been rendered CCR5-negative and HIV-1 resistant (to R5-tropic isolates) by zinc-finger nuclease treatment [21].
In summary, these studies demonstrate a proof of concept that peptides from gp41 HR2 can be delivered to the precise sites of HIV-1 entry when conjugated to chemokine receptors CCR5 or CXCR4, with C34-conjugated CXCR4 conferring particularly potent, broad and durable protection from HIV-1 infection to primary CD4+ T cells in vitro and in humanized mice. The ability to exploit the physiologic trafficking of coreceptors that are essential for HIV-1 entry, and spatial relationships of viral and cellular molecules that interact during fusion provide a novel approach to generate HIV-1 resistance. Whether the creation of HIV-1 resistant CD4 T cells can generate stable and long lasting antiviral immune responses in infected patients remains to be determined; however, the feasibility of safely administering gene-modified peripheral T cells expanded ex vivo, has been well shown in patients with hematologic malignancies [82, 83] and in HIV-1 infection [21, 22, 84, 85].
All humanized mouse experiments were approved by the University of Pennsylvania’s Institutional Animal Care and Use Committee (Protocol 802717) and were carried out in accordance with recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health.
The HIV gp41 HR2 domain (C34) peptide sequence was conjugated to the N-terminus of CCR5, CXCR4, or CD4 with a two amino acids spacer. To generate the pVAX-C34-CXCR4 expressing construct, two paired oligonucleotide (oligo) sequences (C34-left-F1/R1 and C34-right-F1/R1, refer to S2 Table for oligo sequences) were allowed to anneal to each other and then used as inserts to ligate into a NheI and AflII-digested pVAX-X4b construct, which is a pVAX plasmid (Life Technologies, Carlsbad, CA) with the CXCR4 isoform B cDNA cloned in. To generate the scrambled C34 with 4 mutated amino acids (SC34mut4 LLEQEDKEQENQAEEIISHLLSTFNNELRDFEMW), the two oligo pairs were replaced with SC34mut4-left-F1/R1 and SC34mut4-right-F1/R1. To generate the scramble C34 with 8 mutated amino acids (SC34mut8 LLEQEDKEQENQSEEILSHILSTYNNLERDFEMW), the two oligo pairs were replaced with SC34mut8-left-F1/R1and SC34mut8-right-F1/R1. To generate the pVAX-C34-CCR5 expressing construct, a pair of primers (R5 cDNA_F1/R1) were used to PCR amplify CCR5 cDNA (single exon) from genomic DNA (gDNA) of the Jurkat T cell line. The PCR product was digested with AflII and XhoI and ligated to pVAX-C34-CXCR4 digested with the same restriction enzymes. To generate the pVAX-C34-CD4 construct, the CD4 signal peptide portion was PCR amplified from a CD4 cDNA clone with a pair of primers (CD4sig_F1/R1) and then digested with SpeI and BsaI. The C34 portion was PCR amplified from the pVAX-C34-CXCR4 construct with a pair of primers (C34_F1/R1) and then digested with BsaI and EcoRI. The mature CD4 portion was PCR amplified from the CD4 cDNA clone with a pair of primers (CD4_cDNA_F1/R1) and then digested with EcoRI and Bgl2. The three parts described above were then ligated into pVAX plasmid digested with NheI and BglII.
To generate C34 constructs for lenti production, we first introduced NheI and XbaI restriction sites to flank the GFP ORF portion of the pCCLSIN.cPPT.hPGK.EGFP.wPRE construct [86] using a QuickChange Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA) to facilitate subsequent cloning steps. The resultant pCCLSIN.cPPT.hPGK.EGFP.wPRE-Nhe1Xba1 construct was digested with NheI and XbaI and ligated with inserts derived from C34-coreceptor constructs (pVAX-C34-CXCR4 or pVAX-C34-CCR5) digested with the same restriction enzymes to generate pCCLSIN.cPPT.hPGK.C34X4.wPRE or pCCLSIN.cPPT.hPGK.C34R5.wPRE. To generate pCCLSIN.cPPT.hPGK.C34CD4.wPRE, pVAX-C34-CD4 was digested with AseI and XbaI, blunted with DNA Polymerase I Klenow fragment, and then ligated to the digested and blunted pCCLSIN.cPPT.hPGK.EGFP.wPRE vector to replace the EGFP portion. The pTRPE lentivirus vector is previously described [50], and contains the EF1α promoter and cloning sites at 5’ (Nhe I site) and 3’ (Sal I site) ends. Both the C34-R5 and C34-X4 fragments were subcloned into pTRPE using 5’ Nhe1 and 3’ Sal1 sites.
The pCCLSIN or pTRPE constructs described above were used to produce lentivirus, pseudotyped with the VSV-G envelope, by transient co-transfection of four plasmids in 293T cells as described [87]. Expression of the transgene is driven by either the human phosphoglycerate kinase (PGK) promoter in pCCLSIN vector or EF1α in pTRPE vector.
The 293T cells (Invitrogen/Thermo Fisher Scientific, Carlsbad, CA) for producing virus and the CF2-Luc cells (kindly gifted by Dr. Dana H. Gabuzda, Dana-Farber Cancer Institute, Boston, MA) for the entry assays were grown in Dulbecco's modified Eagle medium (DMEM) (high glucose) supplemented with 10% fetal bovine serum (FBS), 2 mM glutamine, and 2 mM penicillin/streptomycin. The human T-cell line SupT1 were obtained from the American Type Culture Collection (ATCC, Manassas, VA) and grown in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 10% FBS, 2 mM glutamine, and 2 mM penicillin/streptomycin. The SupT1 derived A66 CXCR4 negative cell line, in which the endogenous CXCR4 alleles were both disrupted (CXCR4-/-), was created as previous described [36]. To generate A66-derived stable cell lines, A66 cells were transduced with lentivirus to express either wild-type (pTRPE constructs) or C34 fused receptors (pCCLSIN.cPPT.hPGK.wPRE or pTRPE constructs). Receptor-expressing cells were then enriched/selected by either cell sorting or single cell cloning (limited dilution and expansion).
The lentiviral vector pTRPE with the EF1α promoter has cloning sites at 5’ (NheI site) and 3’ (SalI site). NheI and SalI digested C34-CCR5 and C34-CXCR4 fragments from pCCL.SIN.cPPT.hPGK.WPRE were subcloned into these sites to make the pTRPE constructs. These lentiviral vectors were then transfected in 293T cells in six-well plates using 3 μg vector, 1 μg gag, 1 μg pol, and 0.5 μg VSV-Env plasmids with 20 μl Lipofectamine 2000 (Invitrogen) for 1 h. After incubation at 37°C for 24 h. the cell-free supernatant was added to A66 cells and spun in six-well plates at 1,500 g for 1 h. This procedure was repeated a second time at 48 h post-transfection. Individual cells were then isolated through limiting dilution and when the cultures had expanded single clones were chosen based on their expression profile.
To determine which coreceptors were used by HIV-1 to enter cells CF2-Luc cells in six well plates were transfected for 1hr at 37oC with 2 μg per well of pcDNA3.1 containing CD4 and 6 μg per well of pcDNA3.1 containing coreceptors (C34-conjugated or unconjugated) and 25 μl Lipofectamine 2000 (Invitrogen). For experiments in which the ratio of C34-conjugated to unconjugated coreceptors was varied, the total amount of transfected DNA was maintained at 6 μg per well with parental pcDNA3.1 plasmid added, as needed, to maintain the same total input of DNA (e.g., for a ratio of 1.0 of C34-conjugated to unconjugated coreceptor, 3 μg of each coreceptor-containing plasmid were transduced; for a 0.1 ratio, 0.3 μg of C34-conjugated coreceptor, 3.0 μg of unconjugated coreceptor, and 2.7 μg of pcDNA3.1 plasmid were transfected). The transfected cells were incubated over night at 37°C and then plated in duplicate into 24-well plates with each coreceptor in duplicate. The cells were incubated overnight at 37°C and then equal amounts of virus were added to each well. Cells were then incubated for 48 h at 37°C. The level of virus entry was determined by lysing cells with 200 μl of a 0.5% Triton X-100/PBS solution of which 100 μl was then mixed with an equal amount of luciferase substrate (Promega). Luciferase activity was quantified on a Thermo-Labsystems Luminoskan Ascent luminometer.
Cells were aliquoted equally into 13 mm tubes and washed in PBS with 2% FBS. Pelleted cells were then resuspend and stained with particular antibodies on ice for 30 min. CCR5 staining was done with conjugated anti-CCR5 monoclonal antibody 2D7-FITC [fluorescein isothiocyanate] (BD Pharmingen). The anti-CD4 staining was done with mAb #19 [88], and CXCR4 staining was done with mAb 12G5 [88] followed by secondary staining with a FITC-conjugated goat anti-mouse antibody (1:40 dilution; Invitrogen). For the in vitro and in vivo primary CD4 T cell staining the anti-CD4 was labeled with BV421, clone OKT4, (BioLegend, San Diego, CA) and the anti-p24 Gag antibody, clone KC57, was labeled with RD1 (Beckman Coulter, Brea, CA). C34 staining was done with an anti-C34 mAb, generated at Green Mountain Antibodies (Burlington, VT) by immunizing Balb/c mice with the C34 peptide synthesized at ELIM Biopharmaceuticals (Hayward, CA). Fluorescence-activated cell sorter analysis was performed either on a Becton Dickinson FACS Calibur flow cytometer or a BD LSRII.
HIV-1 R3A Env was mutagenized using a QuickChange Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA). Oligonucleotides primer pairs (refer to S1 Table for oligonucleotide sequences) used for the I37A change were I37A_F1/R1; for the V38A were V38A_F1/R1; and for the N43D were N43D_F1/R1. Enfuvirtide was provided by the NIH AIDS Reagent Program Repository and resistance determined with pNL4-3 luc pseudotype virions bearing mutated Envs compared to non-mutated Envs.
Primary CD4 T cells were enriched by negative selection (UPenn CFAR Immunology Core) and then culture in RPMI supplemented with 300 U rIL-2 / ml and 2mM pen/strep. The cells were stimulated with CD3/CD28 Dynabeads (Invitrogen) for 5 days, de-beaded, and then rested for 3 additional days prior to either in vitro HIV-1 challenge or infusion into NSG mice for the in vivo challenge.
Recombinant NL4-3 virus with the R3A, BaL, and HxB Envs inserted were transfected into 293T cells for 4 h using the standard calcium-phosphate (cal-phos) method. At 48 h post-transfection virus was collected and stored at –80°C. Virus concentrations were quantified via enzyme-linked immunosorbent assay (ELISA) for the viral p24 antigen (Perkin-Elmer). The US1, CMU 02, SF2, JRFL, and MN virus were amplified in PBMC and equal amounts of viral stocks were used in all challenges.
Equivalent amounts of virus were added to all cell lines and at 18 hours post infection excess virus was removed by washing the cells in fresh RPMI 1640 medium supplemented with 10% FBS. Replication was monitored by measuring the viral reverse transcriptase (RT) activity in culture supernatants which were collected at the indicated dates, centrifuged at 45,000 g for 30 min, 4C. The supernatant was aspirated and 100 μl of RT buffer was added to each tube before the samples were stored at -20C until the completion of the experiment.
CD4+ T cells were cultured for 12 days after removal of anti-CD3/CD28 coated beads, and then restimulated with PMA/Ionomycin (1.5 ug/mL PMA, 1.0 ug/mL ionomycin; Sigma-Aldrich) or with fresh anti-CD3/CD28 beads (3:1 bead to cell ratio). For PMA/Ionomycin, a 1/1000 dilution of brefeldin-A (BFA, GolgiPlug, Becton Dickinson, cat #555029) was added at the same time, and cells were pulsed for 3 hours before being washed in FACS buffer and fixed in Caltag Fix and Perm buffer A for 15 minutes. Cells restimulated with anti-CD3/CD28 beads were incubated with beads alone for 1 hour, and then with BFA and beads for 4 hours, prior to washing and fixing in buffer A. Cells were washed again in FACS buffer after fixation, and then stained for intracellular cytokines in Caltag Fix and Perm buffer B for 15 minutes. After intracellular staining, cells were washed again in FACS buffer and resuspended in 1X PBS with 2% paraformaldyde, prior to analysis by flow cyotometry.
Transendothelial migration of A66-derived CD4+ T cell lines was assessed using a method described previously[89] with some modifications. Briefly, an endothelial cell line, EA.hy926 (ATCC, Manassas, VA), was pre-cultured in 24-well transwell inserts (Coaster) with a 5-μm pore size for 2–4 days in Dulbecco's Modified Eagle's Medium (DMEM) with 10% FBS. Lower wells were then filled with the migration medium (1:1 of RPMI/DMEM, 0.5% BSA, 20 mM HEPES, pH 7.4) in the presence or absence of the indicated amounts of CCL4 or CXCL12. Transwell inserts were washed and filled with 100 μL of A66-derived T cells. The cells were allowed to migrate through the endothelial cell layer into the lower wells at 37°C for 4 h. The migrated cells in the lower well were then collected and counted using a flow-count technique (Coulter). Responses were calculated as the ratio of migrated cells in comparison to total input cells (% input).
Surface expression of C34-conjugated coreceptors on A66 cells was quantified by flow fluorimetry using rhodamine-labeled 5-Helix, an engineered partial mimetic of the gp41 6-helix bundle that lacks one of three HR2 domains and, consequently, binds with high affinity and specificity to HR2 [37–39]. A 5-Helix variant containing a C-terminal Cys residue was recombinantly expressed, purified, and conjugated to rhodamine-5-maleimide (Anaspec), as previously [37–39]. Specific binding activity of 5-Helix-rhodamine was assessed through stoichiometric titrations using HR2-peptide C37 (KD = 0.65 pM). Briefly, 5-Helix-rhodamine (estimated concentration of 1 nM) was incubated with varying concentrations of C37 (5 pM—10 nM) for 2 hours at room temperature in Tris-buffered saline containing 100 μg/ml bovine serum albumin. Each solution was individually loaded through the flow cell of a KinExA 3000 flow fluorimeter (Sapidyne Instruments). The flow cell contained azlactone-activated polyacrylamide beads (ThermoFisher) covalently conjugated to C37 peptide. The beads captured free (unbound) 5-Helix-rhodamine, resulting in a change in bead fluorescence (Δf) that was directly proportional to the free 5-Helix concentration in solution. The C37-dependence to Δf was fit using a general bimolecular binding model where the real concentration of 5-Helix-rhodamine was assumed to be unknown (Origin Software, OriginLabs):
Δf=Δfmin+Δfmax−Δfmin2[5H]0([5H]0−[C37]0−KD+([5H]0+[C37]0+KD)2−4[5H]0[C37]0)
Here, Δfmin is the minimal fluorescence signal obtained at high C37 concentrations where all 5-Helix-rhodamine is bound; Δfmax is the fluorescence signal obtained in the absence of C37; [5H]0 and [C37]0 are the total concentrations of 5-Helix-rhodamine and C37 used in each incubation; and KD is the equilibrium dissociation constant. The specific binding activity of 5-Helix-rhodamine determined from these experiments (0.61 ± 0.06 nM, compared to the 1 nM estimated concentration) was utilized as the basis for 5-Helix-rhodamine dilutions in cell binding experiments.
To determine expression levels of C34-conjugated coreceptors, A66 cells were first washed extensively in complete RPMI media to remove cellular debris before being visualized with Trypan Blue staining to ensure that the proportion of intact, live cells per sample exceeded 90%. Cells were counted manually with a hemacytometer and subsequently aliquoted (500 μl) at a concentration of 4-8x106 per mL. For titration experiments, C34-conjugated coreceptor-expressing A66 cells were serially diluted (1:2) into an equal number of matching coreceptor-expressing A66 cells so that the total cell concentration for each sample was constant. An equal volume (500 μl) of cell-free media containing 100 pM or 200 pM 5-Helix-rhodamine was added to each cell aliquot and the samples were incubated at room temperature for 2–3 hours. Cells were pelleted by slow speed centrifugation, and supernatants were removed and re-centrifuged at high speeds to remove any additional insoluble material. The amount of 5-Helix-rhodamine remaining in clarified supernatants was quantified by flow fluorimetry as described above. For titration experiments, the dependence of Δf on C34-coreceptor-expressing A66 cell concentration was fit to the bimolecular binding equation above with [5H]0 fixed at either 50 or 100 pM and [C37]0 replaced with the molar concentration of C34-coreceptors:
[C37]0→[C34−CoR]0=1000ncNA
Here, n is the number of C34-coreceptors per A66 cell (unknown), c is the cell concentration in cells per milliliter, and NA is Avagadro’s number. Δfmin was determined from 5-Helix-rhodamine/A66 cell incubations that also included 100 nM C37, while Δfmax was obtained from incubations that lacked C34-coreceptor-experessing cells.
NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were engrafted with 107 untransduced or engineered primary human CD4 T cells, and monitored for engraftment of CD45+CD4+ cells after 3 weeks by TruCount (Becton Dickinson) analysis using 50 μL of blood from each animal. Mice were then normalized based on engraftment into groups of 8–10 animals each per experimental group, and infected with 50 μL each of cell-free US1 and CMU-02 intravenously (50 ng p24 total per animal). Mice were subsequently bled every 10 days prior to necropsy at day 21 or day 28 post-infection, at which point spleens were harvested, processed in 6 mL complete RPMI 1640 (10% FCS) through 70 μm filters, and 50 μL analyzed by TruCount. TruCount analysis was performed using PerCP-Cy5.5 conjugated anti-human CD45 antibody (eBioscience), and BV421-conjugated anti-CD4 antibody (eBioscience). Plasma was saved for viral load analysis when available in sufficient quantities for all animals. Animals were housed at the University of Pennsylvania. Serum viral loads were determined Amplicor HIV-1 Monitor Test (Children's Hospital, Philadelphia), using at least 10 μL of serum per animal.
|
10.1371/journal.pcbi.1007054 | Mathematical modelling reveals unexpected inheritance and variability patterns of cell cycle parameters in mammalian cells | The cell cycle is the fundamental process of cell populations, it is regulated by environmental cues and by intracellular checkpoints. Cell cycle variability in clonal cell population is caused by stochastic processes such as random partitioning of cellular components to progeny cells at division and random interactions among biomolecules in cells. One of the important biological questions is how the dynamics at the cell cycle scale, which is related to family dependencies between the cell and its descendants, affects cell population behavior in the long-run. We address this question using a “mechanistic” model, built based on observations of single cells over several cell generations, and then extrapolated in time. We used cell pedigree observations of NIH 3T3 cells including FUCCI markers, to determine patterns of inheritance of cell-cycle phase durations and single-cell protein dynamics. Based on that information we developed a hybrid mathematical model, involving bifurcating autoregression to describe stochasticity of partitioning and inheritance of cell-cycle-phase times, and an ordinary differential equation system to capture single-cell protein dynamics. Long-term simulations, concordant with in vitro experiments, demonstrated the model reproduced the main features of our data and had homeostatic properties. Moreover, heterogeneity of cell cycle may have important consequences during population development. We discovered an effect similar to genetic drift, amplified by family relationships among cells. In consequence, the progeny of a single cell with a short cell cycle time had a high probability of eventually dominating the population, due to the heritability of cell-cycle phases. Patterns of epigenetic heritability in proliferating cells are important for understanding long-term trends of cell populations which are either required to provide the influx of maturing cells (such as hematopoietic stem cells) or which started proliferating uncontrollably (such as cancer cells).
| All cells in multicellular organisms obey orchestrated sequences of signals to ensure developmental and homeostatic fitness under a variety of external stimuli. However, there also exist self-perpetuating stem-cell populations, the function of which is to provide a steady supply of differentiated progenitors that in turn ensure persistence of organism functions. This “cell production engine” is an important element of biological homeostasis. A similar process, albeit distorted in many respects, plays a major role in cancer development; here the robustness of homeostasis contributes to difficulty in eradication of malignancy. An important role in homeostasis seems to be played by generation of heterogeneity among cell phenotypes, which then can be shaped by selection and other genetic forces. In the present paper, we present a model of a cultured cell population, which factors in relationships among related cells and the dynamics of cell growth and important proteins regulating cell division. We find that the model not only maintains homeostasis, but that it also responds to perturbations in a manner that is similar to that exhibited by the Wright-Fisher model of population genetics. The model-cell population can become dominated by the progeny of the fittest individuals, without invoking advantageous mutations. If confirmed, this may provide an alternative mode of evolution of cell populations.
| The cell cycle is a process leading to cell division. It plays a critical role in tissue growth, development and regeneration of multicellular organisms. It consists of two critical phases: the S phase, in which the cell replicates its DNA, and the M phase where it divides in two progeny cells (mitosis). These phases follow the G1 and G2 phases, respectively. After division, progeny cells usually re-enter the cell cycle and return to the G1 phase [1, 2]. Depending on a variety of factors, they may become quiescent (pass to the dormant G0 phase).
One of the important biological questions is how the dynamics at the cell cycle scale, which is related to family dependencies between the cell and its descendants, affects cell population behavior in the long-run. We address this question using a “mechanistic” model, built based on observations of single cells over several cell generation, and then extrapolated in time. We follow a paradigm recently expressed among others by Sandler et al. [3] and Dolbniak et al. [4] stating that stochastic processes in cells are associated with fluctuations in mRNA [5], protein production and degradation [6, 7], noisy partition of cellular components at division [8], and other cell processes. Variability within a clonal population of cells originates from such stochastic processes, which may be amplified or reduced by deterministic factors [9].
Independently of recent approaches, our work has been inspired by earlier work of Darzynkiewicz et al. [10], who analyzed cycling Chinese hamster ovary (CHO) cells using flow cytometry. They reported variability in G1 phase caused mainly by unequal division of cytoplasmic constituents into progeny cells, and the main conclusion was that the cell-cycle heterogeneity was generated mostly during cytokinesis and to a lesser degree during the G2 phase. These data influenced the mathematical models of Kimmel et al. [11], and Arino and Kimmel [12]. In these models, the heterogeneity has been generated only by unequal division or RNA or cytoplasm, with cell growth and the cell cycle duration being deterministic functions of the birth-size of cell. Models involving cell cycle duration stochasticity followed, with the most recent one being ref. [4]. The latter model is a precursor of the present one, yet with a more limited scope and based only on literature data.
Understanding of the complexity of cell-cycle dynamics and of the specific patterns of cell-cycle progression remains incomplete. Quantitative dynamic imaging combined with mathematical modelling has become an essential approach to understanding such complex dynamics [13]. The recently developed experimental FUCCI (fluorescent ubiquitination-based cell-cycle indicator) reporter system [14, 15] allows continuous imaging of cell-cycle progression in single live cells. In this system, two distinct proteins CDT1 and GEMININ, fused to fluorescent markers, indicate the G0/G1 and S/G2/M phases of the cell cycle, respectively.
FUCCI system has been used to investigate inheritance mechanisms in non-stimulated dividing mammalian cells [3], as well as in reoxygenated [16] and X-ray-irradiated cells [17]. In ref. [3] the authors analyzed the correlation of cell-cycle phase durations between family members. Variability in cell-cycle duration is ubiquitous, and sources of heterogeneity such as extrinsic and intrinsic noise [7] or unequal division [18] have been reported. Division times may also be epigenetically regulated [19].
In the present paper, we analyzed experimental and modelled cell pedigrees to determine patterns of inheritance of cell-cycle phase durations for aggregated G1 and S/G2/M phases, based on dynamic imaging of live NIH 3T3 cells. Based on this, we developed an integrated model involving bifurcation autoregression to describe cell proliferation and cell-cycle phase durations, and an ordinary differential equation (ODE) system to describe single-cell protein dynamics. The idea of bifurcating autoregression is that each line of descent from an ancestral cell follows the autoregression model (descendants inherit certain properties from the ancestor), while the inherited and environmental effects in progeny are correlated. We developed estimates of the parameters of bifurcating autoregression under lognormally distributed noise, given observed cell-cycle phase durations, and fitted single-cell protein trajectories to the ODE model. In this way we found correlations among parameters for single cells. We validated the model, using the cell pedigrees from dynamic imaging data.
Using the validated model, we employed long-term simulations to address the long-term behavior of the population, including homeostasis, memory of initial conditions and heritability. Specifically, we were interested in how regulation mechanisms of the cell cycle may contribute to propagation of new genetic or epigenetic variants in the cell population. This seems important because it has been established that many disease processes in living organisms were caused by the replacement of original cell diversity by clones which either proliferate without control (as in cancer), or dominate tissue-specific stem cells, limiting their resilience and ability to regenerate, as in aging bone marrow (see [20] and references therein). This observation has been explored in a number of deterministic and stochastic models (see review [21]). We summarized our findings using a version of the classical population-genetics Wright-Fisher model, with variable population size; examples and references can be found in [22, 23]. This approach is also related to the branching process paradigm, although our cell proliferation model is not a classical branching process [24].
Remark We employ the following vocabulary convention for family relationships in cell pedigrees. Suppose cell A divides into cells X and Y; then X divides into L and M, while Y divides into N and P. We call the progeny cells of the same parent cell sibling pairs. X and Y, L and M, and N and P, are sibling pairs. Cells whose parents are siblings we call cousin pairs. Thus, L and N, and M and P, are cousin pairs.
The data at our disposal include single-cell observation of NIH3T3 cells using the FUCCI-2A system (see Methods section (Experimental procedure) for details), under two different Fetal Bovine Serum (FBS) concentrations. Cells were grown in constant conditions for 72 hours. During the experiment, films were recorded in randomly selected areas. We collected data from 123 cell lineages, including 890 individual cells from eight recorded films for 15% FBS, and 69 cell lineages including 224 individual cells from five recorded movies for 10% FBS. Based on experimental results we discovered that cell-cycle duration is shorter when higher FBS concentration is used. Faster progression of cell cycle is caused mainly by speeding up of S/G2/M phases progression. Differences between these two experiments are mainly visible (S7E Fig) in the fraction of cells entering dormancy (G0). Also see Supporting Information (S1 Text, Sensitivity of durations of cell-cycle phases to serum stimulation) for a more detailed description of these results. Further analysis was performed for 15% FBS data, since the sample size was significantly higher than in the 10% FBS experiment, and our model is not focused on dormant cells.
Interpretation of results obtained by cell-cycle model may be difficult because it is not known which effects are caused by population growth and which by correlations among family members. To separate these two effects we used a simplified model of cell cycle which does not include correlations among family members. In the model, cell cycle phase duration of each individual was drawn from lognormal distribution and the parameters were estimated based on experimental data. Long-term experiments were designed for different initial numbers of ancestors. Results for four cases introduced in Table 1 are presented in Table 2. Estimated effective population size (K) is almost twice as high as that obtained using the complete model, which includes correlations among family members, and close to harmonic mean value calculated from simulated genealogies.
To verify how the value of effective population size depends on correlations among family members we used the model presented in this paper, which can reproduce behavior of populations with varying correlations among family members. Parameters of the model were estimated, simulations were performed and K values were estimated for all possible cases for three different numbers of ancestor cells (Fig 5).
In the current paper, we related the experimental data from dynamic imaging of proliferating NIH 3T3 cells to a mechanistic view of the cell cycle and dynamics of accumulation and decay of proteins expressed specifically in different cell-cycle phases. The data included recorded observations of cell pedigrees started from single ancestors and then followed for several divisions (72 h). Assuming constant conditions, it seems possible to reconstruct the long-term dynamics of such cells by building a relevant mathematical model, estimating it based on short-term data, and extrapolating the results in silico to longer times.
Importantly, we investigated existence in our simulations of drift-like effects similar to those exhibited by the Wright-Fisher model of population genetics. In brief, our findings support the following paradigm of interplay of regulation mechanisms in the cell cycle and long-term proliferation:
Two important points regarding our approach are (1) selection of the cell system, and (2) generalizability to other cells. The NIH3T3 cells grow in flat-dish cultures, which makes it possible to track progression through the growth and division cycles. Cells grown in suspension pose unsurmountable difficulties in this respect. All publications known to us, which employed growth and division tracking, used flat-dish cultures. The list includes Chinese Hamster Ovary cells [11], NIH 3T3 cells, H1299 non-small cell lung cancer cells [4, 33], and L1210 mouse lymphocytic leukemia cell [3]. Using flat-dish cultures can be considered a limitation. However, it can also be argued that suspension cultures are even less similar to cells in physiological conditions, which always require substrate for proliferation. In addition, mathematical and statistical analyses of the data in papers listed above indicate, despite differences in details, qualitatively similar patterns of cell-cycle regulation.
Many models of cell-cycle progression have been proposed in the literature: age-structured cell population models [34], branching processes [4, 34], transition probability models [35–37] and other novel models [38], many of them based on experimental data [4, 13, 34, 38, 39]. The importance of developing a fully integrated model with different sources of noise and heterogeneity was discussed in ref. [40]. This motivated us to develop heterogeneous population-growth models with protein dynamics included, such as the model in the present paper, and an earlier model in ref. [4].
The important feature of our model is that it reproduces most of the characteristics observed in experimental data, as is evident from Figs 1, 2 and 3. Our model is based in part on bifurcating autoregression [26] applied to cell-cycle phases and on ideas concerning cell-cycle regulation and unequal division that were developed by many authors; specifically, we refer here to the model by Kimmel et al. [11]. In the model, chance and deterministic elements contribute to its ability to accurately fit multiple facets of cell-cycle kinetics in a heterogeneous cell population.
This type of model may be important not only for understanding the kinetics of cell proliferation but also for testing of the individual responses of the cell to stimuli, especially when such a response is cell cycle-dependent. The models should also be useful for predicting the growth rates of populations consisting of subgroups with different properties and/or in which epigenetic effects are strong. We know that tumor growth is the consequence of competition among a few cell populations. It seems that even a small difference in cell characteristics, such as the cell-cycle proliferation time [41], may increase the ability of a cancer to survive chemotherapy and re-enter the cell cycle.
One of the features of our approach is integration of in vitro experiments with statistics and in silico simulations. The function of the latter is to understand the long-term behavior of cell population given the set of rules (i.e. the mathematical model) inferred using statistical tools, based on a limited-time in vitro experiment. Questions that can be answered in this way include the homeostatic properties of the population growth. Specifically, do the rules of cell growth and division lead under constant environmental conditions to stabilization of the distributions of important cell characteristics, such as cell-cycle time and durations of cell-cycle phases, as well as concentrations of cell proteins? A related question concerns the nature of the transients that emerge after a cell with extreme parameters becomes an ancestor of its own population. If this cell has a short cell cycle, will its progeny tend to dominate the population?
Based on simulations, in two extreme cases of initial cell-cycle time 13.6 h and 61.3 h, large differences in the population growth rate have been observed. Within the interval from 0 to 200 h, during which the cell-cycle duration in both populations returned to the equilibrium distributions, the descendants of the cell with the short cell-cycle length have formed a subpopulation n = 40 times larger than the descendants of the cell with the long cell-cycle duration. If these two sub-populations were mixed, the one originating from the ancestor with the shorter cell-cycle length would dominate the other. In bacterial cells, the importance of long-term dynamics of cellular populations was considered in recent studies [42, 43], in which mathematical models were supported by biological experiments using E. coli strains. In the first paper [42], authors discovered that (1) condition-dependent change of mean cell-cycle time is strongly correlated with variability in cell-cycle durations; and (2) increase of the heterogeneity of generation times in a population may be the method to evolve to a higher population growth rate in a constant environment, which is partly parallel to our conclusions. In our case a higher variability of cell cycle times was observed in the population with shorter mean cell-cycle times (10% FBS, mean cell cycle time 21.6 h, MAD = 0.17, CV = 0.25; 15% FBS, mean cell cycle time 20.4 h, MAD = 0.21, CV = 0.30).
In the second paper [43], a method to predict histories of single cells in an exponentially growing population was proposed. Analysis revealed that physiological differences in sister cells have a significant impact on individual cell histories and their contribution to the overall population-growth.
As stated previously, the models of cell proliferation and the models of genetic change in populations have been historically based on two apparently contradictory hypotheses, i.e. the unlimited branching process and a completely constant population size, respectively. By necessity, when it became clear that somatic mutations in proliferating eukaryotic cells are important for growth rates, the population constancy assumption in genetic models has been relaxed. A seminal paper concerns the Wright-Fisher coalescent under exponentially growing population [44], followed by a number of models developed for other growth patterns, such as the model in ref. [45], where linear growth has been considered. Polanski and Kimmel in ref. [22], developed computable expressions for a Wright-Fisher coalescent with arbitrary growth pattern (originally due to [46]). We analyzed the experimental and modelled cell pedigrees to determine patterns of inheritance of cell-cycle phase durations for aggregated G1 and S/G2/M phases, based on dynamic imaging of live NIH 3T3 cells. We developed estimates of the parameters of bifurcating autoregression model with lognormal distributions given observed cell-cycle phase durations, and fitted single-cell protein trajectories to the ODE model and found correlations among parameters between single cells (sib-sib, parent-progeny, and other). Parent-progeny and sib-sib correlations from the experimental data were well-reproduced by our modelling, as demonstrated by comprehensive comparisons. Results showed stronger inheritance of the S/G2/M duration compared to G0/G1. Using the model developed, we simulated its transient and long-term behavior and interpreted it in the terms of population genetics. Long-term simulations demonstrated the model had homeostatic properties. However, progeny of a single cell with a short interdivision time had a high probability of eventually dominating the population, due to heritability of cell-cycle phases. Analysis of model simulations showed that an effect similar to genetic drift was present in the model; however, it was amplified by family relationships among cells. This was manifested by reduction of the effective population size compared to the standard Wright-Fisher model of drift. Such patterns of epigenetic heritability in proliferating cells are important for understanding long-term trends of cell populations which are either required to provide influx of maturing cells (such as hematopoietic stem cells), or which relaxed controls and started proliferating uncontrollably (such as cancer cells).
Specifically, we investigated adherence of our simulations to the Wright-Fisher model. We found that after 300 h, population started from N ancestral cells consists of their descendants in random proportions similar as in the Wright-Fisher model with effective population size K much smaller than the census size (straight count of individuals). This is different from the previous studies fitting cell population drift using a Moran model (which may be considered as a version of Wright-Fisher), in which the effective population size was equal to the census size; cf. [47] and references therein.
In addition, we investigated the dependence of K on the correlations existing in our model. As depicted in Fig 5, K is largest for the case in which both parent-progeny and sib-sib correlations are close to 0. Also in this case, K is almost identical to that obtained from the harmonic mean of population sizes at different times, which is the textbook method for computing the effective population size for expanding populations ([29], Equ. (2.13); also see Methods, Wright-Fisher model and the cell cycle model). For numerical comparisons, see Table 2. These findings illustrate the importance of including the correlations in the model.
The most important conclusion is that in the presence of family relations, the estimated effective population size K is smaller than that obtained by using the harmonic-mean law. Using parent-progeny and sibling correlations estimated from data, we obtain K that is about 45% lower. As a consequence, drift acts in cell populations stronger than under the strict Wright-Fisher model with population growth, which increases the impact of random fluctuations in such populations (see the last section of Results). This seems to be of importance in two contexts, in which continued cell proliferation takes place. One of these is cancer growth, in which an initially small population expands and diversifies by somatic mutations but also epigenetic changes [48–50]. Genetic or epigenetic drift acts at the stage when the tumor is very small, but also in isolated secondary foci in some cancers. A very well-documented study of neutral evolution of this kind has been carried out for hepatocellular carcinoma [51]. The other context is healthy human hematopoiesis, in which a relatively small population (ca. 10,000 cells) of hematopoietic stem cells (HSC) proliferates throughout the lifetime, diversifying into a number of descendant lineages and producing about 109 mature blood cells per day [52]. In the course of infections, the HSC become activated and if the incidents recur, their number and heterogeneity may permanently decrease [20], which also makes the healthy HSC less competitive if a malignant clone arises. Since HSC population is distributed among smaller bone-marrow neighborhoods called niches, drift is likely to act strongly in this population. Reduced K amplifies these effects. This may also be the case in development of other stem-cell types, such as in hippocampal neurogenesis. The role of heterogeneity in this system is becoming an intensive research focus [53]. One of the interesting phenomena is heterogeneity reduction with age, which hypothetically might be due to stem-cell population bottlenecks, which have been demonstrated using the branching process model by Li and co-workers [54].
Replication-defective, self-inactivating retroviral constructs were used for establishing NIH3T3 FUCCI-2A cell line as described in ref. [55].These cells stably express the Cdt1 and Geminin sequences coding for G1 and S/G2/M probes, fused to mKO2 and E2-Crimson fluorescent reporters, respectively. They are separated by a 2A sequence to allow post-translational cleavage and followed by a puromycin-resistance cassette for subsequent selection.
Before recording, cells were seeded at 7–10% confluence (105 cells per well) in a 6-wells plate (Falcon), with white DMEM medium (high glucose) containing 1% Penicillin/Streptomycin, 10 mM HEPES and either 10% or 15% FBS. Cells were left undisturbed for 48 hours. For recording, cells were placed in a Zeiss Axiovert 200M microscope (Zeiss) with a 20X Ph objective. A culture chamber, temperature and CO2 controller (Pecon) were used to ensure constant suitable conditions for long-term recording of the cells. Images were recorded every 15 minutes for 72 hours, using a Coolsnap HQ/Andor Neo sCMOS camera. Cells were briefly illuminated with a FluoArc HBO lamp (Zeiss) at reduced intensity. Epifluorescence signals were recorded as follows: mKO2: 300 ms (filter cube: Ex 534/20 –Di 552 –Em 572/38), E2Crimson: 800 ms (filter cube: Ex 600/37—Di 650—LP 664).
The modelling paradigm we employ is based on the hypothesis that the timing of major events in the cell cycle is heritable in proliferating cells. This timing and its heritability are controlled by an intricate mechanism, which has been partly elucidated [56], but its details require more resolution than we can build into our model. Other processes, such as synthesis of FUCCI proteins, occur within this time. Another driving factor is unequal division of proteins among progeny cells. It has been demonstrated theoretically [12, 57] and confirmed by fitting models to data [4, 11], that models based on similar hypotheses exhibit homeostatic properties. This amounts to regulatory feedbacks acting in the model. However, none of these models has been based on data of such resolution as the present one.
|
10.1371/journal.pntd.0002824 | Reproductive Status of Onchocerca volvulus after Ivermectin Treatment in an Ivermectin-Naïve and a Frequently Treated Population from Cameroon | For two decades, onchocerciasis control has been based on mass treatment with ivermectin (IVM), repeated annually or six-monthly. This drug kills Onchocerca volvulus microfilariae (mf) present in the skin and the eyes (microfilaricidal effect) and prevents for 3–4 months the release of new mf by adult female worms (embryostatic effect). In some Ghanaian communities, the long-term use of IVM was associated with a more rapid than expected skin repopulation by mf after treatment. Here, we assessed whether the embryostatic effect of IVM on O. volvulus has been altered following frequent treatment in Cameroonian patients.
Onchocercal nodules were surgically removed just before (D0) and 80 days (D80) after a standard dose of IVM in two cohorts with different treatment histories: a group who had received repeated doses of IVM over 13 years, and a control group with no history of large-scale treatments. Excised nodules were digested with collagenase to isolate adult worms. Embryograms were prepared with females for the evaluation of their reproductive capacities.
Oocyte production was not affected by IVM. The mean number of intermediate embryos (morulae and coiled mf) decreased similarly in the two groups between D0 and D80. In contrast, an accumulation of stretched mf, either viable or degenerating, was observed at D80. However, it was observed that the increase in number of degenerating mf between D0 and D80 was much lower in the frequently treated group than in the control one (Incidence Rate Ratio: 0.25; 95% CI: 0.10–0.63; p = 0.003), which may indicate a reduced sequestration of mf in the worms from the frequently treated group.
IVM still had an embryostatic effect on O. volvulus, but the effect was reduced in the frequently treated cohort compared with the control population.
| Onchocerciasis, also known as river blindness, is a parasitic disease due to the filarial nematode Onchocerca volvulus. It affects more than 37 million people worldwide, most of them (99%) living in Africa. The control of river blindness is, up to now, based on annual or six-monthly mass treatment with ivermectin. This drug kills O. volvulus microfilariae (mf) present in the skin and the eyes and prevents for 3–4 months the release of new mf by female worms (embryostatic effect). In Ghana, after 10–19 years of repeated treatments, the emergence of adult parasite populations not responding as expected to ivermectin was postulated. In this study, the reproductive status of female worms was compared, just before and 80 days after ivermectin treatment, between frequently treated and ivermectin-naïve cohorts from Cameroon. In both groups, embryogenesis of O. volvulus was not affected by ivermectin. However, the accumulation of microfilariae (mf) in the females uteri expected after ivermectin was less marked in the frequently treated population, suggesting that the temporary sequestration of mf following treatment may have been weakened in this group. After 13 years of repeated annual treatments, the embryostatic effect of ivermectin on O. volvulus still occurs but the present findings, associated with observations of higher rates of skin repopulation by mf in the same individuals, suggest that this effect has been decreased.
| The macrocyclic lactone drug ivermectin (IVM) has a broad spectrum of applications against arthropods and nematodes. In human medicine, one of the major indications for IVM is the treatment of onchocerciasis or river blindness [1]. IVM targets both the microfilariae (mf) and adult stages of Onchocerca volvulus, the filarial nematode causing river blindness. By binding to glutamate-gated chloride (GluCl) channels, IVM may provoke pharyngeal and/or somatic paralysis of nematode parasites [2]–[5]. In Brugia malayi, a filarial nematode closely related to O. volvulus, it has been postulated that IVM may paralyze the muscle associated with the excretory vesicle, leading to a reduction in the release of immunomodulators from the parasite that enable evasion of the host immune system [6]. In synergy with the host immune response, this paralyzing effect possibly leads to the elimination of O. volvulus skin mf. Following a standard therapeutic dose (150 µg/kg of bodyweight), this so-called microfilaricidal effect of IVM leads to a 98% clearance of the skin mf within 2–3 weeks [7]. However, a standard dose of IVM is not adulticidal for O. volvulus even though repeated treatments at short intervals (≤3 months) have a significant effect on the viability of a proportion of adult worms [8]. The effect of IVM on adult male worms is not very well known but multiple doses may reduce their ability to re-inseminate the females [9]. In female worms, the drug prevents, temporarily, the release of mf from the uteri. Apparently, IVM has no effect on the embryogenesis per se, but the newly produced mf accumulate in the uteri and degenerate in situ. This is the so-called embryostatic effect of IVM. This inhibition of release of viable mf for some months is important, together with the initial microfilaricidal effect, for reducing the transmission of the parasite.
Because of the longevity of adult worms, IVM distribution programs need to be sustained for 15–20 years, with a high level of suppression of parasite transmission, if one wants to reach elimination of the parasite in the population [10]–[13]. Unfortunately, suboptimal responses to IVM have been reported from some Ghanaian communities that had been subjected to 10–19 rounds of annual community treatments [14], [15]. In those poorly responding communities, repopulation of the skin by mf after IVM was unexpectedly rapid in a fraction of the population, and the response to IVM by O. volvulus was considered atypical [16]. Indeed, the joint analysis of skin microfilarial dynamics after treatment and of female adult worms' reproductive capacity suggests that in these Ghanaian communities, the strength of the embryostatic effect of IVM has been reduced in some parasites that had been previously exposed repeatedly to this drug.
In a previous paper, we compared the dynamics of O. volvulus skin microfilarial densities after IVM treatment in two cohorts with contrasting exposure to this drug: one which had received repeated treatment for 13 years and one which had no history of large-scale treatments. We observed that the repopulation rate was significantly higher in the frequently treated group than in the controls between 15 and 80 days post-IVM, which suggests that the worms from the frequently treated area had resumed their capacity to release mf earlier [17]. In the present paper, we analyzed the reproductive status of O. volvulus female worms collected and the composition of the different embryonic stages found in utero, in the same cohorts before (D0) and 80 days (D80) after IVM, to assess whether embryo production, development and viability in the females are consistent with our previous findings.
The objective of this study was to assess whether the embryo production and development, and/or the embryostatic effect of IVM on O. volvulus have been altered after several years of drug pressure. To do this, we composed two cohorts of patients and defined the exposure factor as the area of residence (frequently treated area or area naïve to mass IVM administration).
The group from the IVM-naïve area was recruited in 10 neighboring communities of the Nkam valley (Bayon, Ekom-Nkam, Mboue, Mpaka, Mbarembeng, Bakem 1, Bakem 2, Lonze, Manjibo and Mounko), a forested area located in the Littoral Region of Cameroon. These villages were known to be endemic for onchocerciasis but had not benefitted from any mass IVM treatment at the outset of the study. Since the IVM-naïve region was also known to be endemic for loiasis, Loa loa microfilaremia was assessed and those few subjects presenting with more than 30,000 L. loa mf per milliliter of blood were excluded from the study to prevent the occurrence of a Loa-related post-IVM encephalopathy. This IVM-naïve area will be referred, throughout the text, as the control area.
The group of patients subjected to multiple IVM treatments was recruited in 22 communities of the Mbam valley (Babetta, Balamba 1, Balamba 2, Bayomen, Bialanguena, Biamo, Biatsotta, Boalondo, Bombatto, Botatango, Boura 1, Diodaré, Gah-Bapé, Kalong, Kon, Lablé, Lakpang, Ngomo, Ngongol, Nyamanga, Nyamsong and Yébékolo). In these communities, annual large-scale treatments with IVM have been conducted since 1994. In addition, these patients had taken part in a clinical trial conducted between 1994 and 1997 aimed at evaluating the macrofilaricidal potential of IVM [8]. During this clinical trial, eligible patients were randomly allocated to one of the four following IVM treatment groups: 150 µg/kg body weight annually (standard group or group 1); 150 µg/kg three-monthly (group 2); high doses (one dose of 400 µg/kg and then two doses of 800 µg/kg) annually (group 3); and high doses (two doses of 400 µg/kg and then 10 doses of 800 µg/kg) three-monthly (group 4). A “clearing dose” of IVM (150 µg/kg) was given to all volunteers in May 1994 to avoid the possibility of severe reactions developing in any patients subsequently taking their first dose on the high-dose regimen and the first “trial” treatment was given three months later. Thus, over the four-year study period (1994–1997) and depending on their treatment group during the trial, they received 4 to 13 doses of IVM under the direct observation of the investigators.
To date, no vector control has ever been implemented in either study area.
Patients eligible for the present study, either from the control or the frequently treated area, were males aged 25 years and over carrying at least two palpable onchocercal nodules, but otherwise in a good state of health. All eligible subjects, including those from the IVM-naïve area, were questioned about their history of IVM treatment. A small number of patients from the IVM-naïve area declared they had occasionally received the drug during distribution campaigns organized in communities located 10–20 km away, in the West Region where large scale treatments with IVM had been ongoing for more than 10 years. Assuming that the effect of IVM on adult worm reproduction gradually disappears after 9 months [18], all individuals who had taken IVM during the previous 9 months were discarded from the analysis. Consequently, the effect of a single dose of IVM, and not the potential cumulative effect of two doses of IVM given within a short time frame was assessed. A total of 15 individuals from the frequently treated population declared having taken IVM more recently than the previous Community-Directed Treatment with IVM which had taken place about 9 months before the first nodulectomy planned for the present study, and were thus excluded from the analyses.
To assess whether the embryostatic effect of IVM had been reduced in the frequently treated population, the reproductive activity of O. volvulus adult female worms was evaluated in both populations before and 80 days after the administration of IVM.
The study received ethical clearance from the National Ethics Committee of Cameroon and was approved by the Cameroonian Ministry of Public Health. The objectives and schedule of the study were explained to all eligible individuals, and those who agreed to participate signed a consent form and kept a copy of the latter.
The diagnosis and extirpation of onchocercal nodules were performed as previously described [19], with only slight modifications. Briefly, subcutaneous nodules were sought, at the outset of the study, by visual inspection of subjects, then by careful palpation in a closed but well illuminated room. The locations of all palpated nodules were recorded on a body chart. Two of these locations were randomly selected for subsequent surgical removal, the first one just before the administration of IVM, and the second one 80 days after treatment. Nodulectomies were performed under optimal aseptic conditions. All nodules present in the randomly chosen anatomical sites were collected and each was placed individually in a Petri dish containing RPMI-1640 medium (GIBCO, Life Technologies Inc., Burlington, ON, Canada) in which they were cleaned of remaining human tissue. The nodules were then stored in liquid nitrogen until use.
In order to isolate the adult worms contained in the nodules, the latter were digested using the collagenase technique [20]–[22], blinded as to their origin (frequently treated or IVM-naïve area), or date of nodulectomy (pre- or post-treatment). After thawing, each nodule was incubated for 12–19 hours at 35°C or 37°C (time and temperature of digestion depending on the nodule's weight) in five milliliters of the culture medium 199 (GIBCO, Life Technologies Inc., Burlington, ON, Canada) containing type I collagenase (SIGMA, Aldrich Co., Oakville, ON, Canada) at a final concentration of 2.25 mg/ml. Details on the process of the nodule digestion are given as supplementary information (Text S1). The product of digestion (the worm mass and digested human tissues constituting the nodule) was placed in a Petri dish containing 15 ml of medium 199 enriched with Earle's salts (E199), L-glutamine, sodium bicarbonate (GIBCO, Life Technologies Inc., Burlington, ON, Canada), and supplemented with gentamicin sulfate (SIGMA, Aldrich Co., Oakville, ON, Canada) at a final concentration of 2 mg/ml. Individual worms were isolated under a dissecting microscope using entomological and Dumont #5 forceps (Fine Science Tools GmbH, Heidelberg, Germany). Each entire and live worm (dead or calcified and incomplete or broken worms were counted but discarded from the further process) was then spread on a labeled slide and examined under a light microscope (magnification ×40) to confirm the sex of the worm. Entire male worms were individually frozen for subsequent genotyping. In the case of female worms, the head and the tail were localized, and the whole worm examined to determine whether it had been broken during the isolation process. A 15 mm-long section was then removed with a scalpel from the tail end, of each complete and unbroken female, for subsequent genotyping. The rest of the body of the female worm was used to prepare embryograms: it was cut in 1 mm thin slices and crushed in a porcelain mortar containing 1 ml of medium 199. To avoid the shells of embryos breaking during the crushing process, the mortar was placed on a 3 cm thick wet sponge to absorb shocks between the pestle (also in porcelain) and the mortar. Fifteen microliters of the homogenized resulting suspension was then transferred into a 0.2 mm deep Malassez counting chamber and the embryograms were examined under a light microscope (magnification ×100 or ×400).
All embryonic stages were identified and counted according to the following classification: viable stretched mf, degenerating stretched mf, viable coiled mf, degenerating coiled mf, viable morulae and degenerating morulae [21]. The density of oocytes was assessed in a semi-quantitative manner using four categories: absence, rare (less than one oocyte per square of the counting chamber or PSC), few (1–10 oocytes PSC) and numerous (more than 10 oocytes PSC). The suspensions with embryos were examined by two experienced and independent investigators and when any discrepancy was found, the preparation was re-examined by both investigators.
The evaluation of the uterine content was made from 15 µl of the homogenized suspension resulting from the crushing of each female worm; for a matter of simplicity, we shall express the numbers of embryos using this volume (15 µl) as arbitrary unit.
The reproductive status of the female worms was analyzed using one qualitative and three quantitative criteria.
In the control group, 190 individuals underwent nodulectomies before receiving IVM, and 171 (90.0%) of them were present for the second round of nodulectomy, 80 days later. One hundred and eighty eight frequently treated individuals underwent nodulectomies before receiving IVM, and 159 (84.6%) of them took part in the second round of nodulectomy. Overall, the 708 surgical interventions led to the collection of 1110 nodules, of which 1069 were examined and contained 1230 male and 2036 female worms (Table 1). Details on the composition of the nodules for each study site and time of examination are given in Table 1.
At D0, embryograms were performed on 469 and 471 female worms from the control and the frequently treated groups, respectively, and at D80, embryograms were done on 396 and 320 worms from the control and the frequently treated groups, respectively. Thus embryograms were available for 1656 of the 2036 females isolated, the difference consisting of incomplete or broken worms and of dead or calcified worms (Table 1). The distribution of female worms according to the contents of their uteri is summarized in Table 1.
Before IVM, most worms contained oocytes (87.8% in the control group vs 90.0% in the frequently treated). In the frequently treated group, female worms appeared to have a slightly higher oocyte production than in the control group, with higher proportions of females containing oocytes density of 1–10 oocytes PSC and >10 oocytes PSC (Chi-squared: 18.509; 3 degrees of freedom (df); p = 0.0003) (Figure 1). This difference remained at D80 (Chi-squared: 11.544; p = 0.0091). However, in both the control and frequently treated groups, the distribution of worms according to their oocyte production did not change between D0 and D80 (Chi-squared: 2.396; 3 df; p = 0.4943 and Chi-squared: 2.226; 3 df; p = 0.5268, respectively), meaning that oocyte production remained unchanged after the IVM dose given as part of the study.
A reduction in the mean number of viable morulae per female worm was observed between D0 and D80 in the two groups (Table 1, Figure 2a). This reduction was less marked in the frequently treated group (38.4% decrease) than in the control group (67.6% decrease). A very similar pattern was observed for the mean number of viable coiled mf per female worm, with an average decrease of 32.4% in the frequently treated group and 68.7% in the control group (Table 1, Figure 2a).
The mean numbers of embryos per worm are summarized in Figure 2a for each group and for each time of observation. Similar numbers of embryos per worm were observed in the two groups both before IVM (mean (standard deviation, sd): 54.9 (97.0) in the control group vs 54.3 (90.3) in the frequently treated group) and 80 days after IVM (mean (sd): 72.4 (185.7) in the control group vs 72.5 (160.9) in the frequently treated group) (Table 1). Multilevel Poisson regression confirmed a similar evolution in the total number of embryos per worm in the two groups between D0 and D80 (incidence rate ratio, IRR: 0.67; 95% Confidence Interval (95% CI): 0.28–1.61; p = 0.37) (Table S1). The number of males present in the nodule was the only covariate associated with the number of embryos per worm (IRR: 2.10; 95% CI: 1.77–2.49; p = 0.001).
The mean numbers of embryos per productive worm are summarized in Figure 2b for each group and for each time of observation. At D0, 49.3% of the worms from the control group were productive with an average of 102.2 (sd: 116.8) embryos (viable or degenerating) per productive worm (Table 1). In the frequently treated group, we observed similar values, with 45.3% of productive worms and an average of 103.0 (sd: 106.5) embryos (viable or degenerating) per productive worm. At D80, the proportion of productive females decreased slightly in the two groups to reach 43.2% in the control group and to 41.5% in the frequently treated group. On average, at D80, the productive females from the control and frequently treated groups contained 156.5 (sd: 252.5) and 154.1 (sd: 220.4) viable or degenerate embryos per worm, respectively (Table 1). Multilevel logistic regression of the productive status of female worms showed an absence of significant difference between the two groups at each nodulectomy round and that changes in the proportion of productive worms between D0 and D80 were similar in the two groups (OR: 0.97; 95% CI: 0.50–1.26; p = 0.339) (Table S2). The number of male worms in the nodule was the only covariate significantly associated with the productive status of female worms.
At D0, the proportion of female worms with viable stretched mf was significantly higher in the control group than in the frequently treated group (45.2% vs 38.7%, respectively, p = 0.044) (Table 1). At D80, these proportions had slightly decreased and were not anymore significantly different between the two groups (37.4% vs 34.7% in the control and frequently treated group, respectively, p = 0.45). Similarly, before treatment, the number of viable stretched mf per worm (all worms) was slightly higher in the controls than in the frequently treated group (mean (sd): 12.6 (24.9) vs 8.5 (19.7), respectively, p = 0.002). The number of viable stretched mf per worm increased by about 35% in both groups at D80 (mean (sd): 17.1 (80.0) vs 11.4 (44.6) in the control and frequently treated group, respectively, p = 0.127) (Figure 2a). Multilevel Poisson regression did not show a significant difference between the two groups in the evolution of number of viable stretched mf per worm from D0 to D80 (IRR: 1.03; 95% CI: 0.37–2.89; p = 0.949) (Table S3). The numbers of male and of female worms in the nodule were positively and significantly associated with the number of viable stretched mf per worm (p = 0.001 and 0.015, respectively).
At D0, the proportion of females with degenerating stretched mf was significantly higher in the frequently treated than in the control group (48.7% vs 31.6%, respectively, p<0.001) (Table 1). This was associated with a higher number of degenerating stretched mf per worm (all worms) in the frequently treated group (mean (sd): 16.4 (45.6)) than in the control group (mean (sd): 5.4 (22.7)) (Figures 2a and 3). At D80, the proportion of females with degenerating stretched mf was still higher in the frequently treated group (58.1% vs 47.9% in the control group) but the mean number of degenerating stretched mf per worm was the same in the two groups (mean (sd): 42.2 (110.7) vs 42.2 (134.6) in the frequently treated and the control group, respectively) (Figures 2a and 3). However, the multilevel Poisson regression indicated that the increase in number of degenerating stretched mf per worm between D0 and D80 was much lower in the frequently treated group than in controls (IRR: 0.25; 95% CI: 0.10–0.63; p = 0.003) (Table S4). Moreover, it showed that age (IRR: 1.02; 95% CI: 1.00–1.04; p = 0.038) and the number of male worms in the nodules (IRR: 2.08; 95% CI: 1.75–2.48; p = 0.001) were positively associated with the number of degenerating stretched mf.
Oocyte production was unchanged after IVM treatment in both groups. The proportion of productive females was slightly reduced after IVM in both groups but the uteri of those productive females contained about 50% more embryos, all stages considered together, than before treatment. Whereas the numbers of morulae and coiled mf both decreased after IVM, especially in the control group, the number of viable mf increased significantly (by about 35%) in both groups. The number of degenerating mf in the uteri of the worms also increased after IVM in both groups, but this accumulation was more marked in the worms from the control group.
The present study was carried out in a context where many controversies about possible resistance of O. volvulus to IVM still subsist [25]–[28]. As a chapter of a detailed study conducted in Cameroon to address this issue, this investigation aimed at assessing whether the strength of the embryostatic effect of IVM against the parasite has been modified after repeated treatments. To this end, we compared the embryonic populations, before and 80 days after a standard dose of IVM, between worms collected from naïve and frequently treated cohorts of Cameroonians.
In the design of the study, we tried to match the two groups as much as possible, except for the history of drug administration, on all other factors related to the epidemiology of onchocerciasis (age, sex, level of endemicity of river blindness, Simulium species, human activities, individual level of infection). Yet, to account for residual differences between the groups, these individual host factors were included as adjustment covariates in the regression models (either Poisson or logistic) while comparing the effect of IVM on embryonic populations between the two groups.
Embryograms revealed that the worms from the repeatedly treated cohort had a higher oocyte production compared to the naïve worms, suggesting that the former may have a higher capacity of reproduction than the latter. Nonetheless, at D80, the oocyte production was similar to its level at D0 in the two groups. These results confirm that oocyte production is not affected by IVM [29].
Morulae and coiled mf were also found at D80, which confirms that IVM does not interrupt the embryogenesis of O. volvulus [18], [30]. However, despite the unchanged production of oocytes after IVM treatment (Figure 1), we observed a reduction in the mean number of viable morulae and coiled mf per female worm between D0 and D80 (Figure 2a and 2b). Such a reduction has been previously described in O. volvulus [30] and Dirofilaria immitis (dog heartworm) [31]. Maintenance of oocyte production associated with a reduction of morulae and coiled mf suggests that the oocytes were likely not fertilized after treatment, probably due to a lack of female re-insemination [9], [32]. It has been hypothesized that IVM interferes with mate-finding by reducing the number of male worms in the nodules [33]. Migration of male worms away from the nodules might be due to the fact that IVM concentration is higher in the latter than in other human host tissues [34]–[36]. In view of the probable effect of IVM on release of substances from the excretory pore of filariae [6], one could alternatively hypothesize that IVM may block the release of sex pheromones from the female worms which normally attract male worms to the nodule and to mate with the female worms. Investigating the effects of multiple monthly doses of IVM on adult O. volvulus, Duke et al. [37] also provided histological evidences that, after IVM, sperm of male worms can be stuck in the mass of degenerating mf in the anterior parts of the uteri of re-inseminated female worms. This suggests that, despite re-insemination, the sperm would be unable to reach the seminal receptacle of a proportion of female worms.
In the present study, an accumulation of stretched mf (either viable or degenerating) in female worms uteri was observed in both groups after IVM treatment. This indicates that the embryostatic effect of IVM was still operating in the worms from the frequently treated population. However, and this is probably the most interesting finding of our study, we observed a much lower increase in the mean number of degenerating stretched mf between D0 and D80 in the frequently treated cohort compared to the control group. The physiological mechanisms associated with degenerative changes of O. volvulus mf in utero have not been elucidated. In the skin, degeneration of mf results from immunological process induced or facilitated by IVM [38]–[40]. However, in the uteri, mf are not in contact with the host immune cells. As suggested by recent observations on B. malayi, IVM might prevent the release of mature mf by interacting with glutamate-gated chloride channels localized in the uterine wall [5]. A prolonged stay in the uterus may not be suitable to mf survival, especially when they are densely packed and, as an indirect consequence of IVM, sequestrated mf may degenerate quicker than those living in their natural environment, the dermis. The lower increase in the number of degenerating mf in those worms repeatedly exposed to the drug might thus reflect an earlier than expected weakening of the embryostatic effect of IVM, allowing viable mf to move from the uteri. The genetic characterization of the worms collected as part of this study, using genes associated with the mode of action of IVM such as the avr-14 gene coding for GluCl [5], [41], are warranted to confirm possible selection towards resistance. Precisely, correlation between embryogram results and genetic profile of these worms will be particularly informative to assess whether some worms have become less sensitive to IVM, and in which proportion.
A limitation of our study may be related to the observation that, despite matching the two study groups on a number of criteria, a higher mean number of degenerating stretched mf was observed at D0, i.e. about 9 months after the last distribution of IVM in the frequently treated population, in the worms from the latter group as compared to the control group. This might be explained by a cumulative effect of repeated IVM treatments on the uteri wall. This could also be the consequence of a different age structure in the worm population between the two areas. It has been shown that, in areas of the former Onchocerciasis Control Programme in West Africa, a sustained decrease in transmission brings about an ageing of the worm population [42], associated with an increase in the proportion of old female worms harboring degenerating stretched mf [18]. The mean age of the parasites in the frequently treated population is probably higher following the decrease in transmission in this area where large-scale IVM treatments have been ongoing for more than 10 years [43]. However, since we did not score the adult worms for age, we cannot assess the respective roles of previous IVM distributions and of a possible ageing of the worm population on the excess of degenerating mf in the frequently treated group at D0. This being said, we do not think that this difference at D0 may have influenced the effect of the IVM dose given during the study.
As an ancillary result of our analyses, a positive association was observed between the number of female worms in a nodule and the number of viable stretched mf observed in their uteri. This might be explained by a stronger effect of grouped female worms to attract male worms for mating and insemination. In Nippostrongylus brasiliensis and Trichinella spiralis, a strong dosage-dependency to female pheromone was observed in male worms [44]–[47]. This means that the higher the number of female worms, the higher the number of male worms attracted and consequently the higher the chance of mating. The influence of pheromone produced by female worms in the attractiveness of male worms was considered in O. volvulus [29]. In the present study, the number of male worms in a nodule was also positively associated with the number of viable stretched mf observed in the female worms' uteri, indicating that the oocyte fertilization succeeded for a proportion of female worms in those nodules with higher number of female and male worms.
The present study demonstrated that the embryostatic effect of IVM on O. volvulus was still present even after multiple treatments. Nevertheless, this effect appears to weaken earlier after treatment in the frequently treated cohort. The higher repopulation rate of the skin by mf after IVM treatment in the individuals from the frequently treated area is consistent with an earlier recovery of mf productivity of their worms [17]. Genetic selection has been described in worm populations submitted to a high drug pressure, including worms collected from individuals of the frequently treated group of the present study [48]–[50]. The analysis of the genetic profile of the adult worms, mf and infective larvae collected as part of this study would constitute the last piece of the puzzle to complete these investigations.
|
10.1371/journal.pgen.1008370 | Environmental and epigenetic regulation of Rider retrotransposons in tomato | Transposable elements in crop plants are the powerful drivers of phenotypic variation that has been selected during domestication and breeding programs. In tomato, transpositions of the LTR (long terminal repeat) retrotransposon family Rider have contributed to various phenotypes of agronomical interest, such as fruit shape and colour. However, the mechanisms regulating Rider activity are largely unknown. We have developed a bioinformatics pipeline for the functional annotation of retrotransposons containing LTRs and defined all full-length Rider elements in the tomato genome. Subsequently, we showed that accumulation of Rider transcripts and transposition intermediates in the form of extrachromosomal DNA is triggered by drought stress and relies on abscisic acid signalling. We provide evidence that residual activity of Rider is controlled by epigenetic mechanisms involving siRNAs and the RNA-dependent DNA methylation pathway. Finally, we demonstrate the broad distribution of Rider-like elements in other plant species, including crops. Our work identifies Rider as an environment-responsive element and a potential source of genetic and epigenetic variation in plants.
| Transposons are major constituents of plant genomes and represent a powerful source of internal genetic and epigenetic variation. For example, domestication of maize has been facilitated by a dramatic change in plant architecture, the consequence of a transposition event. Insertion of transposons near genes often confers quantitative phenotypic variation linked to changes in transcriptional patterns, as documented for blood oranges and grapes. In tomato, the most widely grown fruit crop and model for fleshy fruit biology, occurrences of several beneficial traits related to fruit shape and plant architecture are due to the activity of the transposon family Rider. While Rider represents a unique endogenous source of genetic and epigenetic variation, mechanisms regulating Rider activity remain unexplored. By achieving experimentally-controlled activation of the Rider family, we shed light on the regulation of these transposons by drought stress, signalling by phytohormones, as well as epigenetic pathways. Furthermore, we reveal the presence of Rider-like elements in other economically important crops such as rapeseed, beetroot and quinoa. This suggests that drought-inducible Rider activation could be further harnessed to generate genetic and epigenetic variation for crop breeding, and highlights the potential of transposon-directed mutagenesis for crop improvement.
| Transposable elements (TEs) replicate and move within host genomes. Based on their mechanisms of transposition, TEs are either DNA transposons that use a cut-and-paste mechanism or retrotransposons that transpose through an RNA intermediate via a copy-and-paste mechanism [1]. TEs make up a significant part of eukaryotic chromosomes and are a major source of genetic instability that, when active, can induce deleterious mutations. Various mechanisms have evolved that protect plant genomes, including the suppression of TE transcription by epigenetic silencing that restricts TE movement and accumulation [2–5].
Chromosomal copies of transcriptionally silenced TEs are typically hypermethylated at cytosine residues and are associated with nucleosomes containing histone H3 di-methylated at lysine 9 (H3K9me2). In addition, they are targeted by 24-nt small interfering RNAs (24-nt siRNAs) that guide RNA-dependent DNA methylation (RdDM), forming a self-reinforcing silencing loop [6–8]. Interference with these mechanisms can result in the activation of transposons. For example, loss of DNA METHYLTRANSFERASE 1 (MET1), the main methyltransferase maintaining methylation of cytosines preceding guanines (CGs), results in the activation of various TE families in Arabidopsis [9–11] and in rice [12]. Mutation of CHROMOMETHYLASE 3 (CMT3), mediating DNA methylation outside CGs, triggers the mobilization of several TE families, including CACTA elements in Arabidopsis [10] and Tos17 and Tos19 in rice [13]. Interference with the activity of the chromatin remodelling factor DECREASE IN DNA METHYLATION 1 (DDM1), as well as various components of the RdDM pathway, leads to the activation of specific subsets of TEs in Arabidopsis. These include DNA elements CACTA and MULE, as well as retrotransposons ATGP3, COPIA13, COPIA21, VANDAL21, EVADÉ and DODGER [14–17]. Similarly, loss of OsDDM1 genes in rice results in the transcriptional activation of TE-derived sequences [18].
In addition to interference with epigenetic silencing, TE activation can also be triggered by environmental stresses. In her pioneering studies, Barbara McClintock denoted TEs as “controlling elements”, thus suggesting that they are activated by genomic stresses and are able to regulate the activities of genes [19, 20]. In the meantime, a plethora of stress-induced TEs have been described, including retrotransposons. For example, the biotic stress-responsive Tnt1 and Tto1 families in tobacco [21,22], the cold-responsive Tcs family in citrus [23], the virus-induced Bs1 retrotransposon in maize [24], the heat-responsive retrotransposons Go-on in rice [25], and ONSEN in Arabidopsis [26,27]. While heat-stress is sufficient to trigger ONSEN transcription and the formation of extrachromosomal DNA (ecDNA), transposition was observed only after the loss of siRNAs, suggesting that the combination of impaired epigenetic control and environmental stress is a prerequisite for ONSEN transposition [28]. Studies have further shown that stress-responsive TEs can affect the expression of surrounding genes, by providing novel regulatory elements and, in some cases, conferring stress-responsiveness [28–30].
The availability of high-quality genomic sequences revealed that LTR (Long Terminal Repeat) retrotransposons make up a significant proportion of plant chromosomes, from approximately 10% in Arabidopsis, 25% in rice, 42% in soybean, and up to 75% in maize [31]. In tomato (Solanum lycopersicum), a model crop plant for research on fruit development, LTR retrotransposons make up about 60% of the genome [32]. Despite the abundance of retrotransposons in the tomato genome, only a limited number of studies have linked TE activities causally to phenotypic alterations. Remarkably, the most striking examples described so far involve the retrotransposon family Rider. For example, fruit shape variation is based on copy number variation of the SUN gene, which underwent Rider-mediated trans-duplication from chromosome 10 to chromosome 7. The new insertion of the SUN gene into chromosome 7 in the variety “Sun1642” results in its overexpression and consequently in the elongated tomato fruits that were subsequently selected by breeders [33,34]. The Rider element generated an additional SUN locus on chromosome 7 that encompassed more than 20 kb of the ancestral SUN locus present on chromosome 10 [33]. This large “hybrid” retroelement landed in the fruit-expressed gene DEFL1, resulting in high and fruit-specific expression of the SUN gene containing the retroelement [34]. The transposition event was estimated to have occurred within the last 200–500 years, suggesting that duplication of the SUN gene occurred after tomato domestication [35].
Jointless pedicel is a further example of a Rider-induced tomato phenotype that has been selected during tomato breeding. This phenotypic alteration reduces fruit dropping and thus facilitates mechanical harvesting. Several independent jointless alleles were identified around 1960 [36–38]. One of them involves a new insertion of Rider into the first intron of the SEPALLATA MADS-Box gene, Solyc12g038510, that provides an alternative transcription start site and results in an early nonsense mutation [39]. Also, the ancestral yellow flesh mutation in tomato is due to Rider-mediated disruption of the PSY1 gene, which encodes a fruit-specific phytoene synthase involved in carotenoid biosynthesis [40,41]. Similarly, the “potato leaf” mutation is due to a Rider insertion in the C locus controlling leaf complexity [42]. Rider retrotransposition is also the cause of the chlorotic tomato mutant fer, identified in the 1960s [43]. This phenotype has been linked to Rider-mediated disruption of the FER gene encoding a bHLH-transcription factor. Rider landed in the first exon of the gene [44,45]. Sequence analysis of the element revealed that the causative copy of Rider is identical to that involved in the SUN gene duplication [45].
The Rider family belongs to the Copia superfamily and is ubiquitous in the tomato genome [34,45]. Based on partial tomato genome sequences, the number of Rider copies was estimated to be approximately 2000 [34]. Previous DNA blots indicated that Rider is also present in wild tomato relatives but is absent from the genomes of potato, tobacco, and coffee, suggesting that amplification of Rider happened after the divergence of potato and tomato approximately 6.2 mya [45,46]. The presence of Rider in unrelated plant species has also been suggested [47]. However, incomplete sub-optimal sampling and the low quality of genomic sequence assemblies has hindered a comprehensive survey of Rider elements within the plant kingdom.
Considering that the Rider family is a major source of phenotypic variation in tomato, it is surprising that its members and their basic activities, as well as their responsiveness and the possible triggers of environmental super-activation, which explain the evolutionary success of this family, remain largely unknown. Contrary to the majority of TEs characterized to date, previous analyses revealed that Rider is constitutively transcribed and produces full-length transcripts in tomato [34], but the stimulatory conditions promoting reverse transcription of Rider transcripts that results in accumulation as extrachromosomal DNA are unknown.
To fill these gaps, we provide here a refined annotation of full-length Rider elements in tomato using the most recent genome release (SL3.0). We reveal environmental conditions facilitating Rider activation and show that Rider transcription is enhanced by dehydration stress mediated by abscisic acid (ABA) signalling, which also triggers accumulation of extrachromosomal DNA. Moreover, we provide evidence that RdDM controls Rider activity through siRNA production and partially through DNA methylation. Finally, we have performed a comprehensive cross-species comparison of full-length Rider elements in 110 plant genomes, including diverse tomato relatives and major crop plants, in order to characterise species-specific Rider features in the plant kingdom. Together, our findings suggest that Rider is a drought stress-induced retrotransposon ubiquitous in diverse plant species that may have contributed to phenotypic variation through the generation of genetic and epigenetic alterations induced by historical drought periods.
We used the most recent SL3.0 tomato genome release for de novo annotation of Rider elements. First, we retrieved full-length, potentially autonomous retrotransposons using our functional annotation pipeline (LTRpred, see Materials and Methods). We detected a set of 5844 potentially intact LTR retrotransposons (S1 Table). Homology search among these elements identified 71 elements that share >85% sequence similarity over the entire element with the reference Rider sequence [45] and thus belong to the Rider family. We then determined the distribution of these Rider elements along the tomato chromosomes (Fig 1A) and also estimated their age based on sequence divergence between 5’ and 3’ LTRs (Fig 1A). We classified these elements into three categories according to their LTR similarity: 80–95%, 95–98% and 98–100% (S1A Fig). While the first category contains relatively old copies (last transposition between 10.5 and 3.5 mya), the 95–98% class represents Rider elements that moved between 3.5 and 1.4 mya, and the 98–100% category includes the youngest Rider copies that transposed within the last 1.4 my (S1A Fig). Out of 71 Rider family members, 14 were found in euchromatic chromosome arms (14/71 or 19.7%) and 57 in heterochromatic regions (80.3%) (Table 1). In accordance with previous observations based on partial genomic sequences [34], young Rider elements of the 98–100% class are more likely to reside in the proximity of genes, with 50% within 2 kb of a gene. This was the case for only 37.5% of old Rider members (85–95% class) (Table 2). Such a distribution is consistent with the preferential presence of young elements within euchromatic chromosome arms (50%, 5/10) compared to old Rider elements (9.4%, 3/32) (Table 2 and S1B Fig). In addition, the phylogenetic distance between individual elements is moderately correlated to the age of each element (Fig 1B) (S2 Table).
To better understand the activation triggers and, thus, the mechanisms involved in the accumulation of Rider elements in the tomato genome, we examined possible environmental stresses and host regulatory mechanisms influencing their activity. Transcription of an LTR retroelement initiates in its 5’ LTR and is regulated by an adjacent promoter region that usually contains cis-regulatory elements (CREs) (reviewed in [48]). Therefore, we aligned the sequence of the Rider promoter region against sequences stored in the PLACE database (www.dna.affrc.go.jp/PLACE/) containing known CREs and identified several dehydration-responsive elements (DREs) and sequence motifs linked to ABA signalling (Fig 2A). First, we tested whether these CREs were present in the LTR promoter sequences of the 71 de novo annotated Rider elements (S3 Table). Comparison of Rider LTRs to a set of gene promoter sequences of the same length revealed significant enrichment of several CREs in Rider LTRs (Fisher’s exact test P<0.001) (S4 Table). It is known, for example, that the CGCG sequence motif at position 89–94 (Fig 2A) is recognized by transcriptional regulators binding calmodulin. These are products of signal-responsive genes activated by various environmental stresses and phytohormones such as ABA [49]. We also detected two MYB recognition sequence motifs (CTGTTG at position 176–181 bp, and CTGTTA at position 204–209 bp) (Fig 2A). MYB recognition sequences are usually enriched in the promoters of genes with transcriptional activation during water stress, elevated salinity, and ABA treatments [50,51]. In addition, an ABA-responsive element-like (ABRE-like) was found at position 332–337 bp in the R region of Rider’s LTR, along with a coupling element (CE3) located at position 357–372 bp (Fig 2A). The co-occurrence of ABRE-like and CE3 has often been found in ABA-responsive genes [52,53].
The simultaneous presence of these five CREs in promoters of Rider elements suggests that Rider transcription may be induced by environmental stresses such as dehydration and salinity that involves ABA mediated signalling. To test whether Rider transcription is stimulated by drought stress, glasshouse-grown tomato plants were subjected to water deprivation and levels of Rider transcripts quantified by RT-qPCR (Fig 2B). When compared to control plants, we observed a 4.4-fold increase in Rider transcript abundance in plants subjected to drought stress. Thus, Rider transcription appears to be stimulated by drought.
To further test this finding, we re-measured levels of Rider transcripts in different experimental setups. In vitro culture conditions with increasing levels of osmotic stress were used to mimic increasing drought severity (Fig 2C). Transcript levels of Rider increased in a dose-dependent fashion with increasing mannitol concentration, corroborating results obtained during direct drought stress in greenhouse conditions. Interestingly, tomato seedlings treated with NaCl also exhibited increased levels of Rider transcripts (Fig 2C).
ABA is a versatile phytohormone involved in plant development and abiotic stress responses, including drought stress [54]. Therefore, we asked whether Rider transcriptional drought-responsiveness is mediated by ABA and whether increased ABA can directly stimulate Rider transcript accumulation. To answer the first question, we exploited tomato mutants defective in ABA biosynthesis. The lines flacca (flc), notabilis (not) and sitiens (sit) have mutations in genes encoding a sulphurylase [55], a 9-cis-epoxy-carotenoid dioxygenase (SlNCED1) [56,57], and an aldehyde oxidase [58], respectively. Both flc and sit are impaired in the conversion of ABA-aldehyde to ABA [55,58], while not is unable to catalyse the cleavage of 9-cis-violaxanthin and/or 9-cis-neoxanthin to xanthoxin, an ABA precursor [57]. Glasshouse-grown flc, not and sit mutants and control wild-type plants were subjected to water deprivation treatment and Rider transcript levels quantified by RT-qPCR (Fig 2D). Rider transcript levels were reduced in flc, not and sit by 43%, 26% and 56%, respectively.
To examine whether ABA stimulates accumulation of Rider transcripts, tomato seedlings were transferred to media supplemented with increasing concentrations of ABA (Fig 2E). The levels of Rider transcripts increased in a dose-dependent manner with increasing ABA concentrations. This suggests that ABA is not only involved in signalling that results in hyper-activation of Rider transcription during drought, but it also directly promotes the accumulation of Rider transcripts. The effectiveness of the treatments was verified by assaying expression of the stress- and ABA-responsive gene SlASR1 (S2A–S2F Fig).
Identification in the U3 region of Rider LTRs of a binding domain for C-repeat binding factors (CBF), which are regulators of the cold-induced transcriptional cascade [52,59], led us to test Rider activation by cold stress. However, Rider transcription was not affected by cold treatment, leaving drought and salinity as the predominant environmental stresses identified so far that stimulate accumulation of Rider transcripts (S2G Fig).
The suppression of transposon-derived transcription by epigenetic mechanisms, which typically include DNA methylation, maintains genome integrity [2,3,5]. We asked whether Rider transcription is also restricted by DNA methylation. Tomato seedlings were grown on media supplemented with 5-azacytidine, an inhibitor of DNA methyltransferases. Rider transcript steady-state levels increased in plants treated with 5-azacytidine compared to controls (Fig 3A). Comparison of Rider transcript accumulation in 5-azacytidine-treated and ABA-treated plants revealed similar levels of transcripts and the levels were similar when the treatments were applied together (P <0.05; Fig 3A).
To further examine the role of DNA methylation in controlling Rider transcription, we took advantage of tomato mutants defective in crucial components of the RdDM pathway, namely SlNRPD1 and SlNRPE1, the major subunits of RNA Pol IV and Pol V, respectively. These mutants exhibit reduced cytosine methylation at CHG and CHH sites (in which H is any base other than G) residing mostly at the chromosome arms, with slnrpd1 showing a dramatic, genome-wide loss of 24-nt siRNAs [60]. To evaluate the role of RdDM in Rider transcript accumulation, we first assessed the consequences of impaired RdDM on siRNA populations at full-length Rider elements. Deficiency in SlNRPD1 resulted in a complete loss of 24-nt siRNAs that target Rider elements (Fig 3B). This loss was accompanied by a dramatic increase (approximately 80-fold) in 21-22-nt siRNAs at Rider loci (Fig 3B). In contrast, the mutation in SlNRPE1 triggered increases in both 21-22-nt and 24-nt siRNAs targeting Rider elements (Fig 3B). We then asked whether altered distribution of these siRNA classes is related to the age of the Rider elements and/or their chromosomal position, and thus local chromatin properties. Compilation of the genomic positions and siRNA data in RdDM mutants didn’t reveal preferential accumulation of 21-22-nt siRNAs (S3A Fig) or 24-nt siRNAs (S3B Fig) over specific Rider classes. Subsequently, we examined whether loss of SlNRPD1 or SlNRPE1 was sufficient to increase levels of Rider transcripts and observed increased accumulation of Rider transcripts in both slnrpd1 and slnrpe1 compared to WT (Fig 3C).
We assessed whether this increase in Rider transcript levels is linked to changes in DNA methylation levels in Rider elements of RdDM mutants. There was no significant change in global DNA methylation in the three sequence contexts in the 71 de novo annotated Rider elements (S3C Fig), despite a tendency for young Rider elements to lose CHH in slnrpd1 and slnrpe1 (S3D Fig). Thus, the RdDM pathway affects the levels of Rider transcripts. Also, features of Rider copies such as age and chromatin location alone cannot predict potential for activation based on DNA methylation levels.
The life cycle of LTR retrotransposons starts with transcription of the element, then the synthesis and maturation of accessory proteins including reverse transcriptase and integrase, reverse transcription, and the production of extrachromosomal linear (ecl) DNA that integrates into a new genomic location [61]. In addition, eclDNA can be a target of DNA repair and can be circularised by a non-homologous end-joining mechanism or homologous recombination between LTRs, resulting in extrachromosomal circular DNA (eccDNA) [62–65]. We searched for eccDNA to evaluate the consequences of increased Rider transcript accumulation due to drought stress or an impaired RdDM pathway on subsequent steps of the transposition cycle. After exonuclease-mediated elimination of linear dsDNA and circular ssDNA, Rider eccDNA was amplified by sequence-specific inverse PCR (Fig 4A). Rider eccDNA was absent in plants grown in control conditions but was detected in plants subjected to drought stress (Fig 4A). Sanger sequencing of the inverse PCR products showed that the amplified eccDNA probably originates from the Rider_08_3 copy, which has 98.2% sequence homology of the 5’ and 3’ LTR sequences (S4A Fig). Residual sequence divergence may be due to genotypic differences between the reference genomic sequence and the genome of our experimental material. Analysis of CREs in the LTR of the eccDNA revealed the presence of all elements identified previously with the exception of a single nucleotide mutation located in the CGCGBOXAT box (S4A Fig).
Examination by quantitative PCR of the accumulation of Rider DNA, which included extrachromosomal and genomic copies, in drought-stressed plants also revealed an increase in Rider copy number due to eccDNA (Fig 4B). Importantly, Rider eccDNA was not detected in sit mutants subjected to drought stress (Fig 4A), suggesting that induced transcription of Rider by drought stress triggers production of extrachromosomal DNA and this response requires ABA biosynthesis.
We also examined the accumulation of Rider eccDNA in plants impaired in RdDM. Interestingly, Rider eccDNA was detected in slnrpd1 and slnrpe1 (Fig 4C) and increase in Rider DNA copy number due to eccDNA accumulation was confirmed by qPCR (Fig 4D). Absence of newly integrated genomic copies has been further validated by genome sequencing. The eccDNA forms differed between the mutants (Fig 4C). Sequencing of Rider eccDNA in slnrpd1 showed a sequence identical to the Rider eccDNA of wild-type plants subjected to drought stress. Thus, the Rider_08_3 copy is probably the main contributor to eccDNA in drought and in slnrpd1. In contrast, eccDNA recovered from slnrpe1 had a shorter LTR (287 bp) and the highest sequence similarity with Rider_07_2 (89.2%) (S4B Fig). Shortening of the LTR in this particular element results in the loss of the upstream MYBCORE as well as the CGCGBOXAT elements (S4B Fig).
We then asked whether DNA methylation and siRNA distribution at these particular Rider copies had changed in the mutants. DNA methylation at CHH sites, but not CG nor CHG, was drastically reduced at Rider_08_3 in slnrpd1 (Fig 4E, S4C–S4E Fig and S5A Fig) together with a complete loss of 24-nt siRNAs at this locus (Fig 4F and S4F Fig) but DNA methylation at Rider_07_2 was not affected, despite the deficiency of SlNRPD1 or SlNRPE1 (Fig 4E, S4C–S4E Fig and S5B Fig). Levels of 21-22-nt siRNAs in both mutants and 24-nt siRNA in slnrpe1 were increased (Fig 4F and S4F and S4G Fig). Altogether, this suggests that RdDM activity on Rider is highly copy-specific and that different components of the RdDM pathway differ in their effects on Rider silencing.
To examine the distribution of Rider retrotransposons in other plant species, we searched for Rider-related sequences across the genomes of further Solanaceae species, including wild tomatoes, potato (Solanum tuberosum), and pepper (Capsicum annuum). We used the Rider reference sequence [45] as the query against genome sequences of Solanum arcanum, S. habrochaites, S. lycopersicum, S. pennellii, S. pimpinellifolium, S. tuberosum, and Capsicum annuum (genome versions are listed in Materials and Methods). Two BLAST searches were performed, one using the entire Rider sequence as the query and the other using only the Rider LTR. Consistent with previous reports, Rider-like elements are present in wild relatives of tomato such as S. arcanum, S. pennellii and S. habrochaites; however, the homology levels and their lengths vary significantly between species (Fig 5A). While S. arcanum and S. habrochaites exhibit high peak densities at 55% and 61% homology, respectively, S. pennellii show a high peak density at 98% over the entire Rider reference sequence (Fig 5A). This suggests that the S. arcanum and S. habrochaites genomes harbour mostly Rider-like elements with relatively low sequence similarity, while S. pennellii retains full-length Rider elements.
To better visualize this situation, we aligned the BLAST hits to the reference Rider copy (Fig 5B). This confirmed that Rider elements in S. pennellii are indeed mostly full-length Rider homologs showing high density of hits throughout their lengths, while BLAST hits in the S. arcanum and S. habrochaites genomes showed only partial matches over the 4867 bp of the reference Rider sequence (Fig 5B). Unexpectedly, this approach failed to detect either full-length or truncated Rider homologs in the close relative of tomato, S. pimpinellifolium. Extension of the same approaches to the genomes of the evolutionary more distant S. tuberosum and Capsicum annuum failed to detect substantial Rider homologs (Fig 5A and 5B), confirming the absence of Rider in the potato and pepper genomes [45]. As a control, we also analysed Arabidopsis thaliana, since previous studies reported the presence of Rider homologs in this model plant [45]. Using the BLAST approach above, we repeated the results provided in [45] and found BLAST hits of high sequence homology to internal sequences of Rider in the Arabidopsis thaliana genome. However, we did not detect sequence homologies to Rider LTRs (Fig 5C and 5D). Motivated by this finding and the possibility that Rider homologs in other species may have highly divergent LTRs, we screened for Rider LTRs that would have been missed in the analysis shown in Fig 5A and 5B due to the use of the full-length sequence of Rider as the query. Using the Rider LTR as a query revealed that S. pennellii, S. arcanum and S. habrochaites retain intact Rider LTR homologs, but S. pimpinellifolium exhibits a high BLAST hit density exclusively at approximately 60% homology. This suggests strong divergence of Rider LTRs in this species (Fig 5C and 5D). Overall, the results indicate intact Rider homologs in some Solanaceae species, whereas sequence similarities to Rider occur only within the coding area of the retrotransposons in more distant plants such as Arabidopsis thaliana. Therefore, LTRs, which include the cis-regulatory elements conferring stress-responsiveness, diverge markedly between species. Finally, we performed a reciprocal BLAST against tomato using Rider-like hits from all other species having sequence similarity over the entire element between 50% - 84% and confirmed that all Rider loci in tomato were among the top reciprocal BLAST hits.
To address the specificity of this divergence in Solanaceae species, we examined whether the CREs enriched in S. lycopersicum (Fig 2A) are present in LTR sequences of the Rider elements in S. pennellii, S. arcanum, S. habrochaites and S. pimpinellifolium (Fig 5C). While the LTRs identified in S. pennellii, S. arcanum and S. habrochaites retained all five previously identified CREs, more distant LTRs showed shortening of the U3 region associated with loss of the CGCG box (S6 Fig and S5 Table). This was observed already in S. pimpinellifolium, where all identified Rider LTRs lacked part of the U3 region containing the CGCG box (S6 Fig). Thus, Rider distribution and associated features differ even between closely related Solanaceae species, correlated with the occurrence of a truncated U3 region and family-wide loss of CREs.
Finally, to test the evolutionary conservation of Rider elements across the plant kingdom, we performed Rider BLAST searches against all 110 plant genomes available at the NCBI Reference Sequence (RefSeq) database (www.ncbi.nlm.nih.gov/refseq). Using the entire Rider sequence as the query to measure the abundance of Rider homologs throughout these genomes, we found Rider homologs in 14 diverse plant species (S7 Fig). The limited conservation of Rider LTR sequences in the same 14 species, revealed using the LTR sequence as the query, suggests that Rider LTRs are highly polymorphic and that drought-responsive CREs may nevertheless be restricted to Solanaceae (S8 Fig).
Comprehensive analysis of individual LTR retrotransposon families in complex plant genomes has been facilitated and become more accurate with the increasing availability of high-quality genome assemblies. Here, we took advantage of the most recent tomato genome release (SL3.0) to characterize with improved resolution the high-copy-number Rider retrotransposon family. Although Rider activity has been causally linked to the emergence of important agronomic phenotypes in tomato, the triggers of Rider have remained elusive. Despite the relatively low proportion (approximately 20%) of euchromatic chromosomal regions in the tomato genome [32]), our de novo functional annotation of full-length Rider elements revealed preferential compartmentalization of recent Rider insertions within euchromatin compared to aged insertions. Mapping analyses further revealed that recent rather than aged Rider transposition events are more likely to modify the close vicinity of genes. However, Rider copies inserted into heterochromatin have been passively maintained for longer periods. This differs significantly from other retrotransposon families in tomato such as Tnt1, ToRTL1 and T135, which show initial, preferential insertions into heterochromatic regions [66]. TARE1, a high-copy-number Copia-like element, is present predominantly in pericentromeric heterochromatin [67]. Another high-copy-number retrotransposon, Jinling, is also enriched in heterochromatic regions, making up about 2.5% of the tomato nuclear genome [68]. The Rider propensity to insert into gene-rich areas mirrors the insertional preferences of the ONSEN family in Arabidopsis. Since new ONSEN insertions confer heat-responsiveness to neighbouring genes [28,69], it is tempting to speculate that genes in the vicinity of new Rider insertions may acquire, at least transiently, drought-responsiveness.
We found that Rider transcript levels are elevated during dehydration stress mediated by ABA-dependent signalling. The activation of retrotransposons upon environmental cues has been shown extensively to rely on the presence of cis-regulatory elements within the retrotransposon LTRs [48]. The heat-responsiveness of ONSEN in Arabidopsis [26,27,70], Go-on in rice [25], and Copia in Drosophila [71] is conferred by the presence in their LTRs of consensus sequences found in the promoters of heat-shock responsive genes. Thus, the host’s heat-stress signalling appears to induce transcriptional activation of the transposon and promote transposition [70]. While ONSEN and Go-on are transcriptionally inert in the absence of a triggering stress, transcripts of Drosophila Copia are found in control conditions, resembling the regulatory situation in Rider. Due to relatively high constitutive expression, increase in transcript levels of Drosophila Copia following stress appears modest compared to ONSEN or Go-on, which are virtually silent in control conditions [25–27,70]. Regulation of Drosophila Copia mirrors that of Rider, where transcript levels during dehydration stress are very high but the relative increase compared to control conditions is rather modest.
The presence of MYB recognition sequences within Rider LTRs suggests that MYB transcription factors participate in transcriptional activation of Rider during dehydration. Multiple MYB subfamilies are involved in ABA-dependent stress responses in tomato, but strong enrichment of the MYB core element CTGTTA within Rider LTRs suggests involvement of R2R3-MYB transcription factors, which are markedly amplified in Solanaceae [72]. Members of this MYB subfamily are involved in the ABA signalling-mediated drought-stress response [73] and salt-stress signalling [74]. This possible involvement of R2R3-MYBs in Rider is reminiscent of the transcriptional activation of the tobacco retrotransposon Tto1 by the R2R3-MYB, member NtMYB2 [75]. Drought-responsiveness has been observed for Rider_08_3 only, despite other individual Rider copies displaying intact MYB core element (S3 Table). This suggests that presence of this CRE is not the only feature required for drought-responsiveness, and other factors, such as genomic location, influence Rider activity. Indeed, Rider_08_3 is located within a gene-rich area, with low TE content that might facilitate its activation. This is strikingly different from Rider_07_2 that is nested in a TE-rich area and isolated from genes (S6 Table).
In addition to environmental triggers, Rider transcript levels are regulated by the RdDM pathway. Depletion of SlNRPD1 and SlNRPE1 increases Rider transcript abundance, resulting in production of extrachromosomal circular DNA. Analysis of Rider-specific siRNA populations revealed that siRNA targeting of Rider elements is mostly independent of their chromatin context. This is somewhat unexpected since RdDM activity in tomato seems to be restricted to gene-rich euchromatin and it was postulated that accessibility of RNA Pol IV to heterochromatin is hindered by the compact chromatin structure [60,76,77]. We identified Rider copies targeted by RdDM, which potentially influences local epigenetic features. Loss of SlNRPD1 and SlNRPE1 leads to over-accumulation of 21-22-nt siRNAs at Rider copies, suggesting that inactivation of canonical RdDM pathway-dependent transcriptional gene silencing triggers the activity of the non-canonical RDR6 RdDM pathway at Rider [78–80].
It is noteworthy that, despite clear effects on Rider transcript accumulation and siRNA accumulation, loss of SlNRPD1 and SlNRPE1 is not manifested by drastic changes in total DNA methylation levels of Rider at the family level. This is in accordance with the modest decrease in genome-wide CHH and CHG methylation described in tomato RdDM mutants, with most of the changes happening on the euchromatic arms while the pericentromeric heterochromatin is unaffected [60]. Distribution of the 71 intact Rider elements in both euchromatic and heterochromatic compartments thus likely hampers detection of major changes DNA methylation over the Rider family. Only young euchromatic Rider elements marginally lose CHH methylation in the slnrpd1 mutant, but this is modest compared to the general decrease in mCHH observed throughout the chromosome arms [60]. As expected, CHH methylation at heterochromatic Rider is not affected. This suggests that SlCMT2 is involved in maintenance of mCHH at heterochromatic Rider copies in the absence of SlNRPD1, as observed previously for pericentromeric heterochromatin [60]. In general, our observations suggest that epigenetic silencing of Rider retrotransposons is particularly robust and involves compensatory pathways.
We identified extrachromosomal circular DNA originating from the Rider copies Rider_08_3 and Rider_07_2 in slnrpd1 and slnrpe1, respectively. In terms of DNA methylation and siRNA distribution at these two specific copies, loss of SlNRPD1 and SlNRPE1 brought different copy-specific outcomes. Rider_08_3, the main contributor to eccDNA in slnrpd1, displayed a reduction in CHH methylation that may contribute to increased transcription and the accumulation of eccDNA. In Rider_07_2, that provides a template for eccDNA in slnrpe1, there was no change in DNA methylation levels. Therefore, transcription and the production of eccDNA from this Rider copy is not regulated by DNA methylation. Consequently, eccDNA from Rider_07_2 was not detected in slnrpd1 despite drastic loss of CHH methylation.
Despite our efforts, we were unable to apply either drought or ABA treatment to the slnrpd1 and slnrpe1 mutants. In contrast to Arabidopsis [81,82], RdDM mutants in tomato are showing severe developmental defects and are sterile [60]. They are particularly difficult to maintain even in optimal growth conditions, precluding the application of stress treatments. Altogether, it appears that transcriptional control and reverse transcription of Rider copies occurs via multiple layers of regulation, possibly specific for individual Rider elements according to age, sequence and genomic location, that are targeted by parallel silencing pathways, including non-canonical RdDM [83,84].
The presence of Rider in tomato relatives as well as in more distantly related plant species has been described previously [34,45,47]. However, the de novo identification of Rider elements in the sampling provided here shows the distribution of the Rider family within plant species to be more complex than initially suggested. Surprisingly, mining for sequences with high similarity, overlapping more than 85% of the entire reference sequence of Rider, detected no full-length Rider elements in Solanum pimpinellifolium but in all other wild tomato species tested. Furthermore, the significant accumulation of only partial Rider copies in Solanum pimpinellifolium, the closest relative of tomato, does not match the established phylogeny of the Solanaceae. The cause of these patterns is unresolved but two scenarios can be envisaged. First, the absence of full-length Rider elements may be due to the suboptimal quality of genome assembly that may exclude a significant proportion of highly repetitive sequences such as Rider. This is supported by the N50 values within the Solanaceae, where the quality of genome assemblies varies significantly between species, with S. pimpinellifolium showing the lowest (S7 Table). An improved genome assembly would allow a refined analysis of Rider in this species. Alternatively, active Rider copies may have been lost in S. pimpinellifolium since the separation from the last common ancestor but not in the S. lycopersicum and S. pennellii lineages. The high-density of solo-LTRs and truncated elements in S. pimpinellifolium is in agreement with this hypothesis.
Comparing the sequences of Rider LTRs in the five tomato species, the unique occurrence of LTRs lacking most of the U3 region in S. pimpinellifolium suggests that loss of important regulatory sequences has impeded maintenance of intact Rider elements. Interestingly, part of the U3 region missing in S. pimpinellifolium contains the CGCG box, which is involved in response to environmental signals [49], as well as a short CpG-island-like structure (position 52–155 bp on reference Rider). CpG islands are usually enriched 5’ of transcriptionally active genes in vertebrates [85] and plants [86]. Despite the presence of truncated Rider LTRs, the occurrence of intact, full-length LTRs in other wild tomato species indicates that Rider is still potentially active in these genomes.
Altogether, our findings suggest that inter- and intra-species TE distribution can be uncoupled and that the evolution of TE families in present crop plants was more complex than initially anticipated. We have further opened interesting perspectives for harnessing transposon activities in crop breeding. Potentially active TE families that react to environmental stimuli, such as Rider, provide an unprecedented opportunity to generate genetic and epigenetic variation from which desirable agronomical traits may emerge. Notably, rewiring of gene expression networks regulating the drought-stress responses of new Rider insertions is an interesting strategy to engineer drought-resilient crops.
Tomato plants were grown under standard greenhouse conditions (16 h at 25°C with supplemental lighting of 88 w/m2 and 8 h at 15°C without). flacca (flc), notabilis (not), and sitiens (sit) seeds (cv. Ailsa Craig) were obtained from Andrew Thompson, Cranfield University; the slnrpd1 and slnrpe1 plants (cv. M82) were described before [60]. For aseptic growth, seeds of Solanum lycopersicum were surface-sterilized in 20% bleach for 10 min, rinsed three times with sterile H2O, germinated and grown on half-strength MS media (16 h light and 8 h dark at 24°C).
For dehydration stress, two-week-old greenhouse-grown plants were subjected to water deprivation for two weeks. For NaCl and mannitol treatments, tomato seedlings were grown aseptically for two weeks prior to transfer into half-strength MS solution containing 100, 200 or 300 nM NaCl or mannitol (Sigma) for 24 h. For abscisic acid (ABA) treatments, tomato seedlings were grown aseptically for two weeks prior to transfer into half-strength MS solution containing 0.5, 5, 10 or 100 μM ABA (Sigma) for 24 h. For 5-azacytidine treatments, tomato seedlings were germinated and grown aseptically on half-strength MS media containing 50 nM 5-azacytidine (Sigma) for two weeks. For cold stress experiments, two-week-old aseptically grown plants were transferred to 4°C for 24 h prior to sampling.
Total RNA was extracted from 200 mg quick-frozen tissue using the TRI Reagent (Sigma) according to the manufacturer’s instructions and resuspended in 50 μL H2O. The RNA concentration was estimated using the Qubit Fluorometric Quantitation system (Thermo Fisher). cDNAs were synthesized using a SuperScript VILO cDNA Synthesis Kit (Invitrogen). Real-time quantitative PCR was performed in the LightCycler 480 system (Roche) using primers listed in S8 Table. Selected Rider primers amplify 64 out of the 71 copies, with 3 mismatches allowed. Localization of Rider primers is shown in S9 Fig. LightCycler 480 SYBR Green I Master premix (Roche) was used to prepare the reaction mixture in a volume of 10 μL. Transcript levels were normalized to SlACTIN (Solyc03g078400). The results were analysed by the ΔΔCt method.
Tomato DNA was extracted using the Qiagen DNeasy Plant Mini Kit (Qiagen) following the manufacturer’s instructions and resuspended in 30 μL H2O. DNA concentration was estimated using the Qubit Fluorometric Quantitation system (Thermo Fisher). Quantitative PCR was performed in the LightCycler 480 system (Roche) using primers listed in S8 Table. Selected Rider primers amplify 64 out of the 71 copies, with 3 mismatches allowed. Localization of Rider primers is shown in S9 Fig. LightCycler 480 SYBR Green I Master premix (Roche) was used to prepare the reaction in a volume of 10 μL. DNA copy number was normalized to SlACTIN (Solyc03g078400). Results were analysed by the ΔΔCt method.
Extrachromosomal circular DNA amplification was derived from the previously published mobilome analysis [11]. In brief, extrachromosomal circular DNA was separated from chromosomal DNA using PlasmidSafe ATP-dependent DNase (EpiCentre) according to the manufacturer’s instructions with the incubation at 37°C extended to 17 h. The PlasmidSafe exonuclease degrades linear DNA and thus safeguards circular DNA molecules. Circular DNA was precipitated overnight at -20°C in 0.1 v/v 3 M sodium acetate (pH 5.2), 2.5 v/v EtOH and 1 μL glycogen (Sigma). The pellet was resuspended in 20 μL H2O. Inverse PCR reactions were carried out with 2 μL of DNA solution in a final volume of 20 μL using the GoTaq enzyme (Promega). The PCR conditions were as follows: denaturation at 95°C for 5 min, followed by 30 cycles at 95°C for 30 s, an annealing step for 30 s, an elongation step at 72°C for 60 s, and a final extension step at 72°C for 5 min. Selected primers amplify 68 out of the 71 Rider copies, with 3 mismatches allowed. Primer localization is shown on Fig 4A and 4C, left panel (grey bar: Rider element, black box: LTR, arrowheads: PCR primers) and sequences are listed in S8 Table. PCR products were separated in 1% agarose gels and developed by NuGenius (Syngene). Bands were extracted using the Qiagen Gel Extraction Kit and eluted in 30 μL H2O. Purified amplicons were subjected to Sanger sequencing. Five amplicons, obtained from two independent experiments, were sequenced for each eccDNA form.
A phylogenetic tree was constructed from the nucleotide sequences of the 71 Rider elements using Geneious 9.1.8 (www.geneious.com) and built with the Tamura-Nei neighbor joining method. Pairwise alignment for the building distance matrix was obtained using a global alignment with free end gaps and a cost matrix of 51% similarity.
Genomic coordinates of each of the 71 Rider elements identified by de novo annotation using LTRpred (https://github.com/HajkD/LTRpred) have been used to establish their chromosomal locations. Coordinates for centromeres were provided before [32] and pericentromeric regions were defined by high levels of DNA methylation and H3K9me2 ([60] and David Baulcombe, personal communication).
The Genbank accession number of the reference Rider nucleotide sequence identified in [45] is EU195798.2. We used Solanum lycopersicum bisulfite and small RNA sequencing data (SRP081115) generated in [60].
Insertion times of Rider elements were estimated using the method described in [45]. Degrees of divergence between LTRs of each individual element were determined using LTRpred. LTR divergence rates were then converted into dates using the average substitution rate of 6.96 x 10−9 substitutions per synonymous site per year for tomato [87].
We collected data from previously published BS-seq libraries of tomato mutants of RNA polymerase IV and V and controls [60]: slnrpe1 (SRR4013319), slnrpd1 (SRR4013316), wild type CAS9 (SRR4013314) and not transformed wild type (SRR4013312). The raw reads were analysed using our previously established pipeline [88] and aligned to the Solanum lycopersicum reference version SL3.0 (www.solgenomics.net/organism/Solanum_lycopersicum/genome). The chloroplast sequence (NC_007898) was used to estimate the bisulfite conversion (on average above 99%). The R package DMRcaller [89] was used to summarize the level of DNA methylation in the three cytosine contexts for each Rider copy.
Tomato siRNA libraries were obtained from [60] and analysed using the same analysis pipeline to align reads to the tomato genome version SL3.0. Briefly, the reads were trimmed with Trim Galore! (www.bioinformatics.babraham.ac.uk/projects/trim_galore) and mapped using the ShortStack software v3.6 [90]. The siRNA counts on the loci overlapping Rider copies were calculated with R and the package GenomicRanges.
Computationally reproducible analysis and annotation scripts for the following sections can be found at http://github.com/HajkD/RIDER.
We retrieved genome assemblies for 110 plant species (S9 Table) from NCBI RefSeq [91] using the meta.retrieval function from the R package biomartr [92]. For Solanum lycopersicum, we retrieved the most recent genome assembly version SL3.0 from the Sol Genomics Network ftp://ftp.solgenomics.net/tomato_genome/assembly/build_3.00/S_lycopersicum_chromosomes.3.00.fa [93].
Functional de novo annotations of LTR retrotransposons for seventeen genomes from the Asterids, Rosids, and monocot clades (Asterids: Capsicum annuum, C. baccatum MLFT02_5, C. chinense MCIT02_5, Coffea canephora, Petunia axillaris, Phytophthora inflata, Solanum arcanum, S. habrochaites, S. lycopersicum, S. melongena, S. pennellii, S. pimpinellifolium, S. tuberosum; Rosids: Arabidopsis thaliana, Vitis vinifera, and Cucumis melo; Monocots: Oryza sativa) were generated using the LTRpred.meta function from the LTRpred annotation pipeline (https://github.com/HajkD/LTRpred; also used in [25]). To retrieve a consistent and comparable set of functional annotations for all genomes, we consistently applied the following LTRpred parameter configurations to all Solanaceae genomes: minlenltr = 100, maxlenltr = 5000, mindistltr = 4000, maxdisltr = 30000, mintsd = 3, maxtsd = 20, vic = 80, overlaps = “no”, xdrop = 7, motifmis = 1, pbsradius = 60, pbsalilen = c(8,40), pbsoffset = c(0,10), quality.filter = TRUE, n.orf = 0. The plant-specific tRNAs used to screen for primer binding sites (PBS) were retrieved from GtRNAdb [94] and plant RNA [95] and combined in a custom fasta file. The hidden Markov model files for gag and pol protein conservation screening were retrieved from Pfam [96] using the protein domains RdRP_1 (PF00680), RdRP_2 (PF00978), RdRP_3 (PF00998), RdRP_4 (PF02123), RVT_1 (PF00078), RVT_2 (PF07727), Integrase DNA binding domain (PF00552), Integrase zinc binding domain (PF02022), Retrotrans_gag (PF03732), RNase H (PF00075), and Integrase core domain (PF00665).
We combined the de novo annotated LTR retrotransposons of the 17 species mentioned in the previous section in a large fasta file and used the cluster program VSEARCH [97] with parameter configurations: vsearch—cluster_fast—qmask none–id 0.85—clusterout_sort—clusterout_id—strand both—blast6out—sizeout to cluster LTR retrotransposons by nucleotide sequence homology (global sequence alignments). Next, we retrieved the 85% sequence homology clusters from the VSEARCH output and screened for clusters containing Rider sequences. This procedure enabled us to detect high sequence homology (>85%) sequences of Rider across diverse species.
To determine the distribution of Rider related sequences across the plant kingdom, we performed BLASTN [98] searches of Rider (= query sequence) using the function blast_genomes from the R package metablastr (https://github.com/HajkD/metablastr) against 110 plant genomes (S9 Table) and the parameter configuration: blastn -eval 1E-5 -max_target_seqs 5000. As a result, we retrieved a BLAST hit table containing 11,748,202 BLAST hits. Next, we filtered for hits that contained at least 50% sequence coverage (= sequence homology) and throughout at least 50% sequence length homology to the reference Rider sequence. This procedure reduced the initial 11,748,202 BLAST hits to 57,845 hits, which we further refer to as Rider-like elements. These 57,845 Rider-like elements are distributed across 21 species with various abundance frequencies. In a second step, we performed an analogous BLASTN search using only the 5’ LTR sequence of Rider to determine the distribution of Rider-like LTR across the plant kingdom. Using the same BLASTN search strategy described above, we retrieved 9,431 hits. After filtering for hits that contained at least 50% percent sequence coverage (= sequence homology) and at least 50% sequence length homology to the reference Rider LTR sequence, we obtained 2,342 BLAST hits distributed across five species.
We tested the enrichment of cis-regulatory elements (CREs) in Rider using two approaches. In the first approach, we compared Rider CREs to promoter sequences of all 35,092 protein coding genes from the tomato reference genome. We retrieved promoter sequences 400 bp upstream of the TSS of the respective genes. We constructed a 2x2 contingency table containing the respective motif count data of CRE observations in true Rider sequences versus counts in promoter sequences. We performed a Fisher’s exact test for count data to assess the statistical significance of enrichment between the motif count data retrieved from Rider sequences and the motif count data retrieved from promoter sequences. In the second approach, due to the unavailability of gene annotation for Solanum arcanum, Solanum habrochaites and Solanum pimpinellifolium we compared Rider CREs to randomly sampled sequence loci from the same genome using the following two step procedure: in step one, we sampled 1000 DNA sequences with the same length as the reference Rider sequence from 1000 randomly sampled loci in the tomato reference genome. When sampling, we also considered the strand direction of the reference Rider sequence. Whenever a Rider sequence was annotated in the plus direction, we also sampled the corresponding set of random sequences in the plus direction of the respective randomly drawn locus. In contrast, when a Rider sequence was annotated in the minus direction, we also sampled the corresponding set of random sequences in the minus direction. In step two, we counted CRE occurrences for each Rider sequence independently and for a set of different CREs. Next, we counted the number of the same CRE occurrences for each random sequence independently to assess how often these CREs were found in random sequences. We then, analogous to the first approach, constructed a 2x2 contingency table containing the respective motif count data of CRE observations in true Rider sequences versus counts in random sequences. We performed a Fisher’s exact test for count data to assess the statistical significance of enrichment between the motif count data retrieved from Rider sequences and the motif count data retrieved from random sequences. The resulting P-values are shown in S4 Table for the first approach and in S5 Table for the second approach. Computationally reproducible scripts to perform the motif count analysis can be found at https://github.com/HajkD/RIDER.
To assess the genome quality of Solanaceae species, we calculated the N50 metric for the genome assemblies of Solanum lycopersicum, S. pimpinellifolium, S. arcanum, S. pennellii, S. habrochaites, and S. tuberosum using the following procedure. First, we imported the scaffolds or chromosomes of each respective genome assembly using the R function read_genome() from the biomartr package. Next, for each species individually we determined the sequence length for each scaffold or chromosome and sorted them according to length in descending order. The N50 value in Mbp was then calculated in R as follows: N50 <- len.sorted[cumsum(len.sorted) > = sum(len.sorted)*0.5][1] / 1000000, where the variable len.sorted denotes the vector storing the ordered scaffold or chromosome lengths of a genome assembly.
SRAtoolkit, v2.8.0 (https://github.com/ncbi/sra-tools) and Biomartr 0.9.9000 (https://ropensci.github.io/biomartr/index.html) were used for data collection.
Phylogenetic trees were constructed using Geneious 9.1.8 (www.geneious.com).
The de novo retrotransposon annotation pipeline LTRpred is available in the GitHub repository (https://github.com/HajkD/LTRpred).
Rider annotation and analysis pipeline is available in the GitHub repository (https://github.com/HajkD/RIDER).
Distribution of Rider elements was done using the R package metablastr (https://github.com/HajkD/metablastr).
DNA methylation levels were assessed using the R package DMRcaller (http://bioconductor.org/packages/release/bioc/html/DMRcaller.html).
Small RNA analysis was done using Trim Galore! (www.bioinformatics.babraham.ac.uk/projects/trim_galore), ShortStack v3.6 (https://github.com/MikeAxtell/ShortStack) and GenomicRanges v3.8 (https://bioconductor.org/packages/release/bioc/html/GenomicRanges.html).
Reference Rider nucleotide sequence (accession number EU195798) is available here (https://www.ncbi.nlm.nih.gov/nuccore/EU195798).
The datasets supporting the conclusions of this article are available at Sequence Read Archive (SRA) (https://www.ncbi.nlm.nih.gov/sra/) under accession numbers "SRP081115", "SRR4013319", "SRR4013316", "SRR4013314" and "SRR4013312".
|
10.1371/journal.pcbi.1003604 | Mechanistic, Mathematical Model to Predict the Dynamics of Tissue Genesis in Bone Defects via Mechanical Feedback and Mediation of Biochemical Factors | The link between mechanics and biology in the generation and the adaptation of bone has been well studied in context of skeletal development and fracture healing. Yet, the prediction of tissue genesis within - and the spatiotemporal healing of - postnatal defects, necessitates a quantitative evaluation of mechano-biological interactions using experimental and clinical parameters. To address this current gap in knowledge, this study aims to develop a mechanistic mathematical model of tissue genesis using bone morphogenetic protein (BMP) to represent of a class of factors that may coordinate bone healing. Specifically, we developed a mechanistic, mathematical model to predict the dynamics of tissue genesis by periosteal progenitor cells within a long bone defect surrounded by periosteum and stabilized via an intramedullary nail. The emergent material properties and mechanical environment associated with nascent tissue genesis influence the strain stimulus sensed by progenitor cells within the periosteum. Using a mechanical finite element model, periosteal surface strains are predicted as a function of emergent, nascent tissue properties. Strains are then input to a mechanistic mathematical model, where mechanical regulation of BMP-2 production mediates rates of cellular proliferation, differentiation and tissue production, to predict healing outcomes. A parametric approach enables the spatial and temporal prediction of endochondral tissue regeneration, assessed as areas of cartilage and mineralized bone, as functions of radial distance from the periosteum and time. Comparing model results to histological outcomes from two previous studies of periosteum-mediated bone regeneration in a common ovine model, it was shown that mechanistic models incorporating mechanical feedback successfully predict patterns (spatial) and trends (temporal) of bone tissue regeneration. The novel model framework presented here integrates a mechanistic feedback system based on the mechanosensitivity of periosteal progenitor cells, which allows for modeling and prediction of tissue regeneration on multiple length and time scales. Through combination of computational, physical and engineering science approaches, the model platform provides a means to test new hypotheses in silico and to elucidate conditions conducive to endogenous tissue genesis. Next generation models will serve to unravel intrinsic differences in bone genesis by endochondral and intramembranous mechanisms.
| Arising as a consequence of trauma, tumor resection, removal of necrotic or infected tissue, and congenital abnormalities, critical-sized defects are too large to heal spontaneously and therefore require surgical intervention. New surgical approaches harness the regenerative power of the periosteum, a tissue membrane covering most bones, which provides a niche for stem cells and plays a key role in healing after injury. The interplay of mechanical, cellular and biochemical mechanisms involved in periosteum-mediated tissue genesis and healing remains elusive, providing the impetus for the current study. Here, we develop a mechanistic, mathematical model to predict the dynamics of tissue genesis by periosteum-derived stem cells within a bone defect surrounded by periosteum or a periosteum substitute. A mechanical finite element model is coupled with a model of cellular dynamics to simulate a tested clinical scenario in which the patient's own periosteum is left around the defect after injury. Model predictions incorporating mechanical feedback match spatiotemporal patterns of bone tissue regeneration observed in a series of in vivo ovine experiments. Through combination of computational, physical and engineering science approaches, the model platform provides a means to test new hypotheses in silico. This will provide criteria conducive to endogenous tissue genesis that can be tested in follow on experiments.
| Critical-sized long bone defects pose a currently intractable challenge in orthopaedics as they do not heal spontaneously without surgical intervention and they are associated with significant disability and health care costs. Drawbacks of currently available treatment options, such as distraction osteogenesis, include long treatment durations, and soft tissue scarring. Alternative tissue engineering approaches offer a means to harness endogenous healing processes. A recently developed one-stage bone transport surgical technique [1], [2] capitalizes on the regenerative capacity of the periosteum, the membrane bounding all non-articular, outer bone surfaces. The periosteum provides rich vascular and nervous connections, as well as a niche for progenitor cell populations [3].
Briefly, the one-stage bone transport technique introduces a new defect, enveloped in situ by the periosteum, by osteotomizing the underlying cortical bone and transporting it distally into the original defect site (Fig. 1A, B). Tested in a 16-week ovine femoral defect model, bridging does not occur in absence of the periosteum (control group), which confirms the critical size of the defect. In contrast, all treated groups (periosteum ± bone graft) exhibit de novo bone tissue genesis within and bridging across the defect. Furthermore, infilling is facilitated in the absence of bone graft within the defect [1]. Using a similar in vivo ovine model, a follow on study was conducted to determine which periosteal factors (e.g. cells, periosteal strips) are essential for the observed periosteum-mediated defect healing. A periosteum substitute, designed such that desired factors can be placed in its pockets, is sutured around the defect [4]. Tissue genesis is rapid when periosteum derived cells (PDCs) seeded on collagen sheets or strips of periosteum with cells in situ are tucked into the pockets. These experiments demonstrate the power of PDCs to generate new bone de novo [4]–[10]. In addition, biochemical or molecular factors intrinsic to the periosteum enhance tissue genesis by PDCs even without a patent blood supply. Finally, periosteal strips tucked into the periosteum substitute result in infilling of the defect with less dense but a greater volume of tissue than vascularized periosteum in situ [4].
Further, bone regeneration and maintenance processes are intrinsically linked to mechanical environment. Phenomenological studies of bone regeneration have assessed the role of specific mechanical signals in regeneration dynamics and tissue formation, where magnitude and type of mechanical stimulus are mapped to a regenerated tissue phenotype [11], [12]. While these predictive models are capable of determining nascent tissue type locally, as a function of mechanical cues, the cellular and subcellular mechanisms of mechanically modulated tissue genesis are still not fully understood.
Recent studies with periosteum progenitor cells indicate their mechanosensitivity in vitro and in situ, with applied stretch, or tensile strain, resulting in upregulation of chondro- and osteogenic growth factors [5], [7], [13], [14]. While a variety of growth factors are implicated in the healing process, bone morphogenetic protein 2 (BMP-2) is widely involved in all stages of bone regeneration [8], [9], [15], [16]. Additionally, periosteal injuries heal predominantly via endochondral [17] and, less frequently via intramembranous [1], [3], ossification mechanisms, motivating a deeper understanding of the interplay of mechanical environment on BMP-2 production during periosteally mediated bone regeneration. Finally, defect healing, including initial tissue genesis and vascular perfusion 16 weeks after surgery, correlate to mechanical loading during the post-surgical healing period [18] as well as net change from baseline of the periosteum's mechanical environment [19].
A quantitative understanding of the endogenous and exogenous cues that facilitate tissue manufacture by resident progenitor cells requires an approach that bridges length scales of tissues (mm-cm), cells (µm) and molecules (nm) as well as time scales of tissue generation and healing (months), secretion of extracellular tissue matrix (ECM, days-weeks), and cellular processes (hrs-days) [3], [20]–[24],[25],[26]. Multi-scale mechanistic models that describe cellular-tissue dynamics provide a unique tool to un-/couple spatial and temporal effects or specific mechanical and/or biological effects. Model simulations predict the effect of parameters that affect system behavior, which can be tested experimentally. The continual interdigitation of simulations with experimental studies is the most efficient and least costly process by which we can make significant improvement in regeneration of large defects in bone [22], [23].
Previously developed mathematical models of bone regeneration have incorporated the processes of cell proliferation, differentiation and ECM secretion, as mediated by growth factor production but with parametric incorporation of mechanical stimuli [27]–[30]. In the current study, we develop a mechanistic model framework to predict the cellular, extracellular and mechanical progression of defect infilling, governed by the mechanically mediated production of BMP-2 by progenitor cells located in the periosteum. In this first generation model, bone morphogenetic protein (BMP) is chosen to represent of a class of factors that may coordinate bone healing. Of particular relevance to our labs' experience with a series of experiments using a common ovine critical sized defect model, periosteum (-substitute) mediated tissue genesis within the defect occurs predominantly in a radially inward fashion with no relation to distance along the defect from the proximal or distal edge [4]. Hence, we hypothesize that mechanoregulatory stimulation of progenitor cells located in the periosteum (OP, for osteochondroprogenitors [3]) can be used to predict tissue genesis in defects, measured as the area of de novo cartilage and bone (in cross section, Fig. 1A–C).
The novelty of the approach lies in the incorporation of a mechanistic model accounting for OP mechanical stimulation at the periosteal surface, with direct rather than parametric mediation by BMP-2 production representing a class of molecules mediating tissue genesis and healing. This enables us to model mechanical stimulation of the periosteum, driving OP cell proliferation and differentiation processes, which in turn result in defect infilling and concomitant stiffening of the callus, and which further provides a mechanism for mechanical feedback.
The following sections describe our experimental and computational modeling approach to characterize mechanical and biochemical factors related to healing of a bone defect. The defect separates two parts of the bone that are stabilized initially along the long bone axis by an interlocked intramedullary nail. Periosteum surrounds the defect and contains the OP cells, the ‘sources of healing’ which produce BMP and other factors that mediate bone healing (Fig. 1). With this model of cellular and tissue dynamics, incorporating mechanical and biochemical factors, simulations are presented that show the effects of each of the rate processes that contribute to tissue genesis and mineralization. Model predictions incorporating mechanical feedback match spatiotemporal patterns of bone tissue regeneration observed in a series of in vivo ovine experiments.
A mechanical finite-element (FE) model of an adult human femur was established to approximate loading conditions at the surface of the periosteum during bone regeneration. Further, the FE model served as an input into a mechanistic mathematical model (Development of a Cellular-Tissue Model, below). The three-dimensional (3D) computer-aided design (CAD) geometry of the Sawbones standard femur model (third generation), created by M. Papini [31], was accessed online through the BEL Repository (https://www.biomedtown.org). The Sawbones femur represents a composite geometry, which has been validated experimentally as well as computationally to closely represent mechanical properties of the healthy femur [32]. Following import of the Sawbones model into a 3D CAD program (SolidWorks, Dessault Systèmes, Waltham, MA) the one-stage bone transport surgery was simulated on the model through the creation of a full 2.54 cm critical sized defect at the mid-diaphysis, measured as the midline between the femoral head and the condyles. The defect is stabilized with a stainless steel intramedullary (IM) nail of 35 cm length, 12 mm diameter, and interlocked to the proximal and distal femur via four locking bolts of 10 cm length and 7 mm diameter (Fig. 2).
Cancellous bone was not accounted for in the mechanical model, as it has been shown previously to alter predictions of strain by less than 1% in a similar linearly elastic model [33]. Joint contact forces, as well as the balancing iliotibial components of the abductors and tensor fascia latae were applied to represent the early stance phase [34], while maintaining the condyles in a fixed position. Meshing and FE analysis was performed (Ansys 14.5, Ansys, Inc. Canonsburg, PA), with a minimum of 150,000 quadratic tetrahedral elements.
The nascent tissue comprises extracellular matrix (ECM) in the form of rapid proliferative woven bone and/or osteochondral tissue in the process of ossification [3], [4], [35]. Tissue genesis proceeds in vivo within the defect throughout the healing process. At any point in time, the tissue (ECM) is idealized as either a cartilaginous and/or osseous template in the process of endochondral ossification. The periosteum is idealized as a membrane of negligible thickness relative to the scale of the defect site. The mechanical environment on the surface of the nascent tissue formed in the defect is therefore assumed to be the same as that of the comparatively soft and elastic periosteum. Material properties are applied based on commonly used values from published studies (Table 1). To assess and account for the evolving mechanical environment at the surface of the periosteum throughout tissue genesis and healing, the material properties of the nascent tissue (ECM, also referred to as callus) evolve over time with repeated simulations. Specifically, at 10 discrete intervals, representing phases of the defect infilling and healing process over time, the Young's modulus and Poisson's ratio are adjusted using mixture theory. The mechanical properties are defined, based on the state of the tissue, falling between the beginning and end states of the endochondral ossification process, with nascent tissue comprising 100% cartilage at one end and 100% cortical bone at the opposite end of the spectrum. The applied material properties are then calculated as a weighted average of the Young's modulus and Poisson's ratio.
Following simulation with the described loading, boundary and material conditions, the strains at the surface of the defect callus surface are extracted (Fig. 3) as inputs for the Cellular-Tissue Model (see below for details). In context of the current study, the axial strain, , which represents the largest measure of normal strain by approximately an order of magnitude, is assessed from each simulation. Strains are recorded as a function of nascent tissue's material properties at the periosteal surface. Based on previous experimental strain mapping studies from our group, positive strains (tensile) are experienced on the lateral aspect of the femur, while negative strains (compressive) are experienced on the medial aspect [19]. For this first generation of the model, only the tensile, positive strains are assessed as they have been more thoroughly described in the literature. Axial strains 90° orthogonal to the lateral aspect are averaged to approximate a representative value, and plotted as a function of Young's modulus (Fig. 3). The further development of mathematical relationships describing the effect of strain on periosteal osteoprogenitor cell behavior is outlined under Parameter Estimation and Simulation Strategy.
A mechanistic, mathematical model is developed to quantify the dynamics of cellular and tissue components that can form in a bone defect surrounded by periosteum (depicted schematically in Fig. 1). Definition of a cylindrical coordinate system best depicts tissue genesis described the experimental model [4], analogous to the geometry of a critical sized defect in cross-section and in cognizance of the small length scale cell activity relative to the span of the defect. Furthermore, nascent periosteum derived cell-modulated bone genesis in critical sized defects enveloped in situ by either native, intact periosteum [1], [36] or periosteum substitute [4], [36] proceeds primarily from the outer radial boundary of the bone defect inwards rather than from the axial proximal and distal edges of the defect toward the middle of the defect length. In this model, the primary regulatory processes of BMP-2 are probed in context of bone tissue genesis via endochondral pathways. While BMP is chosen generally to represent a class of molecules that modulate tissue genesis and healing, BMP-2 exerts unique effects on osteoprogenitor cells, chondrocytes and osteoblasts. In overview, a mechanical feedback loop is established, where chondrocytes produce cartilaginous ECM (cartilage), which is subsequently mineralized into bone by osteoblasts. The process of endochondral ossification results in evolution of material properties during tissue genesis, effectively stiffening the defect site and decreasing the mechanical strain experienced at the bounding periosteal surface during the course of healing. Mechanosensistive osteoprogenitor cells within the periosteum upregulate BMP-2 production as a function of their prevailing mechanical environment (strain). A decrease in production of BMP-2 follows stiffening of the tissue regenerate. BMP-2 in turn regulates the cell processes of proliferation and differentiation (Fig. 4).
The osteochondroprogenitor (OP) cells located within the cambium layer of human periosteum are capable of differentiating along chondrogenic and osteogenic pathways [37]. BMP-2 is known to regulate key biological activities of periosteal OP cells. Human mesenchymal stem cells (MSCs, of which OPs are a subset [38]) proliferate significantly faster following BMP-2 treatment relative to untreated control cells [39]. Additionally, differentiation of periosteal progenitors into chondrogenic and osteogenic cells is regulated by BMP-2 in a dose-dependent manner [8], [16]. While some migration of OP cells may occur, a simplifying assumption of no migration is made in the current model iteration, as cell tracking experiments indicate that periosteal OPs remain close to the periosteal surface [17]. Future versions of the model will be developed to determine eventual roles of migration activity on healing.
The primary behavioral processes of OP cells comprise proliferation and differentiation into chondrocytes (C) or osteoblasts (B), while remaining close to the periosteum at . The OP number per unit surface area at the periosteum changes with time according to:(1)Injury and mechanical stimulus of periosteum results in a rapid proliferation of OP cells [40]; proliferation and differentiation of OP cells serves to maintain a population of multipotent cells in the periosteum throughout healing. As long as the density of OP cells is below a critical density, the rate of OP cell proliferation follows a Monod relationship for a rate-limiting factor (BMP):(2)When the density of OP cells is above the critical density, the rate of proliferation matches the rate of differentiation to chondrocytes and osteoblasts:(3)such that population of OP cells remains constant in the periosteum during healing.
The rate of OP differentiation to chondrocytes or osteoblasts similarly follows a Monod relationship for BMP:(4A)(4B)In the Monod relationship, V represents the maximum rate and K is the bound BMP concentration at V/2.
As BMPs are the most well-known and researched musculoskeletal growth factors [41], they are the focus of the framework for growth factor activity in the model presented here (although future iterations of the model may be expanded to include an array of growth factors and cytokines that modulate tissue genesis and healing). BMPs are widely implicated as important regulatory factors during all stages of bone regeneration including cellular proliferation, differentiation, ECM production and apoptosis [42]. Recently, BMP-2 has also been shown to play an key role in periosteum-mediated bone regeneration [43], where deletion of BMP-2 postnatally almost completely blocks osteogenic and chondrogenic differentiation of periosteal progenitor cells [16]. The OP cells within the periosteum are mechanosensitive, with BMP-2 upregulation detectable within the periosteum in vivo as shortly as one hour after loading stimulation [44]. The periosteum also responds to mechanical stimulation by a robust proliferation of OP cells within the cambium layer [14], [45].
BMP-2 (here labeled as BMP for simplicity) is produced by mechanical stimulation of OP cells and diffuses away from the periosteum into the defect site. The system of BMP anatagonists is complex and not yet fully understood, but it appears to be a self-regulatory negative feedback loop [46]. To keep this aspect as straightforward as possible in the current generation of our model, we idealized deactivation of BMP from the system as a metabolic removal by cell uptake and consumption. As our understanding of the biology gains sophistication, the model will be refined to provide a more realistic reflection of the complex biological situation.
Hence, the number of BMP units per unit volume in the defect, , change according to:(5)Initially, no BMP is present: . Furthermore, at the surface of the nail, BMP cannot penetrate:(6)At the periosteum, the rate of production of BMP, which is proportional to the number density of the OP cells, equals the BMP diffusion flux into the defect. The rate of production depends on the strain at the periosteal surface, modeled as the axial normal strain, :(7)where relates periosteal strain to BMP production.
The mean axial normal strain, is calculated as an empirical function of the average elastic modulus : described in the Parameter Estimation and Simulation Strategy section. The average elastic modulus is integrated over the defect region:(8)The local elastic modulus depends on the area fractions of cartilage () and bone (), and is calculated using a law of mixtures, where the elastic modulus for cartilage () and for bone () are known constants:(9)and where the fraction of ECM at any position in the cross-section of the defect is:(10)
Chondrocytes (C) migrate according to random motility, proliferate, and die by apoptosis, where is the number of chondrocytes per unit volume:(11)The rate coefficient for proliferation depends on the local BMP concentration:(12)Chondrocyte apoptosis occurs at a critical density of the local ECM, i.e. :(13)Migrating cells do not enter into the intramedullary cavity (filled by a nail) so that the cell motility flux is zero; consequently,(14)Close to the periosteum, OP cells that differentiate into C cells migrate into the defect space. The rate of migration per unit surface area equals that of cell differentiation:(15)Initially, no chondrocytes are present in the defect:
Osteoblasts (B), which are formed by the differentiation of OP cells, migrate by random motility, proliferate and die by apoptosis. The number of osteoblasts per unit volume, , change with time and position as follows:(16)The rate coefficient for proliferation depends on the local BMP concentration:(17)Apoptosis of osteoblasts occurs when they are surrounded by a critical density of bone:(18)Since migrating cells do not enter into the intramedullary cavity (filled by a nail), the cell motility flux is zero; consequently,(19)Close to the periosteum, OP cells that differentiate into B cells migrate into the defect space. The rate of migration per unit surface area equals the rate of cell differentiation:(20)Initially, no osteoblasts are present in the defect:
The extracellular matrix (ECM) consists of cartilage and mineralized bone. Cartilage is produced by chondrocytes, and is mineralized (transformed) into bone, mediated by osteoblasts. In any region of the defect (), the ECM formation is considered to be characterized by neighborhood area fractions of cartilage () and bone (), such that(21)
Within the defect (), the local area fraction of cartilage increases in proportion to the local density of chondrocytes, and decreases in proportion to as the rate of mineralization:(22)The rate coefficient of cartilage formation varies with , the local area fraction of total ECM. When is small, the rate of production of cartilage is a maximum. When increases beyond a critical value, , the rate slows as increases due to contact inhibition. Cartilage production stops when reaches a critical maximum density :
Mineralized bone is produced by osteoblasts mineralizing the cartilage template, the presence of which must precede bone formation. The local area fraction of mineralized bone increases in proportion to :(23)The rate coefficient for bone mineralization, , depends upon the local cell density of osteoblasts:
Proliferative rates are estimated based on literature values for osteoprogenitor cells, chondrocytes and osteoblasts as: = 1.5 fold/day [47], = 1.3 fold/day [48] and = 2.4 fold/day [49]. The diffusivity of BMP-2, , is approximated as the diffusivity of protein in cytoplasm: = 0.013 cm2/day [50]. The motility of osteoblasts and chondrocytes is estimated as one order of magnitude lower than that of BMP: 1.3×10−3 cm2/day. The maximal rate of cartilage and bone production per day by chondrocytes and osteoblasts are estimated as: 3×10−6 cm2/(cell day) [51], [52].
The mean axial normal strain is calculated as a function of (GPa), which is determined from the average of for all lateral nodes at the periosteal surface from finite element outputs:(24)where is in GPa (Pa9) and is in millistrain (ε−3).
To estimate , we consider a periosteal tensile strain of 2.5 millistrain, experienced at the lateral surface in a rat forelimb model [53] in context of strain magnitudes predicted on the corresponding surface of our current FE model. In the experimental rat model, the strain induces a four-fold upregulation of BMP production at the periosteal surface, where a one-fold increase is comparable to the non-loaded side. Although the alignment of the strain gage during measurements was not reported in this study, compressive and tensile strains are reported, and we assume that the strains represent axial components. Simply put, a 100% increase in BMP production represents a two-fold upregulation, and a 300% increase represents a four-fold increase in BMP production at the periosteal surface (e.g. 100 pg BMP increasing by 300% would be 100 plus 300 pg, resulting in 400 pg total or a four-fold increase). We then apply the experimental observations relating strain (2.5 millistrain or 0.25%) and upregulation of BMP production (300%) in the rat model [53] to our FE model, which predicts axial strains on the surface of the human femur to range from zero to a maximum of 12 millistrain or 1.2%, with most values on the order of magnitude of 0.25% (per method of calculation outlined in Mechanical Model to Estimate Stain Environment at the Periosteum). Hence, assuming a linear relationship between strain and BMP production, the following value of kMech is established, which represents the percent increase in BMP production with a given strain: = 1.2 as a factor increase in BMP production over baseline per unit of microstrain.
The governing equations described previously are transformed to dimensionless versions (Fig. S1). Subsequently the spatial derivatives are discretized so that the model can be represented as an initial-value problem (Fig. S2). Numerical solution of this problem was obtained by applying a code for stiff differential equations “ode15s” in MATLAB R2011b (MathWorks, Natick, MA). For the first set of simulations, all dimensionless parameters are set to 1, except the calculated cell motilities, and and the mechanical stimulus parameter, . In subsequent simulations, parameters are varied independently to determine the relative effect on known outcome measures of ECM area fractions, and .
Accounting for the experimentally observed, near complete infilling of the defect site with mineralized bone after 16 weeks of healing, a baseline of dimensionless parameters was established to describe the ideal healing state (ECM outcome) at 16 weeks. The model was used to predict mechanically mediated growth-factor concentration gradients, cell density dynamics, as well as ensuing tissue regeneration outcomes consistent with defect infilling.
At the onset of healing, mechanical stimulation results in a rapid proliferation of osteoprogenitor cells within the periosteum, and an increase in BMP concentration (Fig. 5). The rapid diffusion of BMP from the periosteum to the intramedullary nail, relative to the expected total time course for tissue mineralization, results in a small spatial gradient of BMP (Fig. 6). Following increases in chondrocyte and osteoblast densities, metabolic consumption of BMP, coupled with decreased BMP production by osteoprogenitors via increasing nascent tissue stiffness, results in a gradual decrease in BMP concentration over time.
A rapid proliferation of osteoprogenitor cells within the periosteum is followed by saturation at the critical density, . The relatively faster differentiation of chondrocytes from osteoprogenitor cells contributes to a large area of cartilage formation, mechanically stabilizing the defect at early time points, and providing a template for subsequent mineralization by osteoblasts. To achieve defect infilling in the model, chondrocyte proliferation must proceed at a faster rate than osteoblast proliferation; this differs from experimentally measured relative rates, which indicate two-fold faster proliferation rates of osteoblasts compared to chondrocytes [48], [49]. Inhibition of efficient nutrient diffusion due to tissue generation in the defect (increased ECM area fraction) is idealized to trigger apoptosis in the model [54]. Osteoblasts are assumed to be sensitive only to the surrounding fraction of mineralized tissue, as they actively convert cartilage to bone. The idealized representation of osteoblastic apoptosis in the model would likely be observed biologically as apoptosis or transformation to osteocytes, as a subset of osteoblasts become embedded in their surrounding mineralized matrix, and form a network of osteocytes, for nutrient exchange [55], [56]. Accounting for the idealized nature of the current model, it will be desirable to include explicit biochemical, cellular and environmental cues triggering apoptosis of chondrocytes and osteoblasts in next generation models [57], [58], [59].
Rapid chondrocyte proliferation results in early formation of an immature tissue template. ECM area fraction is higher in close proximity to the periosteum, attributable to the motility of chondrocytes into the defect space following differentiation from osteoprogenitor cells. The slowly increasing population of osteoblasts subsequently transforms the cartilage template into mineralized bone, at half the rate of cartilage production by chondrocytes. At the final time-point, approximately 80% of the tissue regenerate comprises de novo mineralized bone, which is reflected in the progressive increase in elastic modulus.
Additionally, using the model, we probe the relative effects of key parameters with respect to the ideal healing outcome condition in several biologically relevant scenarios. Increasing the rate of differentiation of osteoprogenitor cells to chondrocytes, , contributes to a more rapid increased density of chondrocytes, as well as more rapid callus formation (Fig. 7). Similarly, increasing the rate of differentiation of osteoprogenitor cells to osteoblasts, , results in an increased density of osteoblasts, slightly decreased density of chondrocytes, as well as a more rapid mineralization of cartilage to bone (Fig. 8). Increasing the rate of proliferation of both chondrocytes () and osteoblasts () dramatically increases the cell density of each population (Fig. 9, 10). Increasing the rate of consumption of BMP by chondrocytes and osteoblasts () results in negative values for , and is therefore not physiologically plausible given the current definition of model parameters. Decreasing leaves considerably more BMP in the defect space, increasing most notably the density of chondrocytes and production of cartilage (Fig. 11). Increasing the maximum rate of cartilage production by chondrocytes () dramatically increases the area fraction of cartilage while simultaneously decreasing the density of chondrocytes as the density of ECM reaches the threshold for apoptosis sooner (Fig. 12).
In vivo experiments harnessing the regenerative capability of the periosteum to infill critical sized defects have been performed in ovine models [1], [4]. Two experiments provide ideal case studies to explore the power of the model to predict potential biological mechanisms leading to observed outcomes. In the first case study, resected autologous periosteal graft is tucked into a periosteal substitute membrane, which is then sutured around the critical sized defect, and stabilized by an intramedullary nail. In the second case study, a patent (intact vascularity) periosteal sleeve is sutured in situ after removal of underlying cortical bone and similar placement of an intramedullary nail for mechanical stabilization. The case studies are of particular interest, as they share a common final desired outcome of full tissue generation and healing of the defect at 16 weeks after surgery. However, previous studies indicate that the two case studies each exhibit a distinct time course for tissue generation as well as mechanism of mineralization.
Healing outcomes are assessed at 16 weeks, where tissue blocks are prepared for hard tissue histology, including Giemsa-eosin staining and fluorochrome microscopy. Giemsa-eosin staining dyes cartilage and cell nuclei blue, and mineralized bone tissue pink (Fig. 13A), offering an ideal comparison between model parameters and biological outcomes at a given time point. The nature of histological staining, however, does not enable temporal analysis of key variables as tissue must be fixed and processed. The chelation of fluorochromes, administered at distinct time-points (e.g. 2 weeks, 4 weeks) enables a semi-quantitative assessment of the extent of mineralization, where unique fluorescence wavelengths are utilized to indicate mineralization occurring during a known time span (Fig. 13B).
Comparing final outcome measures between the two experimental case studies, a larger area (in cross section, volume in full tissue block) of callus generation was observed when periosteum graft is incorporated in a periosteum substitute implant than when periosteum is sutured around the defect in situ. From micro-computed tomography (μCT) of the entire callus regenerated via periosteum sutured in situ, callus volume comprised 3500 mm3 out of the 4000 mm3, or 87.5% of total defect space [1], with cross-sectional area of tissue regenerate measured in histological cross sections proportional to representative volume. The μCT-measured volume corresponds well to the computational model parameter phi value of 0.1, corresponding to 90% callus infilling. Based on μCT measures of the case study in which periosteum is sutured in situ around the defect, total bone volume comprises approximately 40% of callus tissue regenerate. In contrast, periosteum mediated bone generation in the case where the periosteum substitute is used results in approximately 60% filling of the defect with bone; in this case study, quantitative μCT measures could not be made due to retention of the IM nail which leads to imaging artifacts. Though of the same order of magnitude, differences in bone generation between the two case studies may be attributed to differences in tissue regenerate composition, which result from parameters including relative cell populations, as well as differentiation and proliferation rates.
To begin to elucidate which predictive model parameters may lead to these observed differences in outcomes, model parameters are varied parametrically to achieve experimentally relevant ECM area outcomes. As an initial approach midway between the two experimental case study outcomes, ECM area outcomes were targeted at 50% bone and 50% cartilage comprising the total, final callus cross-section (Fig. 14). To achieve the experimentally relevant outcomes from the complete set of parameters of relevance for healing, the rate of differentiation of osteoprogenitor cells to osteoblasts, as well as the proliferation rate of chondrocytes and osteoblasts must be reduced. Additionally, cartilage and bone are formed at the same rate, whereas complete healing outcome analyzed previously (Fig. 5) requires a faster rate of cartilage production from chondrocytes.
Histological experimental measures including fluorescence intensity of the fluorochrome administered after two weeks of healing are comparable with computational predictions. Specifically, the radial intensity of the chelated fluorochrome, a measure of chelated fluorochrome and thus mineral concentration, significant correlates to periosteal proximity, where mineral concentration increases with increasing proximity to the periosteum and distance from the IM nail [18]. These data match the predicted gradients in BMP, cells and tissue fractions over time, as predicted by the computational model (Fig. 15). Taken together, the data from these two case studies demonstrate the feasibility of the predictive model.
In the preceding we demonstrate the development of a novel model framework, including cellular, mechanical and biochemical factors, and dynamics of tissue genesis. The mechanistic model that pairs FE mechanics and cellular-tissue dynamics successfully predicts effects of each rate process contributing to endochondral bone formation in postnatal critical sized bone defects, as observed in data from a series of experimental studies using a common ovine defect model. Together with data from experiments using the one-stage bone transport and periosteum substitutes, the model framework provides a novel means to elucidate the inherently complex process of in vivo, postnatal bone neogenesis in tissue defects.
The initial outcomes of the model motivate study of the mechano-regulatory process of progenitor cells to explain key spatial and temporal aspect of bone regeneration, resulting here in a simple model framework for testing mechanobiological hypotheses. Taken as a whole, the one-stage bone transport model studies present an interesting new clinical approach to promote healing via periosteally mediated bone regeneration in situ. Additionally, the one-stage bone transport model provides a clinically relevant lens from which to focus on modeling the biomechanical processes of bone regeneration in a critical sized defect covered by periosteum. Interestingly, the experimental model offers intrinsic advantages with regard to defining the boundary conditions of the computational model. For example, by virtue of the IM nail, periosteum (or substitute), and proximodistal bone at defect edges (1.27 cm from the defect center), the defect boundary conditions are uniquely defined. In addition, as it defines the outer boundary of the defect and the medullary niche (a source of MSCs) is completely filled by the IM nail, periosteum is the primary source of progenitor cells during defect healing.
While the initial outcomes of the integrated model compare well to in vivo large animal regeneration outcomes, experimental determination of key parameters will enable more accurate and complete model predictions. The importance of mechanical modulation of factors such as BMP is highlighted as a key regulator of cellular processes, in particular proliferation and differentiation rates, capable of predicting trends in defect infilling. However, it should be noted that in this first generation model, BMP represents a class of factors that my coordinate tissue genesis and bone healing.
Designed and tested for its capacity to predict observed outcomes in an experimental model with well characterized initial and boundary conditions as well as endpoints, next generations of the model can be refined to test other chemical factors or mechanical scenarios in the future. Specifically, increased sophistication with regard to several key idealizations will make future generations of the model more physiological and will potentially increase its predictive value. For instance from an anatomical perspective, cancellous bone was not accounted for in the mechanical model; while this idealization may be appropriate in consideration of its effect on strain at the middiaphysis, it limits the application of the model to cortical bone defects and ignores the metaphyseal compartment as a potential longer range source of progenitor cells. Furthermore, the model current addresses the process of bone formation via endochondral ossification alone, while it is known that osteogenesis can also proceed directly via an intramembranous pathway [1]. Finally, the current model does not incorporate cell motility or cell-cell interactions, which are known to be important mediators of cell signaling as well as modulator of emergent tissue architecture [3], [60].
While the presented model framework is limited by a number of assumptions and simplifications, its utility will be potentiated as our understanding of the complex process of tissue genesis and healing becomes better understood. For example, with increased understanding of cell signaling and cell behavior during tissue genesis, inclusion of additional complexity in the model will allow for testing of hypotheses, prioritization of experiments, and may contribute to a more complete understanding of the mechanically mediated process established here. The multi-scale component of integrating cellular and biochemical processes with tissue-scale mechanics and quantification contributes to a small but growing body of work.
This work additionally underscores the necessity for a deeper quantitative understanding of the basic biological process of bone regeneration. Notably, the biological signal transduction of mechanical environment is not yet well understood in terms of the time scale, magnitude, duration and cascade of growth factors produced in response to specific mechanical stimuli [3]. Immunohistochemistry and biochemical tools such as RT-PCR, Western blotting, and cell sorting will help quantify factor production following a given mechanical stimulus, in particular as these processes begin to be elucidated in progenitor cells from human periosteum [61]. Additionally, the effect of growth factor concentration on the relative rates of cellular differentiation and proliferation, and the extent to which spatial and temporal presentation alter pathways is an interesting area of study in context of future model development. Many growth factors are involved in the process of cartilage and bone tissue regeneration [62], and understanding their relative and synergistic contributions will be vital to improving model predictions.
Measurement of the inherent delay between triggering of cellular processes such as mechanotransduction resulting in up- or down-regulation of gene transcription, as well as ECM protein secretion and posttranslational modification may also contribute to estimating actual values of proliferation and differentiation rates, and should be assessed in future studies. Additionally, the rate of production of ECM components is one of several specific factors implicated in triggering rapid formation of structural tissue from cells, as well as their rate-limiting processes. From a therapeutic perspective, speeding the formation of a cartilage template and triggering a temporal increase in osteoblast density may help speed bridging time. Direct intramembranous bone formation, an endogenous means for rapid repair [1], [3], is a further natural paradigm that would lend itself well for study with the current model framework.
Further inherent limitations of the current model relate to the number of idealizations necessary to build and test the feasibility of the initial model platform. Next generation, follow on models may also incorporate additional biological and mechanical factors known to alter tissue regeneration in healing defects. Notably, the magnitude and duration of deviatoric and dilatational mechanical signals are known to modulate proliferation and differentiation pathways [3], [63]. Additionally, the early formation of vascular supply is implicated as playing an important role in regeneration [64], where the role of oxygen tension alters chondro- and osteogenesis in the healing callus [65], [66]. Models that describe the relationship between angiogenesis and bone regeneration have been previously established [29], [67] and may be readily incorporated into the mechanistic model framework presented here. Finally, the explicit depiction of cell motility as well as cell apoptosis in future models will add a further dynamic aspect that may better account for inherent differences in bone tissue genesis via intramembranous and endochondral mechanisms, which themselves represent variations on tissue genesis algorithms via epithelial to mesenchymal and mesenchymal to epithelial transitions [3].
Future versions of tissue genesis models may also integrate the mechanical model to provide a real-time strain stimulus, rather than a fitted-relationship value. This integration will allow for the analysis of the effects of dynamic loading conditions such as walking versus running, or therapeutic treatments to optimize stimulus for maximum quantity and quality of tissue regeneration. Individual-specific anatomic data may also be integrated into the mechanical model simulation to assess injury-specific regimens.
Looking forward to the next generation of periosteal implants and tissue-engineered replacements, specific application tissue healing may be modeled to test in silico, thereby providing a high-throughput test for critical parameters. More complex models may assess the material properties of the periosteum substrate in context of transmitting mechanical cues to underlying progenitor cells, or from a poroelastic and permeability perspective to guide nutrients into the defect space [4], [68], [69].
In conclusion, the model framework presented here offers a novel integration of a mechanistic feedback system based on the mechanosensitivity of periosteal progenitor cells to model and predict tissue regeneration on multiple length and time scales. The complex process of de novo bone regeneration involves many additional cellular and biochemical processes that should be incorporated in the future to improve the model's applicability. Mechanistic models offer great potential to both clinicians and researchers hoping to develop new techniques and insight into the process of bone regeneration, ultimately looking forward to novel therapies to improve patient outcomes.
|
10.1371/journal.pntd.0003616 | Prevalence and Determinants of the Gender Differentials Risk Factors of Child Deaths in Bangladesh: Evidence from the Bangladesh Demographic and Health Survey, 2011 | The number of child deaths is a potential indicator to assess the health condition of a country, and represents a major health challenge in Bangladesh. Although the country has performed exceptionally well in decreasing the mortality rate among children under five over the last few decades, mortality still remains relatively high. The main objective of this study is to identify the prevalence and determinants of the risk factors of child mortality in Bangladesh.
The data were based on a cross-sectional study collected from the Bangladesh Demographic and Health Survey (BDHS), 2011. The women participants numbered 16,025 from seven divisions of Bangladesh – Rajshahi, Dhaka, Chittagong, Barisal, Khulna, Rangpur and Sylhet. The 𝟀2 test and logistic regression model were applied to determine the prevalence and factors associated with child deaths in Bangladesh.
In 2011, the prevalence of child deaths in Bangladesh for boys and girls was 13.0% and 11.6%, respectively. The results showed that birth interval and birth order were the most important factors associated with child death risks; mothers’ education and socioeconomic status were also significant (males and females). The results also indicated that a higher birth order (7 & more) of child (OR=21.421 & 95%CI=16.879-27.186) with a short birth interval ≤ 2 years was more risky for child mortality, and lower birth order with longer birth interval >2 were significantly associated with child deaths. Other risk factors that affected child deaths in Bangladesh included young mothers of less than 25 years (mothers’ median age (26-36 years): OR=0.670, 95%CI=0.551-0.815), women without education compared to those with secondary and higher education (OR =0 .711 & .628, 95%CI=0.606-0.833 & 0.437-0.903), mothers who perceived their child body size to be larger than average and small size (OR= 1.525 & 1.068, 95%CI=1.221-1.905 & 0.913-1.249), and mothers who delivered their child by non-caesarean (OR= 1.687, 95%CI=1.253-2.272).
Community-based educational programs or awareness programs are required to reduce the child death in Bangladesh, especially for younger women should be increase the birth interval and decrease the birth order. The government should apply the strategies to enhance the socioeconomic conditions, especially in rural areas, increase the awareness program through media and expand schooling, particularly for girls.
| Children are a significant asset of a country. Child deaths are an important way to determine the health sector development. The effectiveness of the interventions is required to prevent child deaths. The purpose of this study is to identify the prevalence and risk factors of child deaths in Bangladesh. Data were collected from the Bangladesh Demographic and Health Survey, 2011. The results indicate that in Bangladesh there is an association with child deaths and mothers’ age, mothers’ education, social-economic status, birth interval, birth order, baby size and place delivered. For Bangladesh, this study recommends expanding female education to increase mothers’ knowledge, an awareness program about birth order (take one child) and an increase in the birth interval.
| Child deaths, which are one of the most important health indicators for a country, represent the socio-economic development of a population and have received considerable attention from national and international agencies in the last few decades because of their inclusion in the Millennium Development Goals (MDGs) [1, 2, 3]. A significant number of projects and programs have been conducted worldwide to reduce the under-five child deaths, especially in resource-limited countries, by two-thirds in 2015. In 2013, WHO reported that among the total of 6.3 million deaths of children aged under-five, 74% of them (4.6 million) died within the infancy period. However, 45% died within the first few months after birth. Almost 83% of this child mortality was due to neonatal, infectious or nutritional conditions. Although the number of under-five deaths has declined worldwide in recent decades (from 12.7 million in 1990 to 6.3 million in 2013), the number is still alarmingly high [4]. In addition, the high burden of child mortality still exists in South Asia (one child dies out of 15 before reaching 5 years of age) as well as Sub-Saharan Africa (one child dies out of 8) [2, 5]. About 33% of global child mortality occurs in South Asian countries, compared to less than 1% in high-income countries [6]. In Bangladesh, the mortality of children has followed a declining trend, having reduced from 133 deaths per 1000 in 1993 to 53 deaths per 1000 in 2011, which confirms that Bangladesh is more likely to reach target 4 of the MDG (48 deaths per 1000 children under 18 years of age) by 2015 [7].
Child deaths, as reported in previous studies, is associated with various socio-demographic, and health related characteristics, e.g., lower parental education, lower socioeconomic status, and higher order of birth are associated with increasing risk of child mortality [8, 9, 10]. Whereas, large birth spacing, lower birth order, urban dwelling, and high socioeconomic status are associated with a lower risk of child mortality [10, 11, 12]. Moreover, another study in Bangladesh identified parent’s education, parent’s occupation, delivery and size of child as significant determinants of child mortality [13, 14]. A few studies also reported a significant difference in child deaths between urban and rural areas [15, 16]. In addition, a multi-country study confirmed that higher national income is associated with a lower rate of child deaths [17].
In developing countries like Bangladesh, women are neglected in almost all aspects of life. In addition, such negligence starts from the childhood, as households, especially from rural areas and among the uneducated have a desire for a male child. Even the family is unhappy for the birth of a female baby. Moreover, social disparity and gender differences for health care exist; for example, girls experience a delay in the intervention for their illness compared to boys [18]. Therefore, the same factor may affect child deaths in a different fashion (severity) for male and female children. Although some studies have been conducted in the field of child mortality in Bangladesh [10, 13, 14, 19], they did not study child deaths separately for male and female children. Therefore, the objective of this study is to determine the risk factors that influence child deaths in Bangladesh among males and females separately.
This is a cross-sectional study using data from the BDHS, 2011.
There are seven administrative divisions in Bangladesh—Dhaka, Rajshahi, Rangpur, Chittagong, Khulna, Barisal and Sylhet. One division is subdivided into districts (zilas), and each district is divided into administrative units (upazilas), which are further divided into urban and rural areas. An urban area is divided into wards and city corporation units (mohallas) within a ward, while a rural area is divided into union parishes (UP) and villages (mouzas) within a union parish. The 2011 population and housing census, together with the Bangladesh Bureau of Statistics (BBS), were used as the sampling frame for the list of enumeration areas (EAs) in this survey [7].
A two-stage stratified sample of households was the basis for this survey. In the first stage, 207 clusters in urban areas and 393 in rural areas were selected totaling 600 enumeration areas with proportional probability. In the second stage, on average, 30 households were selected based on the demographic and health survey variables, for both the urban and rural areas in seven divisions by systematic sample. The survey selected 17,842 residential households within this study design [7].
The main objective of the study is to determine the prevalence and risk factors of child deaths in Bangladesh.
The statistical analysis of the results was measured using the IBM statistics Version 21. The χ2 test was used to determine the significant associations with the child deaths (boys and girls) and respondents age, place of residence, educational level, socioeconomic status, total number of children ever born, birth interval, mode of delivery and size of child at last birth. Logistic regression analysis was conducted to determine the risk factors with child deaths (boys and girls) and respondents age, place of residence, educational level, socioeconomic status, total number of children ever born, birth interval, mode of delivery and size of child at last birth (Table 1). A P value of 0.05 was considered significant at the 95% CI (Confidence Interval) level. The dependent variable used in the model was the dichotomous binary variable: Y = 1 if have child deaths (boys and girls) and Y = 0, otherwise. Respondents age, place of residence, educational level, socioeconomic status, total number of children ever born, birth interval, mode of delivery and size of child at last birth were considered as predictor variables in this model. Perform preliminary analyses using univariate tests such as chi-square test. Such initial analyses provide a better grasp of what is happening in the data and may point the potential important variable(s) to be used for the multivariate analysis. As a general rule, perform univariate analyses for each independent variable, if they are significant at p-value of 0.25 then select them to the multivariate analysis. If the χ2 test for a single independent variable is not significant there is no need to include it in the logistic regression analysis.
The rural population refers to the people living in rural areas and urban refers to the people living in urban or city areas. There is a difference between urban and rural in that urban people have access to medical facilities, health care centers, good communication facilities, living standards and schooling compared to rural areas. Poor people refers to those with a monthly income of less than 3000 Bangladeshi Taka, middle class people refers to those with a monthly income of 3000–15,000 and rich people refers to those with a monthly income of more than 15,000 Bangladeshi Taka. If the baby size is 2700–4000 grams (6–9 pounds) that means average baby size, less than 2700 grams is small and more than 4000 grams is large [7].
Ethical approval was obtained from the Ministry of Health and Family Welfare, Dhaka, Bangladesh. International Credit Finance (ICF) provided financial and technical assistance for the survey through USAID/Bangladesh. The BDHS is part of the worldwide Demographic and Health Surveys program, which is designed to collect data on child health [7]. Informed written consent was obtained from all women prior to the study.
The total women participants included in the study numbered 16,025. The results of the associations of child mortality for several socio-demographic characteristics of women are presented in Table 2. The findings from the present study indicated that the prevalence of male children dying was 13.0% and 11.6% for female children. For both sexes, child deaths was significantly associated with mothers’ age, place of residence, mothers’ educational level, total number of children ever born, socio-economic status of household, preceding birth interval, mode of delivery, and size of child at birth. In observing the reported child mortality from several socio-demographic characteristics, an increasing percentage of child mortality was observed in the case of the mothers’ age. Child mortality was significantly higher in rural areas (14.2% for male child and 12.8% for female child) compared to those in urban areas (10.7% for male child and 9.4% for female child). The mothers’ educational level was significantly associated with child deaths. The child mortality rate for males was higher among women with no formal education (22.0%) followed by primary educated (14.5%), secondary educated (6.3%) and higher educated women (3.4%). A similar pattern of mortality was also observed in the case of female children; however, for each level of mothers’ education, the percentage was comparatively lower than the mortality for males. For both sexes, the reported child mortality was high among poor households (17.4% for males and 14.9% for females) and the percentage decreased with the increasing socio-economic status.
Women with more than six children reported a higher percentage of child mortality for both sexes, whereas it was lower among women with 1–2 children. However, the percentage of child mortality was higher among those children whose preceding birth interval was less than 24 months and lower among those children whose preceding birth interval was more than 48 months (12.5% for males and 11.8% for females). The percentage of child deaths was higher among those children who were delivered normally (13.5% for males and 12.2% for females) than for children who were delivered by caesarean section (6.0% for males and 4.7% for females). The percentage of mortality was also higher among those children whose size at birth was small (16.2% for males and 14.6% for females); however, the percentage was lower among those whose size at birth was large (9.0% for males and 7.4% for females).
The results in table 3 indicated a positive high relationship with the place of residence and socioeconomic status, a negative high correlation with educational level and socioeconomic status, a positive correlation with the total number of children ever born and the mode of delivery and the other variables are negatively or positively correlated with each other.
From table 4, the results showed an inverse moderate relationship with respondents’ age and total number of children ever born, a positive high relationship with place of residence and socioeconomic status, a positive moderate relationship with the total number of children ever born, and positive moderate relationship with the mode of delivery and total number of children ever born.
Table 5 represents the risk factors associated with child deaths. Two separate binary logistic regression models were fitted to identify the socio-economic correlates of child mortality for males and females. The selected socio-demographic variables considered in the two models are mothers’ age, place of residence, mothers’ educational level, total number of children ever born, socio-economic status of household, preceding birth interval, mode of delivery, and size of child at birth. According to the fitted models, all of the selected variables except place of residence and child size at last birth appeared to be statistically significantly correlated to child mortality in the case of males; whereas all of the selected variables except place of residence appeared to be significant correlates of child mortality in the case of females.
The findings from this study revealed that women who are aged 26–36 years have 33% and 26% lower risk of child mortality for their male children (OR = 0.67, 95%CI = 0.55–0.81) and female children (OR = 0.74, 95%CI = 0.59–0.91), respectively, compared to women aged 25 years and below. There was a negative association between higher education and child mortality. The risk was significantly lower among women who were higher educated followed by secondary and primary educated compared to women with no formal education. For example, in the case of female mortality, the risk was about 50% lower among higher educated women (OR = 0.51, 95% CI = 0.35–0.77) compared to women with no formal education.
Household socio-economic status was an important indicator that was correlated with child deaths between both males and females. Middle class and upper households were 0.81 times (OR = 0.81, 95%CI = 0.69–0.93) and 0.71 times (OR = 0.71, 95%CI = 0.61–0.81) less likely to have sons die, respectively, than poor households; whereas, rich households had 0.81 times less (OR = 0.81, 95% CI = 0.70–0.93) chance of having daughters die. In addition, women whose total number of children ever born was 3–6 and more than 6 were 5.33 times and 21.42 times more likely of having sons die compared to women whose total number of child ever born was one to two. An almost similar risk was seen in the case of girl child’s mortality.
The preceding birth interval also appeared to be another important correlate for both models. The risk of death was 0.71 (OR = 0.71, 95% CI = 0.62–0.81) times less for those whose preceding birth interval was 25–48 months compared to those whose preceding birth interval was less than 25 months. Sons who were delivered by normal section had 1.68 times (OR = 1.68, 95% CI = 1.25–2.27) more risk of mortality compared to sons delivered by caesarean. However, the risk was 1.47 times (OR = 1.47, 95% CI = 1.06–2.04) more in the case of daughters. Daughters whose size at birth was average were 1.53 times (OR = 1.53, 95% CI = 1.22–1.91) more likely to face mortality compared to daughters whose size at birth was large, which indicates that large size at birth is a protective factor for child deaths. For the fitted model, in respect of the Cox and Snell R2 and Nagelkerke R2, 12% and 20%, respectively, of the variance for boys, and 12% and 21%, respectively, of the variance for girls can be predicted from the linear relationship of the eight independents variables. In respect of the overall percentage, 88% and 87% women were predicted correctly. The overall model was significant when all eight independent variables were entered.
The child death rate reflects a country’s level of socioeconomic improvement and quality of life. It depends on monitoring and evaluating the population and health programs and policies. These rates are also useful in identifying promising directions for the health and nutrition programs in Bangladesh [7]. The results of this study indicate that for the death of children less than 18 years of age by gender difference, a child’s death depends on mothers’ age, mothers’ education, socioeconomic status, geographic difference and mass media awareness program. Births to younger women can generally be identified as high-risk births [2, 20]. This study confirms that children born (especially boy) to women of more than 25 years of age are at lower risk of child deaths. In Bangladesh, majority younger women are not developed economically, emotionally, physically, and are less mature, which, together with other aspects of their life, have a detrimental impact on their children.
Women’s education has an inverse relation with children’s deaths. Educational level is highly association with lower mortality risks because women’s education provides the information about better pregnancy and child health care [7, 21]. The results show that the child deaths for boys is 0.711 times lower risk for those whose mothers have completed higher education than for those whose mothers are illiterate (6.3% to 22.0% deaths), and for the child deaths of girls it is 0.588 times lower for those whose mothers have completed higher education than for those whose mothers are illiterate (5.2% to 21.0% deaths). Mothers with higher education reduce the risk of child deaths by about half. Educated women have better income, higher health literacy and power to make healthier decisions for their health and that of their children [3].
Child deaths in urban and rural areas in Bangladesh have a statistically significant risk difference. Some studies have identified that child mortality in rural areas is at higher risk than in urban areas [3, 10–14, 15, 22]. This study shows that there is a significant association with urban-rural and gender difference, such as the deaths of urban boys and girls (10.7% to 9.4%) and that of rural boys and girls (14.2% to 12.8%). This is because urban areas have good access to basic medical facilities and health care compared to rural areas.
The wealth index of women is the main way to determine the health status in a country. Some studies have shown a positive statistical association between low income and child deaths but an opposite relationship between high income and child deaths [3, 8, 17]. Similarly, in this study the lower income women are at greater risk for child deaths (boys and girls). Because lower income women have access to fewer medical facilities in rural areas, such as hospitals, doctors, paramedics or community based health workers.
Birth order is an essential measure of child deaths. Although some previous studies have shown that higher ranked birth order present a higher risk of child deaths, a few other studies have indicated that lower ranked birth order experience an increased risk of child deaths. The infants of first birth have a higher risk of neonatal mortality than fourth or higher-ranked births in India [23, 24, 25]. A study conducted in Taiwan [26] showed that children with first and fifth-ranked births are at higher risk of early child deaths, with same in Nigeria [27]. In this study, women who have children ever born 3–6 is 5.33 times more risk of boys child death, and children ever born more than 7 is 21.42 times more risk of sons child death as compared to women who had at least one child ever born. An almost similar risk is seen in the case of girl child deaths. Short birth interval and child deaths have a significant association in Bangladesh [3, 28]. Another study also found a similar risk association [29]. The risk of death is 0.71 times less for those women having a birth interval of 25–48 months compared to those women having a birth interval of less than 25 months. One of the limitations in this study is that the information was self-reported.
The causes of child deaths is important for health sector planning, especially in determining the program needs, and monitoring for improving intervention and reassessment of health priorities. Increasing mothers’ education and making them productive to improve their income are important aspects for reducing child deaths. Hence, it is having needed to increase rural based community education programs about child deaths. Being a younger mother at birth, short birth interval and birth order have been identified as risk factors for increased child deaths in Bangladesh. Interventions targeted at empowering women and much effort in emerging rural areas is necessary. Reducing young motherhood and increasing satisfactory birth spacing are also necessary to reduce child deaths in Bangladesh.
|
Subsets and Splits